Research on Provincial-Level Soil Moisture Prediction Based on Extreme Gradient Boosting Model

Yifang Ren; Fenghua Ling; Yong Wang

doi:10.3390/agriculture13050927

Research on Provincial-Level Soil Moisture Prediction Based on Extreme Gradient Boosting Model

Ren, Yifang;Ling, Fenghua;Wang, Yong 2023-04-24 00:00:00 agriculture Article Research on Provincial-Level Soil Moisture Prediction Based on Extreme Gradient Boosting Model 1 2 3 , Yifang Ren , Fenghua Ling and Yong Wang * Jiangsu Provincial Climate Center, Nanjing 210008, China; renyifang2023@gmail.com Institute for Climate and Application Research (ICAR)/CIC-FEMD/KLME/ILCEC, School of Atmospheric Sciences, Nanjing University of Information Science and Technology, Nanjing 210044, China; 20211101018@nuist.edu.cn School of Applied Meteorology, Nanjing University of Information Science and Technology, Nanjing 210044, China * Correspondence: ywang@nuist.edu.cn Abstract: As one of the physical quantities concerned in agricultural production, soil moisture can effectively guide ﬁeld irrigation and evaluate the distribution of water resources for crop growth in various regions. However, the spatial variability of soil moisture is dramatic, and its time series data are highly noisy, nonlinear, and nonstationary, and thus hard to predict accurately. In this study, taking Jiangsu Province in China as an example, the data of 70 meteorological and soil moisture automatic observation stations from 2014 to 2022 were used to establish prediction models of 0–10 cm soil relative humidity (RH ) via the extreme gradient boosting (XGBoost) algorithm. Before constructing the s10cm model, according to the measured soil physical characteristics, the soil moisture observation data were divided into three categories: sandy soil, loam soil, and clay soil. Based on the impacts of various factors on the soil water budget balance, 14 predictors were chosen for constructing the model, among which atmospheric and soil factors accounted for 10 and 4, respectively. Considering the differences in soil physical characteristics and the lagged effects of environmental impacts, the best inﬂuence times of the predictors for different soil types were determined through correlation analysis to improve the rationality of the model construction. To better evaluate the importance of soil factors, two sets of models (Model and Model ) were designed by taking soil factors as optional predictors _atmo _soil&atmo put into the XGBoost model. Meanwhile, the contributions of predictors to the prediction results Citation: Ren, Y.; Ling, F.; Wang, Y. were analyzed with Shapley additive explanation (SHAP). Six prediction effect indicators, as well as Research on Provincial-Level Soil a typical drought process that happened in 2022, were analyzed to evaluate the prediction accuracy. Moisture Prediction Based on The results show that the time with the highest correlations between environmental predictors and Extreme Gradient Boosting Model. RH varied but was similar between soil types. Among these predictors, the contribution rates of s10cm Agriculture 2023, 13, 927. https:// maximum air temperature (T ), cumulative precipitation (P ), and air relative humidity (RH ) amax sum a doi.org/10.3390/agriculture13050927 in atmospheric factors, which functioned as a critical factor affecting the variation in soil moisture, Academic Editors: Tarendra are relatively high in both models. In addition, adding soil factors could improve the accuracy of soil Lakhankar and Gerard Arbat moisture prediction. To a certain extent, the XGBoost model performed better when compared with artiﬁcial neural networks (ANNs), random forests (RFs), and support vector machines (SVMs). The Received: 28 February 2023 values of the correlation coefﬁcient (R), root mean square error (RMSE), mean absolute error (MAE), Revised: 14 April 2023 Accepted: 17 April 2023 mean absolute relative error (MARE), Nash–Sutcliffe efﬁciency coefﬁcient (NSE), and accuracy (ACC) Published: 24 April 2023 of Model were 0.69, 11.11, 4.87, 0.12, 0.50, and 88%, respectively. This study veriﬁed that _soil&atmo the XGBoost model is applicable to the prediction of soil moisture at the provincial level, as it could reasonably predict the development processes of the typical drought event. Copyright: © 2023 by the authors. Keywords: soil moisture; prediction; XGBoost algorithm; SHAP Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons 1. Introduction Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ Soil moisture is a critical climate variable that regulates climate change by facilitating 4.0/). the exchange and distribution of water and energy in land–air interaction. Additionally, Agriculture 2023, 13, 927. https://doi.org/10.3390/agriculture13050927 https://www.mdpi.com/journal/agriculture Agriculture 2023, 13, 927 2 of 17 soil moisture plays a signiﬁcant role in agricultural production, as deﬁcits or overﬂows of soil moisture during critical periods can impact crop growth and yields [1]. Integrating information on available soil moisture and crop water demands can help the development of timely and appropriate irrigation schedules [2], which is particularly important in areas with poor water conditions. The variations and differences in soil moisture across regions are determined by its budget balance, which is inﬂuenced by several factors. Soil moisture is sourced from atmospheric precipitation and artiﬁcial irrigation, and its expenditure depends on physical processes such as evapotranspiration and runoff, which are inﬂuenced by local weather conditions, soil characteristics, land cover, and other factors [3]. Usually, soil moisture can be expressed using physical variables such as relative humidity, weight water content, and volume water content. Among these variables, relative humidity, calculated as the percentage of soil water content and ﬁeld capacity, can comprehensively reﬂect the soil moisture status and surface hydrological processes [4,5]. Consequently, soil relative humid- ity is an essential reference in irrigation, enabling an analysis of soil moisture differences between regions. Soil moisture prediction based on relative humidity can enhance the defense against waterlogging and drought in farmland. Numerous studies have investigated soil moisture prediction using various methods. Traditional approaches include the water balance method [6–8], statistical empirical for- mula method [9], time series method [10,11], and physical models based on hydrological processes [12]. These methods typically consider the soil water budget balance principle, relationships between soil water and environmental factors, change characteristics of soil water over time, and land–air interaction. They use model building or time series analysis to forecast soil moisture. With advances in information technology, various applications of machine learning (ML) in agricultural production have been widely developed, including predictions of the crop growth period, yield, and soil moisture [13–16]. ML technologies such as artiﬁcial neural networks (ANNs) [17], support vector machines (SVMs) [18], and gradient boosting regression trees (GBRTs) [19] offer a novel perspective for soil moisture prediction due to their advantages of having a low computational cost, strong self-learning ability, high prediction accuracy, and wide suitability [20–22]. For instance, a GA-BP neural network regression model was tested to perform well in predicting the soil moisture of high side slopes [23]. A proposed novel encoder–decoder model with residual learning played an excellent role in solving the nonlinear problem of soil moisture prediction, which was tested using data from 13 FLUXNET sites with varying plant function types and climatic characteristics [24]. In the research of soil moisture prediction based on machine learning, besides ﬁnding suitable prediction models [25], selecting the appropriate input factors for the prediction model is crucial. Many studies have selected meteorological factors directly related to soil moisture, such as precipitation, transpiration, sunshine, and surface temperature [26]. For instance, Xu et al. (2010) developed and tested an integrated soil moisture prediction model based on artiﬁcial neural networks (ANNs) with meteorological data in the semi-arid region of eastern China, and the model performed well at basin scales [27]. Li et al. (2018) applied the adaptive genetic ANN method to improve the quality of soil moisture prediction using atmospheric forcing data, which include air temperature, relative humidity, wind speed, radiation, and precipitation, as well as soil forcing data, such as soil temperature at 5 cm depth and lagged soil moisture at 0–10 cm [28]. Moreover, with the advancement of remote sensing technology, remote sensing monitoring indexes based on multi-source data, including optical, thermal infrared, microwave, and other data, have also been widely used for soil moisture monitoring and prediction [29–31]. However, current research on soil moisture prediction has some limitations, including discontinuity in remote sensing images, an inadequate use of data from automatic observa- tion stations, and unclear inﬂuencing factors of soil moisture [24,32]. Therefore, this study utilized the soil moisture data and corresponding meteorological data from 70 automatic stations in Jiangsu Province, determined the optimal inﬂuence times of the input factors Agriculture 2023, 13, x FOR PEER REVIEW 3 of 18 observation stations, and unclear inﬂuencing factors of soil moisture [24,32]. Therefore, Agriculture 2023, 13, 927 3 of 17 this study utilized the soil moisture data and corresponding meteorological data from 70 automatic stations in Jiangsu Province, determined the optimal inﬂuence times of the in- put factors for prediction models using a correlation analysis method, and applied ex- treme gradient boosting (XGBoost) to establish two sets of soil relative humidity predic- for prediction models using a correlation analysis method, and applied extreme gradient tion models (i.e., Model_soil&atmo and Model_atmo). To bett er interpret the inﬂuences of the boosting (XGBoost) to establish two sets of soil relative humidity prediction models (i.e., input factors on these two models and evaluate their performance, Shapley additive ex- Model and Model ). To better interpret the inﬂuences of the input factors on _atmo _soil&atmo planation (SHAP) was applied, and six metrics were utilized as the predicting eﬀect indi- these two models and evaluate their performance, Shapley additive explanation (SHAP) cators to compare the models’ (e.g., ANN, RF, and SVM) prediction accuracy. Further- was applied, and six metrics were utilized as the predicting effect indicators to compare more, a typical drought development process in August 2022 in Jiangsu Province was an- the models’ (e.g., ANN, RF, and SVM) prediction accuracy. Furthermore, a typical drought alyzed in depth. This study aimed to establish a provincial-level and understandable soil development process in August 2022 in Jiangsu Province was analyzed in depth. This m study oistuaimed re pred to ict establish ion mode al b pr yovincial-level applying a mand achine understandable learning algorit soil hm, moistur which co e pruld ediction pro- vid model e a ca by se stu applying dy for other a machine region learning s. algorithm, which could provide a case study for other regions. 2. Materials and Methods 2. Materials and Methods 2.1. Study Area 2.1. Study Area Jiangsu Province (see Figure 1) is located on the east coast of China, in the mid-lati- Jiangsu Province (see Figure 1) is located on the east coast of China, in the mid-latitude tude zone, with a geographical location between 30°46’–35°07’ N and 116°22’–121°55’ E. 0 0 0 0 zone, with a geographical location between 30 46 –35 07 N and 116 22 –121 55 E. It It lies in the climate transition zone between the subtropical and warm temperate zones lies in the climate transition zone between the subtropical and warm temperate zones and belongs to the East Asian monsoon climate zone. The average annual temperature, and belongs to the East Asian monsoon climate zone. The average annual temperature, precipitation, and sunshine hours in Jiangsu Province are between 13.6–16.1 °C, 704–1250 precipitation, and sunshine hours in Jiangsu Province are between 13.6–16.1 C, 704–1250 mm, and 1816–2503 h, respectively [33]. The terrain is generally ﬂat, with the Taihu Plain, mm, and 1816–2503 h, respectively [33]. The terrain is generally ﬂat, with the Taihu Plain, Yanjiang, and Lixia River areas being low-lying and having dense water networks. The Yanjiang, and Lixia River areas being low-lying and having dense water networks. The low mountains and hills account for only 14.33% and are mainly distributed in the west low mountains and hills account for only 14.33% and are mainly distributed in the west and north regions. There are various soil types in Jiangsu, including zonal soils such as and north regions. There are various soil types in Jiangsu, including zonal soils such as cinnamon, brown soil, yellow-brown soil, and yellow soil, and non-zonal soils such as cinnamon, brown soil, yellow-brown soil, and yellow soil, and non-zonal soils such as saline soil, meadow soil, and marsh soil. With a long history of agriculture, natural soil in saline soil, meadow soil, and marsh soil. With a long history of agriculture, natural soil in Jiangsu has evolved into various types of farming soil with diﬀerent soil textures under Jiangsu has evolved into various types of farming soil with different soil textures under the the inﬂuence of diﬀerent farming systems and utilization methods [34]. inﬂuence of different farming systems and utilization methods [34]. Fi Figure gure 1. 1. Ov Overview erview o of f th the e s study tudy area area of of J Jiangsu iangsu Prov Province, ince, China, a China, and nd its geographic its geographical al d distribution istribution map of soil moisture observation stations. map of soil moisture observation stations. 2.2. Data Source 2.2. Data Source Automatic moisture observation instruments have been gradually incorporated into Automatic moisture observation instruments have been gradually incorporated into the meteorological operational observation system since 2010, resulting in the availability the meteorological operational observation system since 2010, resulting in the availability of high regional density and continuous soil moisture observation data across Chinese of high regional density and continuous soil moisture observation data across Chinese provinces [35]. Consequently, daily 0–10 cm soil relative humidity data, measured by provinces [35]. Consequently, daily 0–10 cm soil relative humidity data, measured by 70 70 automatic soil moisture observation stations in Jiangsu Province from 2014 to 2022, along automatic soil moisture observation stations in Jiangsu Province from 2014 to 2022, along with with meteorological data collected by automatic weather stations and soil temperature data meteorological data collected by automatic weather stations and soil temperature data meas- measured by soil temperature instruments at the corresponding 70 soil moisture station ured by soil temperature instruments at the corresponding 70 soil moisture station locations, locations, were used for predicting 0–10 cm soil relative humidity. These atmospheric and soil observation data were obtained from the Jiangsu Meteorological Information Center. Based on the principle of soil water budget balance and considering the inﬂuence of various factors on the 0–10 cm soil relative humidity (RH , %), the predictive factors s10cm were divided into two categories: atmospheric and soil factors. There are ten atmospheric factors, including the mean air temperature (T , C), minimum air temperature (T , C), a amin Agriculture 2023, 13, 927 4 of 17 maximum air temperature (T , C), air relative humidity (RH , %), precipitation (P, mm), amax a sunshine hours (S, h), wind speed (W, ms ), atmospheric pressure (P , hPa), water vapor pressure (e, hPa), and potential evapotranspiration (ET , mm). Additionally, there are four soil factors, including the mean surface temperature (T , C), maximum soil surface temperature (T , C), minimum soil surface temperature (T , C), and 0–10 cm soil smax smin temperature (T , C). s10cm 2.3. Data Classiﬁcation Soil textures and hydrological constants varied signiﬁcantly in Jiangsu Province. Even when weather conditions are identical, different regions may exhibit distinct soil water dynamics due to the differences in soil physical properties [36]. Therefore, it is necessary to consider regional soil characteristics and hydrological constants when predicting soil moisture. To this end, according to the soil hydrological and physical characteristics measured by 70 automatic soil moisture observation stations in Jiangsu Province, the soil moisture observation data were classiﬁed into three categories: sandy soil, loam soil, and clay soil. The statistics of physical parameters corresponding to the different soil types are shown in Table 1. Table 1. Classiﬁcation results and corresponding soil physical characteristics of soil moisture obser- vation data. Soil Bulk Density Field Water Capacity Withering Humidity Soil Type Samples (gcm ) (%) (%) Sand 1.43 25.46 4.04 40,880 Loam 1.40 26.50 5.29 75,920 Clay 1.36 26.62 5.72 87,600 2.4. Methodology Description 2.4.1. Selection of Predictive Factors Soil relative humidity changes are mainly affected by previous and current weather conditions and the state of the soil itself. By distinguishing different soil types, we correlated RH with the averaged or accumulated value (including precipitation and sunshine s10cm hours) of 14 predictor factors on the same day as the soil moisture observed, and 1–10 days in the previous period, to determine the maximum impact time of each predictor (see Table 2). We used the time with the largest correlation coefﬁcient of each predictor as its maximum impact time on RH . The corresponding sample numbers for each soil type s10cm used to take correlation analysis are shown in Table 1. Table 2. List of predictor factors of 0–10 days prior, which are used for correlation analysis with RH . s10cm Names Units Descriptions Range Sunshine hours h Accumulated sunshine hours 0–128.6 Precipitation mm Cumulative precipitation 0–595.4 Evapotranspiration mm Averaged potential evapotranspiration 0.1–10.2 Wind speed ms Averaged wind speed 0–15.9 Relative humidity % Averaged mean air relative humidity 19–100 Averaged water vapor pressure 0.6–42.0 Pressure hPa Averaged atmospheric pressure 983.5–1042.4 Agriculture 2023, 13, 927 5 of 17 Table 2. Cont. Names Units Descriptions Range Averaged mean air temperature 11.1–36.0 Averaged minimum air temperature 15.6–31.9 Averaged maximum air temperature 7.2–40.9 Temperature Averaged mean soil surface temperature 7.0–45.8 Averaged minimum soil surface temperature 14.7–31.2 Averaged maximum soil surface temperature 0.9–70.2 Averaged 0–10 cm mean soil temperature 2.7–39.0 2.4.2. XGBoost Model The XGBoost is an ensemble learning method based on boosting [37]. The boosting technique combines multiple decision trees and aggregates their predictions to obtain a ﬁnal prediction that is more accurate than any individual tree. XGBoost is designed to prevent over-ﬁtting. The XGBoost model builds multiple trees sequentially, with each subsequent tree intended to reduce the errors of the previous tree. As the training proceeds iteratively, new trees are added to predict the error of the prior tree. Such a ﬁtting process is repeated several times until a stopping criterion is met, such as when the root mean square error (RMSE) reaches an asymptotic value. The ultimate prediction of the model is the sum of the predictions from all of the trees. The formula for the prediction at the step t and site location i can be deﬁned as follows [37]: (t1) y ˆ = f (x ) = y ˆ + f (x ) (1) å k i t i k=1 i (t1) where f (x ) is the tree model at step t, y ˆ and y ˆ are the predictions at steps t and t i i i t 1, and x are the predictor variables. The parameters of the model f (x ) are selected i i by optimizing the objective function, and the objective function is deﬁned by root mean square error. Additionally, XGBoost offers several other advanced features [37] that can further enhance the model’s performance. For instance, early stopping allows the training process to be stopped early if the performance on a validation set stops improving. This advanced feature prevents the model from overﬁtting to the training data and can improve its ability to generalize to new data. Cross-validation is another useful technique that can estimate the model’s generalization performance and help to select the optimal hyperparameters. By incorporating these and other advanced features, XGBoost has emerged as one of the most popular and inﬂuential machine learning models. The ﬂow chart depicting the XGBoost model is presented in Figure 2. 2.4.3. The Key Parameters of XGBoost Model In this study, we focused on optimizing several crucial parameters of the XGBoost algorithm, including the number of boost rounds, maximum depth, minimum weight in a child, and learning rate. The number of boost rounds determines the maximum number of boosting iterations, while the maximum depth sets the maximum depth of an individual tree. The minimum weight in a child parameter is utilized to prevent overﬁtting, and the learning rate parameter controls the model’s shrinkage at every step (i.e., a lower learning rate indicates more steps used to achieve the optimum) (see Figure 2). To optimize these parameters, we applied a tuning technique called grid search [38]. This approach computes the optimal values of hyperparameters by exhaustively searching over a range of possible parameter values. We utilized third-fold cross-validation [39] to evaluate the performance of different parameter combinations. In total, we searched through 1500 combinations of parameter values. Ultimately, our XGBoost model achieved Agriculture 2023, 13, 927 6 of 17 the best performance with the maximum depth, minimum weight needed in a child, and learning rate equal to 15, 10, and 0.02, respectively. In addition, we set the maximum number of boosting rounds to 5000 during training and used the early stop technique to Agriculture 2023, 13, x FOR PEER REVIEW 6 of 18 stop the training. The ﬁnal number of iterations was 4218 when the loss on the validation set no longer decreased. Figure 2. The ﬂowchart of the XGBoost model. Figure 2. The ﬂowchart of the XGBoost model. 2.4.4. Shapley Additive Explanations (SHAPs) 2.4.3. The Key Parameters of XGBoost Model SHAP is a local attribution method that is based on the use of Shapley values. The In this study, we focused on optimizing several crucial parameters of the XGBoost Shapley values originate from the ﬁeld of cooperative game theory and represent each algorithm, including the number of boost rounds, maximum depth, minimum weight in play’s average expected marginal contribution in a cooperative game after all possible a child, and learning rate. The number of boost rounds determines the maximum number combinations of players have been considered. It can be formulated as follows [40]: of boosting iterations, while the maximum depth sets the maximum depth of an individ- jSj!(FjSj 1)! ual tree. The minimum weight in a child parameter is utilized to prevent overﬁtt ing, and f = [ f (S[fig) f (S)] (2) i å x x F! the learning rate para Sm Fentfe igr controls the model’s shrinkage at every step (i.e., a lower learning rate indicates more steps used to achieve the optimum) (see Figure 2). where f is the weighted average of all marginal contributions of the predictor i, F is the total To optimize these parameters, we applied a tuning technique called grid search [38]. number of features, S is the subset of predictors from all predictors except for predictor i, This approach computes the optimal values of hyperparameters by exhaustively search- jSj!(FjSj1)! and is the weighting factor counting the number of permutations of the subset F! ing over a range of possible parameter values. We utilized third-fold cross-validation [39] S. f (S) is the expected output given the predictors subset S. [ f (S[fig) f (S)] is the x x x to evaluate the performance of diﬀerent parameter combinations. In total, we searched difference made by the predictor i. through 1500 combinations of parameter values. Ultimately, our XGBoost model achieved 2.4.5. Model Construction and Application the best performance with the maximum depth, minimum weight needed in a child, and learni This ng ra study te eq aimed ual to to15, develop 10, ana d soil 0.02 moistur , respec e tiv prediction ely. In ad model dition for , w dif e fer set t ent he m soila types ximum num- using relevant atmospheric and soil factors. To achieve this, 14 most related factors were ber of boosting rounds to 5000 during training and used the early stop technique to stop obtained by calculating the correlation. Additionally, to account for the different impacts of the training. The ﬁnal number of iterations was 4218 when the loss on the validation set soil types, the variable St was included in the model, with values of 1, 2, and 3 repre- ﬂag no longer decreased. senting sandy, loam, and clay soils, respectively. To further evaluate the importance of soil factors in predicting 0–10 cm soil relative 2.4.4. Shapley Additive Explanations (SHAPs) humidity, two sets of data used as the model’s independent variables were constructed using S 14 HAP optimal is a pr loc edictors al att ribu (including tion metatmospheric hod that is ba and sed soil on variabl the ues) se o and f Sh 10 apl optimal ey values. The predictors (including atmospheric variables only) from 70 stations in Jiangsu Province Shapley values originate from the ﬁeld of cooperative game theory and represent each between 2014 and 2021. Before prediction, missing values in these two data sets were com- play’s average expected marginal contribution in a cooperative game after all possible pleted with the mean values, and the dataset was normalized. A tri-fold cross-validation combinations of players have been considered. It can be formulated as follows [40]: approach [39] was employed to train, validate, and evaluate the model. The data were SF !( −−S 1)! φ=∪ [(fS {i})− f(S)] (2) ix  x F! SF ⊆ \{}i where φ is the weighted average of all marginal contributions of the predictor i , F is the total number of features, S is the subset of predictors from all predictors except for SF !( −−S 1)! predictor i , and is the weighting factor counting the number of permuta- F! tions of the subset S . is the expected output given the predictors subset S . f () S is the diﬀerence made by the predictor i . [( f Si∪− {}) f(S)] xx Agriculture 2023, 13, 927 7 of 17 randomly divided into three sets: 80% (163,520 samples) as the model training dataset, 10% (20,440 samples) as the model validation dataset for parameter optimization, and the remaining 10% (20,440 samples) as the model prediction evaluating dataset. 2.4.6. Model Prediction Effect Interpretation and Veriﬁcation After building the prediction model, the SHAP method was applied to obtain each predictive factor ’s positive and negative effects separately for both Model and _soil&atmo Model . In addition, six metrics were used on the evaluating dataset to evaluate the _atmo performance of XGBoost and other state-of-the-art predictive models, including correlation coefﬁcient (R), root mean square error (RMSE), mean absolute error (MAE), mean absolute relative error (MARE), Nash–Sutcliffe efﬁciency coefﬁcient (NSE), and accuracy (ACC). These indicators are calculated as follows [41]: (y y )(y ˆ y ˆ ) i i i i i=1 R = (3) n n (y y ) (y ˆ y ˆ ) å å i i i i i=1 i=1 u (y y ˆ ) i i i=1 RMSE = (4) MAE = j(y y ˆ )j (5) å i i i=1 1 (y y ˆ ) i i MARE = (6) n y i=1 (y y ˆ ) i i i=1 NSE = 1 (7) n i i=1 (y ˆ ) i=1 1 (y y ˆ ) i i ACC = 1 100% (8) n y i=1 where y is the observed value, y is the predicted value, n is the number of samples, y is i i the mean of observations, and y ˆ is the mean of the prediction. To further verify the prediction capabilities of Model and Model based _atmo _soil&atmo on XGBoost, we compared these models with three state-of-the-art machine learning models (i.e., ANN [42], RF [43], and SVM [44]) for soil moisture prediction over 70 sites in Jiangsu. The comparison was based on the values of these above metrics and the scatter distributions of predicted and observed soil moisture values. Furthermore, we evaluated the performance of Model and Model during a typical drought in August _atmo _soil&atmo 2022 in Jiangsu Province. The ﬂow chart depicting the establishment, interpretation, and evaluation of the prediction models for soil moisture is presented in Figure 3. Agriculture 2023, 13, x FOR PEER REVIEW 8 of 18 y yˆ where is the observed value, is the predicted value, n is the number of samples, i i y is the mean of observations, and y is the mean of the prediction. i i To further verify the prediction capabilities of Model_soil&atmo and Model_atmo based on XGBoost, we compared these models with three state-of-the-art machine learning models (i.e., ANN [42], RF [43], and SVM [44]) for soil moisture prediction over 70 sites in Jiangsu. The comparison was based on the values of these above metrics and the scatt er distribu- tions of predicted and observed soil moisture values. Furthermore, we evaluated the per- formance of Model_soil&atmo and Model_atmo during a typical drought in August 2022 in Agriculture 2023, 13, 927 8 of 17 Jiangsu Province. The ﬂow chart depicting the establishment, interpretation, and evalua- tion of the prediction models for soil moisture is presented in Figure 3. Figure 3. Flow chart of establishing, interpreting, and evaluating soil moisture models. Figure 3. Flow chart of establishing, interpreting, and evaluating soil moisture models. 3. Results 3. Results 3.1. Correlation Analysis between Soil Moisture and Predictive Factors 3.1. Correlation Analysis between Soil Moisture and Predictive Factors After analyzing the correlations between 0–10 cm soil relative humidity (RH ) and s10cm After analyzing the correlations between 0–10 cm soil relative humidity (RHs10cm) and various predictors for different soil types with different advance days (See Figure 4), it was various predictors for diﬀerent soil types with diﬀerent advance days (See Figure 4), it observed that, among the atmospheric factors, RH had a high positive correlation with s10cm was observed that, among the atmospheric factors, RHs10cm had a high positive correlation the mean air relative humidity (RH ) and cumulative precipitation (P ). The correlation a sum with the mean air relative humidity (RHa) and cumulative precipitation (Psum). The corre- coefﬁcients were between 0.17–0.33 and 0.13–0.26, respectively, and their absolute values gradually lation coeincr ﬃcients w eased with ere betw the leading een 0.time, 17–0.3 peaking 3 and 0.13 8–10 –0. days 26, respect prior. Additionally ively, and their ab , RH solute s10cm had a high negative correlation with the mean water vapor pressure (e) and accumulated values gradually increased with the leading time, peaking 8–10 days prior. Additionally, sunshine hours (S ). The absolute correlation coefﬁcients were between 0.24–0.33 and sum RHs10cm had a high negative correlation with the mean water vapor pressure (e) and accu- 0.15–0.33, respectively. The absolute values also increased with the leading time, reaching mulated sunshine hours (Ssum). The absolute correlation coeﬃcients were between 0.24– the maximum at 8 and 10 days prior, respectively. Among the soil factors, RH had a s10cm 0.33 and 0.15–0.33, respectively. The absolute values also increased with the leading time, high negative correlation with the mean maximum surface temperature (T ), with its Agriculture 2023, 13, x FOR PEER REVIEW smax 9 of 18 reaching the maximum at 8 and 10 days prior, respectively. Among the soil factors, RHs10cm maximum absolute value appearing 4–5 days prior. The correlations between RH and s10cm had a high negative correlation with the mean maximum surface temperature (Tsmax), with other factors were relatively low, but all passed the signiﬁcance test of p = 0.01. its maximum absolute value appearing 4–5 days prior. The correlations between RHs10cm and other factors were relatively low, but all passed the signiﬁcance test of . p =0.01 Figure 4. Correlation coefﬁcients between 0–10cm soil relative humidity and various predictive Figure 4. Correlation coeﬃcients between 0–10cm soil relative humidity and various predictive fac- Commented [M1]: 虚线，M 是否需要解释 factors of different soil types, which are (a) sandy soil, (b) loam, and (c) clay, respectively. tors of diﬀerent soil types, which are (a) sandy soil, (b) loam, and (c) clay, respectively. Overall, the correlations between RH and various predictor factors, as well as s10cm Overall, the correlations between RHs10cm and various predictor factors, as well as their change rules with the days advanced, were relatively consistent among different soil their change rules with the days advanced, were relatively consistent among diﬀerent soil types, with the times taken to reach the maximum value being similar (see Figure 4a–c). types, with the times taken to reach the maximum value being similar (see Figure 4a–c). The variabilities of positive–negative correlation with RH were mainly reﬂected in the s10cm factors of the minimum surface temperature and wind speed. Thus, a ﬁxed optimal impact The variabilities of positive–negative correlation with RHs10cm were mainly reﬂected in the factors of the minimum surface temperature and wind speed. Thus, a ﬁxed optimal impact time was set for each predictor factor as the model input, and its corresponding diﬀerences in the impact times between diﬀerent soil types were no longer distinguished. 3.2. Interpretability of Model We analyzed the relationships between the predictor variables and the soil moisture using the XGBoost model and presented the results through SHAP summary plots for each variable. In Figure 3, for each predictor variable displayed on the y-axis, each colored point represents a value of this variable in the dataset and the SHAP values displayed on the x-axis denoting the contributions of that predictor variable, which can be a positive or negative eﬀect on the prediction of soil moisture. The gradient color of each point indi- cates the value of the predictor variable, ranging from low (blue) to high (red), providing a visual representation of the relationships between the predictors and soil moisture. From the SHAP summary chart of Model_soil&atmo in Figure 5a, we observed that Tsmax, Ts10cm, and Tamax had a signiﬁcant negative contribution to the model prediction, consider- ing both atmospheric and soil variables. Conversely, the eﬀects of other factors on the prediction results were either opposite or insigniﬁcant. Among them, Psum had the most considerable positive contribution to the model prediction, followed by RHa. According to the importance of each predictor, the order of the top ﬁve predictors was Tsmax > Psum > Ts10cm > RHa > Ts. Figure 5. SHAP summary chart of (a) Model_soil&atmo and (b) Model_atmo. From the SHAP summary chart of Model_atmo in Figure 5b, we found that the greater value of Tamax, e, and W had a greater negative contribution to the model prediction, con- sidering only atmospheric variables. In contrast, other factors have opposite effects on the Agriculture 2023, 13, x FOR PEER REVIEW 9 of 18 Figure 4. Correlation coeﬃcients between 0–10cm soil relative humidity and various predictive fac- Commented [M1]: 虚线，M 是否需要解释 tors of diﬀerent soil types, which are (a) sandy soil, (b) loam, and (c) clay, respectively. Overall, the correlations between RHs10cm and various predictor factors, as well as their change rules with the days advanced, were relatively consistent among diﬀerent soil types, with the times taken to reach the maximum value being similar (see Figure 4a–c). The variabilities of positive–negative correlation with RHs10cm were mainly reﬂected in the factors of the minimum surface temperature and wind speed. Thus, a ﬁxed optimal impact time was set for each predictor factor as the model input, and its corresponding diﬀerences Agriculture 2023, 13, 927 9 of 17 in the impact times between diﬀerent soil types were no longer distinguished. 3.2. timeIwas nterpret set for abieach lity of Mode predictor l factor as the model input, and its corresponding differences in the impact times between different soil types were no longer distinguished. We analyzed the relationships between the predictor variables and the soil moisture using the XGBoost model and presented the results through SHAP summary plots for 3.2. Interpretability of Model each variable. In Figure 3, for each predictor variable displayed on the y-axis, each colored We analyzed the relationships between the predictor variables and the soil moisture point represents a value of this variable in the dataset and the SHAP values displayed on using the XGBoost model and presented the results through SHAP summary plots for th each e x- variable. axis deno In ti Figur ng t eh3 e , for con each tribu pr ti edictor ons ofvariable that pre displayed dictor va on ria the bley , w -axis, hiceach h can color be ed a positive or point represents a value of this variable in the dataset and the SHAP values displayed on negative eﬀect on the prediction of soil moisture. The gradient color of each point indi- the x-axis denoting the contributions of that predictor variable, which can be a positive or cates the value of the predictor variable, ranging from low (blue) to high (red), providing negative effect on the prediction of soil moisture. The gradient color of each point indicates a visual representation of the relationships between the predictors and soil moisture. the value of the predictor variable, ranging from low (blue) to high (red), providing a visual From the SHAP summary chart of Model_soil&atmo in Figure 5a, we observed that Tsmax, representation of the relationships between the predictors and soil moisture. Ts10cm, and Tamax had a signiﬁcant negative contribution to the model prediction, consider- From the SHAP summary chart of Model in Figure 5a, we observed that _soil&atmo ing both atmospheric and soil variables. Conversely, the eﬀects of other factors on the T , T , and T had a signiﬁcant negative contribution to the model prediction, smax amax s10cm considering both atmospheric and soil variables. Conversely, the effects of other factors prediction results were either opposite or insigniﬁcant. Among them, Psum had the most on the prediction results were either opposite or insigniﬁcant. Among them, P had sum considerable positive contribution to the model prediction, followed by RHa. According the most considerable positive contribution to the model prediction, followed by RH . to the importance of each predictor, the order of the top ﬁve predictors was Tsmax > Psum > According to the importance of each predictor, the order of the top ﬁve predictors was Ts10cm > RHa > Ts. T > P > T > RH > T . smax sum a s s10cm Figure 5. SHAP summary chart of (a) Model_ and (b) Model . soil&atmo _atmo Figure 5. SHAP summary chart of (a) Model_soil&atmo and (b) Model_atmo. From the SHAP summary chart of Model in Figure 5b, we found that the greater _atmo value of T , e, and W had a greater negative contribution to the model prediction, con- From the SHAP summary chart of Model_atmo in Figure 5b, we found that the greater amax sidering only atmospheric variables. In contrast, other factors have opposite effects on the value of Tamax, e, and W had a greater negative contribution to the model prediction, con- prediction results, or their positive–negative characteristics were insignificant. Among them, sidering only atmospheric variables. In contrast, other factors have opposite effects on the P had the most significant positive contribution to the model prediction, followed by RH , sum a which was consistent with the results of Model . According to the importance of each _soil&atm predictor, the order of the top five predictors was P > T > RH > e > W. sum amax a 3.3. Model Prediction Evaluation 3.3.1. Analysis of Model Prediction Accuracy To further verify the prediction capabilities of Model and Model based _soil&atmo _atmo on XGBoost, we compared them with three other state-of-the-art machine learning models (i.e., ANN, RF, and SVM) based on the scatter distributions of the predicted and observed values of soil moisture, and the values of six metrics (i.e., R, RMSE, MAE, MARE, MSE, and ACC). The scatter distributions of the model predictions based on XGBoost and the ac- tual observations of the 0–10 cm soil relative humidity are presented in Figure 6a1,a2. Model and Model showed an even distribution of predicted and observed _soil&atmo _atmo values around the 1:1 diagonal, with Model exhibiting a slightly more clustered _soil&atmo distribution. The mean and standard deviation of Model ’s predictions (79.28% and _soil&atmo Agriculture 2023, 13, x FOR PEER REVIEW 10 of 18 prediction results, or their positive–negative characteristics were insignificant. Among them, Psum had the most significant positive contribution to the model prediction, followed by RHa, which was consistent with the results of Model_soil&atm. According to the importance of each predictor, the order of the top five predictors was Psum > Tamax > RHa > e > W. 3.3. Model Prediction Evaluation 3.3.1. Analysis of Model Prediction Accuracy To further verify the prediction capabilities of Model_soil&atmo and Model_atmo based on XGBoost, we compared them with three other state-of-the-art machine learning models (i.e., ANN, RF, and SVM) based on the scatter distributions of the predicted and observed values of soil moisture, and the values of six metrics (i.e., R, RMSE, MAE, MARE, MSE, and ACC). The scatt er distributions of the model predictions based on XGBoost and the actual observations of the 0–10 cm soil relative humidity are presented in Figure 6a1,a2. Agriculture 2023, 13, 927 10 of 17 Model_soil&atmo and Model_atmo showed an even distribution of predicted and observed val- ues around the 1:1 diagonal, with Model_soil&atmo exhibiting a slightly more clustered dis- tribution. The mean and standard deviation of Model_soil&atmo ’s predictions (79.28% and 10.32%, respectively) were similar to those of the observations (79.30% and 15.77%, re- 10.32%, respectively) were similar to those of the observations (79.30% and 15.77%, respec- spectively). Model_atmo ’s prediction results were comparable to those of Model_soil&atmo, tively). Model ’s prediction results were comparable to those of Model , with _atmo _soil&atmo with only minor diﬀerences. However, overall, the prediction performance of only minor differences. However, overall, the prediction performance of Model _soil&atmo Model_soil&atmo was slightly bett er than that of Model_atmo. was slightly better than that of Model . _atmo Figure 6. Scatter plot of soil moisture observations and predictions of Model and Model _atmo Figure 6. Scatt er plot of soil moisture observations and predictions of Model_soil&_soil&atmo atmo and Model_atmo based on (a1,a2) XGBoost, (b1,b2) ANN, (c1,c2) RF, and (d1,d2) SVM. (The 1:1 diagonal is shown by based on (a1,a2) XGBoost, (b1,b2) ANN, (c1,c2) RF, and (d1,d2) SVM. (The 1:1 diagonal is shown by the gray dashed line, the regression line is shown by the red solid line, and the observed and predicted means and standard deviations are shown by the red dots and dashed boxes, respectively). After comparing the scatter distributions of observations with model predictions based on XGBoost, ANN, RF, and SVM (see Figure 6), it was observed that the lines between the predicted and observed soil moisture for XGBoost were much closer to the ideal line (y = x) than those for the other predictive models. Additionally, the prediction results of the other models presented a relatively smaller standard deviation. Table 3 shows the comprehensive predictive performances of XGBoost, ANN, RF, and SVM over 70 sites in Jiangsu Province. The values of R, RMSE, MAE, MARE, NSE, and ACC for Model and Model based on XGBoost were 0.69, 11.11, 4.87, 0.12, _soil&atmo _atmo 0.50, and 88%, as well as 0.66, 11.49, 4.96, 0.14, 0.47, and 86%, respectively. Comparing the values of the six evaluated indexes of other LM models, it was found that models based on XGBoost always had the lowest RMSE, MAE, and MARE, as well as the highest R, NSE, and ACC. In addition, for XGBoost, compared with Model having an average prediction _atmo accuracy of 86%, Model had better precision, with an average accuracy of 88%. _soil&atmo Notably, Model ’s prediction effects were always slightly better than those of _soil&atmo Model , which was also evident from the prediction results of other models, whether _atmo from the scatter charts or metrics. Agriculture 2023, 13, 927 11 of 17 Table 3. Comparison of XGBoost, ANN, RF, and SVM performances in soil moisture prediction using two data sets as the model’s input. ML Models R RMSE MAE MARE NSE ACC (%) Model 0.69 11.11 4.87 0.12 0.50 88% _soil&atmo XGBoost Model 0.66 11.49 4.96 0.14 0.47 86% _atmo Model 0.59 12.85 6.55 0.16 0.27 84% _soil&atmo ANN Model 0.56 13.19 6.71 0.17 0.23 83% _atmo Model 0.64 12.08 6.07 0.15 0.36 85% _soil&atmo RF Model 0.63 12.25 6.19 0.16 0.34 84% _atmo Model 0.54 13.68 7.56 0.17 0.19 83% _soil&atmo SVM Model 0.51 13.58 6.86 0.18 0.18 82% _atmo Furthermore, the spatial distribution map of the model evaluation indexes (i.e., R and MAE) showed that both Model and Model based on XGBoost had a _atmo _soil&atmo high accuracy in soil moisture prediction, and their spatial distribution patterns were very similar, with differences only at individual stations (see Figure 7). Stations with relatively small correlation coefﬁcients and large average absolute errors of predictions Agriculture 2023, 13, x FOR PEER REVIEW 12 of 18 and observations of both models were mainly concentrated along the northern area of the Yangtze River and in the northeastern area of Jiangsu Province. Figure Figure 7. 7. Spatial Spatial distribution distribution of of predic prediction tion ac accuracy curacy ev evaluation aluation indicators indicators ofof (a ()aModel ) Model_soil&atmo and and _soil&atmo (b) Model_atmo. (b) Model . _atmo In addition, we found that the prediction accuracy of both models varied greatly 3.3.2. Analysis of Typical Drought Process between sites from the spatial distribution maps. According to the statistical analysis, for During 2–23 August 2022, a third round of persistent high temperature occurred in Model , the R between the predicted and measured values ranged from 0.34 to _soil&atmo Jiangsu Province, with the ﬁrst two rounds taking place on 16–22 June and 8–15 July, re- 0.87, with a mean value of 0.69, and the MAE ranged from 0.12% to 14.52%, with a mean spectively. The south of Huaihe region experienced 14–19 days of a maximum tempera- value of 4.87%. The number of sites with R > 0.60 reached 58, accounting for more than ture ≥ 37 °C, with the average temperature between 32–33.7 °C. Compared to the same 82%, and the number of sites with MAE < 5% reached 40, accounting for more than 57%. period in a normal year, the temperature in 2022 was approximately 4 °C higher and the For Model , the R between the predicted and measured values ranged from 0.34 to 0.85, _atmo precipitation was less than 90%. In particular, southern Jiangsu faced widespread high with an average value of 0.66, and the MAE ranged from 0.05% to 13.96%, with an average temperatures above 40 °C from 12–15 August, resulting in a rapid expansion of drought value of 5.04%. The number of sites with R > 0.60 reached 53, accounting for more than across the province. By 15 August, most of the southern Huaihe Basin experienced mod- 75%, and the number of sites with MAE < 5% reached 38, accounting for more than 50%. erate or above meteorological drought, with some areas suﬀering from severe drought. However, the high temperature gradually receded from 24 August, and the precipitation 3.3.2. Analysis of Typical Drought Process gradually increased, mainly in the Huaibei and Sunan areas. As a result, the moisture During 2–23 August 2022, a third round of persistent high temperature occurred conditions across the province improved eﬀectively, and the moisture content reached an in Jiangsu Province, with the ﬁrst two rounds taking place on 16–22 June and 8–15 July, appropriate level. respectively. The south of Huaihe region experienced 14–19 days of a maximum tempera- According to the distribution of a 0–10 cm soil relative humidity on 1, 15, and 30 ture 37 C, with the average temperature between 32–33.7 C. Compared to the same August, which was interpolated from the measurement of the automatic soil moisture sta- period in a normal year, the temperature in 2022 was approximately 4 C higher and the tion (see Figure 8a1–a3), we found on 1 August, aﬀected by antecedent precipitation, the precipitation was less than 90%. In particular, southern Jiangsu faced widespread high soil moisture in most areas of northern Jiangsu was saturated, and the ﬁeld humidity was relatively high, while the 0–10 cm soil relative humidity in some areas of southern Jiangsu was less than 60%. By August 15th, there was a severe soil water shortage in most of the southern Huaihe Basin. The 0–10 cm soil relative humidity was only 40% to 50%, which had reached moderate drought, and was even less than 40% in some regions, reaching severe drought. Aﬀected by precipitation, by 30 August, the ﬁeld soil humidity in some areas of Huaibei was relatively high, and the 0–10 cm soil relative humidity in most south- ern Huaihe Basin had generally improved to more than 60%, with only sporadic areas still suﬀering from the drought. Thus, it can be seen that the variation in farmland drought perfectly corresponds with the beginning, aggravation, and extinction of the entire high- temperature process. Agriculture 2023, 13, 927 12 of 17 temperatures above 40 C from 12–15 August, resulting in a rapid expansion of drought across the province. By 15 August, most of the southern Huaihe Basin experienced mod- erate or above meteorological drought, with some areas suffering from severe drought. However, the high temperature gradually receded from 24 August, and the precipitation gradually increased, mainly in the Huaibei and Sunan areas. As a result, the moisture conditions across the province improved effectively, and the moisture content reached an appropriate level. According to the distribution of a 0–10 cm soil relative humidity on 1, 15, and 30 Au- gust, which was interpolated from the measurement of the automatic soil moisture station (see Figure 8a1–a3), we found on 1 August, affected by antecedent precipitation, the soil moisture in most areas of northern Jiangsu was saturated, and the ﬁeld humidity was relatively high, while the 0–10 cm soil relative humidity in some areas of southern Jiangsu was less than 60%. By 15th August, there was a severe soil water shortage in most of the southern Huaihe Basin. The 0–10 cm soil relative humidity was only 40% to 50%, which had reached moderate drought, and was even less than 40% in some regions, reach- ing severe drought. Affected by precipitation, by 30 August, the ﬁeld soil humidity in some areas of Huaibei was relatively high, and the 0–10 cm soil relative humidity in most southern Huaihe Basin had generally improved to more than 60%, with only sporadic Agriculture 2023, 13, x FOR PEER REVIE areas W still suffering from the drought. Thus, it can be seen that the variation in farmland 13 of 18 drought perfectly corresponds with the beginning, aggravation, and extinction of the entire high-temperature process. Figure 8. Relative humidity of 10 cm soil relative humidity of (a1–a3) observations, (b1–b3) Figure 8. Relative humidity of 10 cm soil relative humidity of (a1–a3) observations, (b1–b3) Model_soil&atmo predictions, and (c1–c3) Model_atmo predictions on 1, 15, and 30 August 2022. Model predictions, and (c1–c3) Model predictions on 1, 15, and 30 August 2022. _soil&atmo _atmo The spatial distribution patt erns of the corresponding prediction results of the mod- The spatial distribution patterns of the corresponding prediction results of the models els agreed with the observation results. The prediction results reﬂected not only the de- agreed with the observation results. The prediction results reﬂected not only the devel- velopment process of drought but also the distribution areas of diﬀerent levels of farm- opment process of drought but also the distribution areas of different levels of farmland dr lan ought. d droug However ht. Howe , v the er, the predicted predicte dr d dr ought ough situation t situatiowas n war s elatively relativelyweak weak c compar ompared ed t to o the observation results. Overall, the diﬀerences in the distribution patt ern and numeric the observation results. Overall, the differences in the distribution pattern and numeric value between the predictions and observations of Model_soil&atmo were less than those of Model_atmo (see Figure 8b1–b3 and Figure 8c1–c3, respectively). 4. Discussion Based on the observation, soil types, and meteorological data, this study adopted XGBoost to predict soil moisture variations. Diﬀerent atmospheric and soil factor combi- nations were selected as input variables to establish two sets of prediction models (Model_soil&atmo and Model_atmo) for RHs10cm. At the same time, the contributions of the pre- dictive factors were discussed using SHAP. The prediction accuracy was evaluated by comparing six evaluated indexes with other popular ML methods and analyzing a typical drought process in 2022. The variation in soil moisture is a complex coupling system that exhibits high noise, nonlinearity, and unstable random time series data [22]. Compared to traditional statisti- cal models, machine learning algorithms use multiple processing layers consisting of com- plex structures or multiple nonlinear transformations to highly abstract data, which could overcome the inﬂuence of white noise on the prediction accuracy and eﬀectively improve the simulation accuracy [25]. However, diﬀerent ML methods have diﬀerent applicabili- ties for the same dataset. For example, in a study predicting soil moisture based on three diﬀerent datasets, machine learning techniques such as multiple linear regression (MLR), support vector regression (SVR), and recurrent neural networks (RNNs) were compared, and MLR was found to have a bett er performance than the others. Our study used auto- matic soil moisture observations to compare the prediction accuracies of two models Agriculture 2023, 13, 927 13 of 17 value between the predictions and observations of Model were less than those of _soil&atmo Model (see Figure 8b1–b3 and Figure 8c1–c3, respectively). _atmo 4. Discussion Based on the observation, soil types, and meteorological data, this study adopted XG- Boost to predict soil moisture variations. Different atmospheric and soil factor combinations were selected as input variables to establish two sets of prediction models (Model _soil&atmo and Model ) for RH . At the same time, the contributions of the predictive factors _atmo s10cm were discussed using SHAP. The prediction accuracy was evaluated by comparing six evaluated indexes with other popular ML methods and analyzing a typical drought process in 2022. The variation in soil moisture is a complex coupling system that exhibits high noise, nonlinearity, and unstable random time series data [22]. Compared to traditional statistical models, machine learning algorithms use multiple processing layers consisting of complex structures or multiple nonlinear transformations to highly abstract data, which could overcome the inﬂuence of white noise on the prediction accuracy and effectively improve the simulation accuracy [25]. However, different ML methods have different applicabilities for the same dataset. For example, in a study predicting soil moisture based on three different datasets, machine learning techniques such as multiple linear regression (MLR), support vector regression (SVR), and recurrent neural networks (RNNs) were compared, and MLR was found to have a better performance than the others. Our study used automatic soil moisture observations to compare the prediction accuracies of two models based on XGBoost with ANN, RF, and SVM. It showed that Model based on _soil&atm XGBoost was superior, providing the lowest RMSE (11.11), MAE (4.87), and MARE (0.12), and highest R (0.69), NSE (0.50), and ACC (88%). Due to different research and application purposes, the dataset applied in soil moisture prediction studies based on machine learning algorithms is varied, including in situ sites [45], remote sensing [46], reanalysis [47], and ﬂux stations [24]. These datasets usually belong to diverse regions with different spatial and temporal resolutions, so it is still challenging to make direct comparisons even if the same method is applied. The analysis of a typical drought process showed that the XGBoost model based on site data had a good performance and was a feasible method for soil water content prediction, as it could capture a reasonable spatial distribution of the soil moisture. In addition, several advantages were considered for choosing the data observed from the automatic observation stations. Firstly, for a speciﬁc site, the data of the automatic observation station have lower errors than the data obtained by remote sensing instruments and reanalysis data, where the problems of insufﬁcient time resolution and delayed acquisition also exist [47]. Hence, we can more accurately explore the relationship between soil moisture and environmental parameters. Secondly, soil moisture and its related meteorological or soil data are commonly available with the exact temporal resolution, so abundant data could be provided for training the predictive model. It is important to note that the predictivity of soil moisture depends on the data’s time steps and spatial resolutions due to their different distribution and variation [24,48]. Moreover, the wideness of the application of soil moisture prediction usually depends on its spatial representativeness. Therefore, as more automatic weather stations are installed, the proposed model based on site data could be helpful for the operational studies on soil moisture prediction over larger regions and could provide information for timely and optimal irrigation scheduling. However, considering the spatial variability of soil moisture, in-depth future research is still needed, using situ data, remote sensing, and reanalysis data. The appropriate selection of model input factors could promote the accuracy of the prediction model [49]. In this research, we correlated the RH with 14 predictors s10cm 1–10 days before to determine each predictor ’s maximum impact time. The selected predictors were taken as inputs for the model, which would make the model establishment more reasonable, but still needs to be tested in the future. In addition, the contributions of Agriculture 2023, 13, 927 14 of 17 each predictor on the modeling results of two sets of models were discussed via SHAP. The analysis revealed that soil factors in Model played a positive role in the prediction _soil&atmo of soil moisture. Overall, the prediction accuracy of Model was higher than that of _soil&atmo Model . Therefore, introducing soil factors such as T , T , and T could improve _atmo smax s s10cm the prediction accuracy of soil moisture to some extent. For atmospheric factors, T , amax P , and RH are crucial for improving the soil moisture prediction accuracy. These sum a results are consistent with the view that temperature and precipitation are the main factors affecting the variations in soil moisture by adjusting the water budget balance [50,51]. This study aimed to predict the 0–10cm soil relative humidity, which is a crucial parameter for drought and waterlogging prevention, as well as farmland fertilization and irrigation. Generally, the cultivation layer of crops is 0–20 cm, and the water condition of this layer has a good characterization of crop drought. However, compared with the deep soil layer, the 0–10 cm soil layer is more directly affected by meteorological conditions such as precipitation and temperature. When the temperature is high and the amount of evapotranspiration increases, the lack of moisture in crop ﬁelds appears gradually from top to bottom. The moisture deﬁcit in surface soil is easily detected and can serve as the evaluation index for preventing and controlling crop drought. In addition, there is an excellent linear correlation between the soil relative moisture at different levels of depth [52], and hence the surface soil moisture condition is a good indicator of deep soil moisture conditions. This study deeply integrated the XGBoost with meteorological data to establish a provincial-level soil moisture prediction model, which can provide a reference for soil moisture prediction research in other regions. The model can be used to analyze historical soil water change rules and typical drought and ﬂood cases during the period lacking soil moisture observation while high-density meteorological observation is available (mainly from the 1960s to 2010s). However, there are some deﬁciencies and uncertainties in this study. For instance, only four frequently used machine learning algorithms were used in the study. In the future, multiple machine learning algorithms or other methods [53–55] could be used to conduct soil moisture prediction research to analyze the advantages and disadvantages of different methods and applicable conditions. Based on the XGBoost algorithm, the positive and negative contributions of most factors in the Model and _soil&atmo Model for soil moisture prediction analyzed by SHAP were consistent and conformed _atmo to the actual physical meaning. However, there were some cases where the same factor had the opposite contribution to the prediction results, which needs further investigation. 5. Conclusions Soil moisture is the characterization of farmland drought and ﬂood and the basis for irrigation schemes. The prediction of soil relative humidity was achieved based on the XGBoost model using continuous daily atmospheric and soil observation data from automatic stations. The methods of correlation analysis and SHAP were applied to select model predictors and evaluate the contribution of model factors. In addition, six effect indicators and a typical drought process were analyzed to compare the predictive accuracy of the XGBoost model with the other three machine learning models (i.e., ANN, RF, and SVM) to assess the predictive power of the model. Through correlative analysis, we found that the time with the highest correlations between environmental predictors and RH varied but was similar between soil types. s10cm Among atmospheric factors, the mean RH and P exhibited strong positive correlations a sum with RH , with correlation coefﬁcients ranging from 0.17 to 0.33 and 0.13 to 0.26. The s10cm correlation gradually increased over time, reaching the maximum 8~10 days ago. On the other hand, the mean e and S displayed strong negative correlations with RH , sum s10cm with correlation coefﬁcients ranging from 0.24 to 0.33 and from 0.15 to 0.33. Their absolute values also gradually increased over time, peaking at the time of 8 days ago and 10 days ago, respectively. Among the soil factors, the mean T showed a strong smax negative correlation with RH , and its maximum absolute value appeared 4~5 days s10cm Agriculture 2023, 13, 927 15 of 17 ago. Furthermore, via SHAP analysis, it showed that the contributions and impacts of the predictors on the modeling results for Model and Model were different. _atmo _soil&atmo According to the importance of each predictor, the orders of the top ﬁve predictors of these two models were T > P > T > RH > T and P > T > RH > e > W, smax sum s10cm a s sum amax a respectively. Overall, among the predictors, the contribution rates of T , P , and RH amax sum a in atmospheric factors, which functioned as a critical factor affecting the variation in soil moisture, were relatively high in both models. The overall performances of Model and Model based on XGBoost ex- _soil&atmo _atmo hibited lower error values when compared to ANN, RF, and SVM, thereby verifying the prediction capabilities of the XGBoost model. The values of R, RMSE, MAE, MARE, NSE, and ACC for Model and Model based on XGBoost were 0.69, 11.11, 4.87, 0.12, _atmo _soil&atmo 0.50, and 88%, and 0.66, 11.49, 4.96, 0.14, 0.47, and 86%, respectively. Both Model _soil&atmo and Model using XGBoost outperformed the other machine learning models in the _atmo scatter distribution of the predicted and measured values. In addition, by integrating the re- sults of SHAP analysis and comparisons of Model and Model , it showed that _soil&atmo _atmo Model ’s prediction effects were always slightly better than those of Model . _atmo _soil&atmo Hence, it is worth noting that introducing soil factors (e.g., T , T , and T ) can smax s s10cm positively improve the soil moisture prediction accuracy. Furthermore, the XGBoost model was applicable for provincial-level soil moisture prediction as it captured the spatial distribution characteristics of different levels of drought and effectively predicted the dynamic change process of the “occurrence–development– termination” of a speciﬁc drought event. Therefore, the excellent establishment of a soil moisture prediction model based on automatic observation stations, which effectively overcomes the temporary discontinuity of remote sensing inversion and the problem of a low prediction accuracy, could not only effectively guide farmland irrigation but also validly compensate for the insufﬁcient historical observation of soil moisture stations. Author Contributions: Conceptualization, Y.R. and Y.W.; methodology, F.L.; software, Y.R. and F.L.; validation, Y.R. and F.L.; formal analysis, F.L.; investigation, Y.R. and Y.W.; resources, Y.R.; data curation, Y.R.; writing—original draft preparation, Y.R. and F.L.; writing—review and editing: Y.W.; visualization: Y.R. and F.L.; supervision: Y.W.; project administration: Y.W. All authors have read and agreed to the published version of the manuscript. Funding: This research was funded by the National Natural Science Foundation of China Project (41805049). Institutional Review Board Statement: Not applicable. Data Availability Statement: The model prediction results presented in this study are available upon request from the corresponding author. The original observations are not publicly available due to the privacy policy. Acknowledgments: We thank the editors and reviewers for their comments to improve our manuscript. Conﬂicts of Interest: The authors declare no conﬂict of interest. References 1. Ahmad, N.; Malagoli, M.; Wirtz, M.; Hell, R. Drought stress in maize causes differential acclimation responses of glutathione and sulfur metabolism in leaves and roots. BMC Plant Biol. 2016, 16, 247. [CrossRef] [PubMed] 2. Isabel Ferreira, M.; Valancogne, C. Experimental Study of a Stress Coefﬁcient: Application on a Simple Model for Irrigation Scheduling and Daily Evapotranspiration Estimation. IFAC Proc. Vol. 1997, 30, 33–38. [CrossRef] 3. Dai, Y.; Zeng, X.; Dickinson, R.E.; Baker, I.; Bonan, G.B.; Bosilovich, M.G.; Denning, A.S.; Dirmeyer, P.A.; Houser, P.R.; Niu, G.; et al. The Common Land Model. Bull. Am. Meteorol. Soc. 2003, 84, 1013–1024. [CrossRef] 4. Kunstmann, H.; Jung, G.; Wagner, S.; Clottey, H. Integration of atmospheric sciences and hydrology for the development of decision support systems in sustainable water management. Phys. Chem. Earth Parts A/B/C 2008, 33, 165–174. [CrossRef] 5. Dan, B.; Zheng, X.; Wu, G. Assimilating Shallow Soil Moisture Observations into Land Models with a Water Budget Constraint. Hydrol. Earth Syst.Sci. 2020, 24, 5187–5201. [CrossRef] 6. Robinson, J.M.; Hubbard, K.G. Soil Water Assessment Model for Several Crops in the High Plains. Agron. J. 1990, 82, 1141–1148. [CrossRef] Agriculture 2023, 13, 927 16 of 17 7. Mahmood, R.; Hubbard, K.G. An Analysis of Simulated Long-Term Soil Moisture Data for Three Land Uses under Contrasting Hydroclimatic Conditions in the Northern Great Plains. J. Hydrometeorol. 2004, 5, 160–179. [CrossRef] 8. Zhang, X.; Ma, Y.H.; Anlauf, R. Forecast and Analysis of Soil Moisture Based on SIMPEL model. J. Agric. Sci. Technol. 2013, 14, 490–493. 9. Holland, J.E.; Biswas, A. Predicting the mobile water content of vineyard soils in New South Wales, Australia. Agric. Water Manag. 2015, 148, 34–42. [CrossRef] 10. Hu, W.; Si, B.C. Soil water prediction based on its scale-speciﬁc control using multivariate empirical mode decomposition. Geoderma 2013, 193–194, 180–188. [CrossRef] 11. Prasad, R.; Ravinesh, C.; Li, Y.; Maraseni, T. Weekly soil moisture forecasting with multivariate sequential, ensemble empirical mode decomposition and Boruta-random forest hybridizer algorithm approach. Catena 2019, 177, 149–166. [CrossRef] 12. Shoaib, M.; Shamseldin, A.Y.; Melville, B.W.; Khan, M.M. A comparison between wavelet based static and dynamic neural network approaches for runoff prediction. J. Hydrol. 2016, 535, 211–225. [CrossRef] 13. Kamilaris, A.; Francesc, X. Deep learning in agriculture: A survey. Comput. Electron. Agric. 2018, 147, 70–90. [CrossRef] 14. Yalcin, H. An Approximation for A Relative Crop Yield Estimate from Field Images Using Deep Learning. In Proceedings of the International Conference on Agro-Geoinformatics (Agro-Geoinformatics), Istanbul, Turkey, 16–19 July 2019. 15. Yu, J.; Tang, S.; Zhangzhong, L.; Zheng, W.; Xu, L. A Deep Learning Approach for Multi-Depth Soil Water Content Prediction in Summer Maize Growth Period. IEEE Access 2020, 8, 199097–199110. [CrossRef] 16. Fathi, M.T.; Ezziyyani, M.; Ezziyyani, M.; Mamoune, S.E. Crop Yield Prediction Using Deep Learning in Mediterranean Region. In Proceedings of the Advanced Intelligent Systems for Sustainable Development (AI2SD’2019), Marrakech, Morocco, 8–11 July 2019. 17. Ji, R.; Li, X.; Zhang, S.; Zheng, L. Prediction of soil moisture in multiple depth based on time delay neural network. Trans. Chin. Soc. Agric. Eng. 2017, 33, 132–136. 18. Gill, M.K.; Asefa, T.; Kemblowski, M.W.; McKee, M. Soil moisture predition using support vector machines. J. Am. Water Resour. Assoc. 2006, 42, 1033–1046. [CrossRef] 19. Pan, J.; Shangguan, W.; Li, L.; Yuan, H.; Zhang, S.; Lu, X.; Wei, N.; Dai, Y. Using data-driven methods to explore the predictability of surface soil moisture with FLUXNET site data. Hydrol. Process. 2019, 33, 2978–2996. [CrossRef] 20. Tharani, P.P.; Baranidharan, B. An Analysis on Application of Deep Learning Techniques for Precision Agriculture. In Pro- ceedings of the International Conference on Inventive Research in Computing Applications (ICIRCA), Coimbatore, India, 2–4 September 2021. 21. Gumiere, S.J.; Camporese, M.; Botto, A.; Lafond, J.A.; Paniconi, C.; Gallichand, J.; Rousseau, A.N. Machine Learning vs. Physics-Based Modeling for Real-Time Irrigation Management. Front. Water 2020, 2, 8. [CrossRef] 22. Li, P.; Zha, Y.; Shi, L.; Tso, C.-H.; Zhang, Y.; Zeng, W. Comparison of the use of a physical-based model with data assimilation and machine learning methods for simulating soil water dynamics. J. Hydrol. 2020, 584, 124692. [CrossRef] 23. Liu, D.; Liu, C.; Tang, Y.; Gong, C. A GA-BP Neural Network Regression Model for Predicting Soil Moisture in Slope Ecological Protection. Sustainability 2022, 14, 1386. [CrossRef] 24. Li, Q.; Li, Z.; Shangguan, W.; Wang, X.; Li, L.; Yu, F. Improving soil moisture prediction using a novel encoder-decoder model with residual learning. Comput. Electron. Agric. 2022, 195, 106816. [CrossRef] 25. Prakash, S.; Sharma, A.; Sahu, S.S. Soil Moisture Prediction Using Machine Learning. In Proceedings of the Second International Conference on Inventive Communication and Computational Technologies (ICICCT), Coimbatore, India, 20–21 April 2018. 26. Adeyemi, O.; Grove, I.; Peets, S.; Domun, Y.; Norton, T. Dynamic Neural Network Modelling of Soil Moisture Content for Predictive Irrigation Scheduling. Sensors 2018, 18, 3408. [CrossRef] [PubMed] 27. Xu, J.W.; Zhao, J.F.; Zhang, W.C.; Xu, X.X. A Novel Soil Moisture Predicting Method Based on Artiﬁcial Neural Network and Xinanjiang Model. Adv. Mater. Res. 2010, 121–122, 1028–1032. [CrossRef] 28. Li, N.; Zhang, Q.; Yang, F.X.; Deng, Z.L. Research of adaptive genetic neural network algorithm in soil moisture prediction. Comput. Eng. Appl. 2018, 54, 54–59+69. 29. Notarnicola, C.; Angiulli, M.; Posa, F. Soil moisture retrieval from remotely sensed data: Neural network approach versus Bayesian method. IEEE Trans. Geosci. Remote Sens. 2008, 46, 547–557. [CrossRef] 30. Wei, W.; Zhang, J.; Zhou, L.; Xie, B.; Zhou, J.; Li, C. Comparative evaluation of drought indices for monitoring drought based on remote sensing data. Environ. Sci. Pollut. Res. 2021, 28, 20408–20425. [CrossRef] 31. Sandholt, I.; Rasmussen, K.; Andersen, J. A simple interpretation of the surface temperature/vegetation index space for assessment of surface moisture status. Remote Sens. Environ. 2002, 79, 213–224. [CrossRef] 32. Zheng, W.; Zhangzhong, L.; Zhang, X.; Wang, C.; Zhang, S.; Sun, S.; Niu, H. A Review on the Soil Moisture Prediction Model and Its Application in the Information System. In Proceedings of the Computer and Computing Technologies in Agriculture XI, Jilin, China, 12–15 August 2017. 33. Jiang, A.J.; Peng, H.Y.; Wang, B.M. The analyses of Jiangsu climate variety in forty years. J. Meteorol. Sci. 2006, 26, 525–529. 34. Qi, Y.; Darilek, J.L.; Huang, B.; Zhao, Y.; Sun, W.; Gu, Z. Evaluating soil quality indices in an agricultural region of Jiangsu Province, China. Geoderma 2009, 149, 325–334. [CrossRef] 35. Wang, J.Q.; Zhao, Y.F.; Ren, Z.H.; Gao, J. Design and Veriﬁcation of Quality Control Methods for Automatic Soil Moisture Observation Data in China. Meteorology 2018, 44, 244–257. Agriculture 2023, 13, 927 17 of 17 36. Wang, S.; Fu, G. Modelling soil moisture using climate data and normalized difference vegetation index based on nine algorithms in alpine grasslands. Front. Environ. Sci. 2023, 11, 1130448. [CrossRef] 37. Chen, T.; Guestrin, C. XGBoost: A Scalable Tree Boosting System. In Proceedings of the ACM, San Francisco, CA, USA, 13–17 August 2016. 38. Bergstra, J.; Bengio, Y. Random search for hyper-parameter optimization. J. Mach. Learn. Res. 2012, 13, 281–305. 39. Kohavi, R. A study of cross-validation and bootstrap for accuracy estimation and model selection. Int. Jt. Conf. Artif. Intell. 1995, 14, 1137–1145. 40. Eisenman, R.L. A proﬁt-sharing interpretation of shapley value for n-person games. Syst. Res. Behav. Sci. 1967, 12, 396–398. [CrossRef] 41. Niazkar, M. Assessment of artiﬁcial intelligence models for calculating optimum properties of lined channels. J. Hydroinform. 2020, 22, 1410–1423. [CrossRef] 42. Agatonovic-Kustrin, S.; Beresford, R. Basic concepts of artiﬁcial neural network (ANN) modeling and its application in pharma- ceutical research. J. Pharm. Biomed. Anal. 2000, 22, 717–727. [CrossRef] 43. Biau, G. Analysis of a random forests model. J. Mach. Learn. Res. 2012, 13, 1063–1095. 44. Cherkassky, V.; Ma, Y. Practical selection of SVM parameters and noise estimation for SVM regression. Neural Netw. 2004, 17, 113–126. [CrossRef] 45. Matei, O.; Rusu, T.; Petrovan, A.; Mihu¸ t, G. A Data Mining System for Real Time Soil Moisture Prediction. Procedia Eng. 2017, 181, 837–844. [CrossRef] 46. Nguyen, T.T.; Ngo, H.H.; Guo, W.; Chang, S.W.; Nguyen, D.D.; Nguyen, C.T.; Zhang, J.; Liang, S.; Bui, X.T.; Hoang, N.B. A low-cost approach for soil moisture prediction using multi-sensor data and machine learning algorithm. Sci. Total Environ. 2022, 833, 155066. [CrossRef] 47. Filipovi, N.; Brdar, S.; Mimi, G.; Marko, O.; Crnojevi, V. Regional soil moisture prediction system based on long short-term memory network. Biosyst. Eng. 2022, 213, 30–38. [CrossRef] 48. Li, Q.; Zhu, Y.; Shangguan, W.; Wang, X.; Li, L.; Yu, F. An attention-aware LSTM model for soil moisture and soil temperature prediction. Geoderma 2022, 409, 115651. [CrossRef] 49. Cai, Y.; Zheng, W.; Zhang, X.; Zhangzhong, L.; Xue, X. Research on soil moisture prediction model based on deep learning. PLoS ONE 2019, 14, e0214508. [CrossRef] [PubMed] 50. Bell, J.E.; Sherry, R.; Luo, Y. Changes in soil water dynamics due to variation in precipitation and temperature: An ecohydrological analysis in a tallgrass prairie. Water Resour. Res. 2010, 46, W03523. [CrossRef] 51. Feng, H.; Liu, Y. Combined effects of precipitation and air temperature on soil moisture in different land covers in a humid basin. J. Hydrol. 2015, 531, 1129–1140. [CrossRef] 52. Ragab, R. Towards a continuous operational system to estimate the root-zone soil moisture from intermittent remotely sensed surface moisture. J. Hydrol. 1995, 173, 1–25. [CrossRef] 53. Yan, H.; Dechant, C.; Hamid, M. Improving Soil Moisture Proﬁle Prediction with the Particle Filter-Markov Chain Monte Carlo Method. IEEE Trans. Geosci. Remote Sens. 2015, 53, 6134–6147. [CrossRef] 54. Huang, Y.; Jiang, H.; Wang, W.F.; Wang, W.; Sun, D. Soil moisture content prediction model for tea plantations based on SVM optimised by the bald eagle search algorithm. Cogn. Comput. Syst. 2021, 3, 351–360. [CrossRef] 55. Wang, X.; Lv, J.; Wang, C.; Xie, D. Soil moisture content prediction using wavelet transform and support vector machine with genetic algorithm optimization. ICIC Express Lett. Part B Appl. 2014, 5, 1141–1148. Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Agriculture Multidisciplinary Digital Publishing Institute http://www.deepdyve.com/lp/multidisciplinary-digital-publishing-institute/research-on-provincial-level-soil-moisture-prediction-based-on-extreme-RbPOkF4V8q

Loading next page...

References (50)

Qingliang Li, Yuheng Zhu, Shangguan Wei, Xuezhi Wang, Lu Li, Fanhua Yu (2022)
An attention-aware LSTM model for soil moisture and soil temperature prediction
Geoderma
(2017)
Prediction of soil moisture in multiple depth based on time delay neural network
(2018)
Research of adaptive genetic neural network algorithm in soil moisture prediction
B. Dan, Xiaogu Zheng, Guocan Wu, Tao Li (2020)
Assimilating shallow soil moisture observations into land models with a water budget constraint
Hydrology and Earth System Sciences
R. Eisenman (1967)
A profit-sharing interpretation of Shapley value for N-person games.
Behavioral science, 12 5
Jingxin Yu, Song Tang, Lili Zhangzhong, Wengang Zheng, Long Wang, A. Wong, Linlin Xu (2020)
A Deep Learning Approach for Multi-Depth Soil Water Content Prediction in Summer Maize Growth Period
IEEE Access, 8
H. Kunstmann, G. Jung, S. Wagner, H. Clottey (2008)
Integration of atmospheric sciences and hydrology for the development of decision support systems in sustainable water management
Physics and Chemistry of The Earth, 33
G. Biau (2010)
Analysis of a Random Forests Model
J. Mach. Learn. Res., 13
R. Prasad, R. Deo, Yan Li, T. Maraseni (2019)
Weekly soil moisture forecasting with multivariate sequential, ensemble empirical mode decomposition and Boruta-random forest hybridizer algorithm approach
CATENA
Oliviu Matei, T. Rusu, A. Petrovan, G. Mihuţ (2017)
A Data Mining System for Real Time Soil Moisture Prediction
Procedia Engineering, 181
Wei Hu, B. Si (2013)
Soil water prediction based on its scale-specific control using multivariate empirical mode decomposition
Geoderma, 193194
Xuan Wang, Jiake Lv, Chenqi Wang (2014)
SOIL MOISTURE CONTENT PREDICTION USING WAVELET TRANSFORM AND SUPPORT VECTOR MACHINE WITH GENETIC ALGORITHM OPTIMIZATION
ICIC Express Letters, Part B: Applications An International Journal of Research and Surveys, 5
R. Mahmood, K. Hubbard (2004)
An Analysis of Simulated Long-Term Soil Moisture Data for Three Land Uses under Contrasting Hydroclimatic Conditions in the Northern Great Plains
Journal of Hydrometeorology, 5
M. Ferreira, C. Valancogne (1997)
Experimental Study of a Stress Coefficient: Application on a Simple Model for Irrigation Scheduling and Daily Evapotranspiration Estimation
IFAC Proceedings Volumes, 30
(2013)
Forecast and Analysis of Soil Moisture Based on SIMPEL model
(2006)
The analyses of Jiangsu climate variety in forty years
J. Bell, R. Sherry, Yiqi Luo (2010)
Changes in soil water dynamics due to variation in precipitation and temperature: An ecohydrological analysis in a tallgrass prairie
Water Resources Research, 46
V. Cherkassky, Yunqian Ma (2004)
Practical selection of SVM parameters and noise estimation for SVM regression
Neural networks : the official journal of the International Neural Network Society, 17 1
S. Agatonovic-Kustrin, R. Beresford (2000)
Basic concepts of artificial neural network (ANN) modeling and its application in pharmaceutical research.
Journal of pharmaceutical and biomedical analysis, 22 5
A. Kamilaris, F. Prenafeta-Boldú (2018)
Deep learning in agriculture: A survey
ArXiv, abs/1807.11809
Shaohua Wang, Gang Fu (2023)
Modelling soil moisture using climate data and normalized difference vegetation index based on nine algorithms in alpine grasslands
, 11
Ying Huang, Hao Jiang, Wenfeng Wang, Weixing Wang, Daozong Sun (2021)
Soil moisture content prediction model for tea plantations based on SVM optimised by the bald eagle search algorithm
Cogn. Comput. Syst., 3
M. Niazkar (2020)
Assessment of artificial intelligence models for calculating optimum properties of lined channels
Journal of Hydroinformatics
Ron Kohavi (1995)
A Study of Cross-Validation and Bootstrap for Accuracy Estimation and Model Selection
Yu Cai, Wengang Zheng, Xin Zhang, Lili Zhangzhong, X. Xue (2019)
Research on soil moisture prediction model based on deep learning
PLoS ONE, 14
Y. Qi, J. Darilek, Biao Huang, Yongcun Zhao, Weixia Sun, Zhiquan Gu (2009)
Evaluating soil quality indices in an agricultural region of Jiangsu Province, China.
Geoderma, 149
Qingliang Li, Zhongyan Li, W. Shangguan, Xuezhi Wang, Lu Li, Fanhua Yu (2022)
Improving soil moisture prediction using a novel encoder-decoder model with residual learning
Comput. Electron. Agric., 195
J. Robinson, K. Hubbard (1990)
Soil Water Assessment Model for Several Crops in the High Plains
Agronomy Journal, 82
C. Notarnicola, M. Angiulli, F. Posa (2008)
Soil moisture retrieval from remotely sensed data: Neural network approach versus Bayesian method
IEEE Transactions on Geoscience and Remote Sensing, 46
M. Shoaib, A. Shamseldin, B. Melville, M. Khan (2016)
A comparison between wavelet based static and dynamic neural network approaches for runoff prediction
Journal of Hydrology, 535
J. Holland, Asim Biswas (2015)
Predicting the mobile water content of vineyard soils in New South Wales, Australia
Agricultural Water Management, 148
M. Gill, T. Asefa, M. Kemblowski, M. McKee (2006)
SOIL MOISTURE PREDICTION USING SUPPORT VECTOR MACHINES 1
JAWRA Journal of the American Water Resources Association, 42
I. Sandholt, K. Rasmussen, J. Andersen (2002)
A simple interpretation of the surface temperature/vegetation index space for assessment of surface moisture status
Remote Sensing of Environment, 79
J. Bergstra, Yoshua Bengio (2012)
Random Search for Hyper-Parameter Optimization
J. Mach. Learn. Res., 13
Dunwen Liu, Chaoxiang Liu, Yu Tang, Chun Gong (2022)
A GA-BP Neural Network Regression Model for Predicting Soil Moisture in Slope Ecological Protection
Sustainability
Nemanja Filipović, S. Brdar, G. Mimić, Oskar Marko, V. Crnojevic (2022)
Regional soil moisture prediction system based on Long Short-Term Memory network
Biosystems Engineering
Thu Nguyen, H. Ngo, Wenshan Guo, S. Chang, D. Nguyen, Chi Nguyen, Jian Zhang, S. Liang, X. Bui, N. Hoang (2022)
A low-cost approach for soil moisture prediction using multi-sensor data and machine learning algorithm.
The Science of the total environment
J. Xu, Junfang Zhao, Wanting Zhang, X. Xu (2010)
A Novel Soil Moisture Predicting Method Based on Artificial Neural Network and Xinanjiang Model
Advanced Materials Research, 121-122
O. Adeyemi, I. Grove, S. Peets, Yuvraj Domun, Tomas Norton (2018)
Dynamic Neural Network Modelling of Soil Moisture Content for Predictive Irrigation Scheduling
Sensors (Basel, Switzerland), 18
Tianqi Chen, Carlos Guestrin (2016)
XGBoost: A Scalable Tree Boosting System
Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
R. Ragab (1995)
Towards a continuous operational system to estimate the root-zone soil moisture from intermittent remotely sensed surface moisture
Journal of Hydrology, 173
Wei Wei, Jing Zhang, Liang Zhou, B. Xie, Jun-ju Zhou, Chuanhua Li (2021)
Comparative evaluation of drought indices for monitoring drought based on remote sensing data
Environmental Science and Pollution Research, 28
(2016)
Drought stress in maize causes differential acclimation responses of glutathione and sulfur metabolism in leaves and roots
BMC Plant Biology, 16
Peijun Li, Y. Zha, Liangsheng Shi, C. Tso, Yonggen Zhang, W. Zeng (2020)
Comparison of the use of a physical-based model with data assimilation and machine learning methods for simulating soil water dynamics
Journal of Hydrology, 584
Huihui Feng, Yuanbo Liu (2015)
Combined effects of precipitation and air temperature on soil moisture in different land covers in a humid basin
Journal of Hydrology, 531
Hongxiang Yan, C. DeChant, H. Moradkhani (2015)
Improving Soil Moisture Profile Prediction With the Particle Filter-Markov Chain Monte Carlo Method
IEEE Transactions on Geoscience and Remote Sensing, 53
Jinjing Pan, W. Shangguan, Lu Li, Hua Yuan, Shupeng Zhang, X. Lu, Nan Wei, Yongjiu Dai (2019)
Using data‐driven methods to explore the predictability of surface soil moisture with FLUXNET site data
Hydrological Processes, 33
Yongjiu Dai, X. Zeng, R. Dickinson, I. Baker, G. Bonan, M. Bosilovich, A. Denning, P. Dirmeyer, P. Houser, G. Niu, K. Oleson, C. Schlosser, Zong‐Liang Yang (2003)
The Common Land Model
Bulletin of the American Meteorological Society, 84
S. Gumiere, M. Camporese, A. Botto, J. Lafond, C. Paniconi, J. Gallichand, A. Rousseau (2020)
Machine Learning vs. Physics-Based Modeling for Real-Time Irrigation Management
, 2
(2018)
Design and Verification of Quality Control Methods for Automatic Soil Moisture Observation Data in China

Publisher: Multidisciplinary Digital Publishing Institute
Copyright: © 1996-2023 MDPI (Basel, Switzerland) unless otherwise stated Disclaimer Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. Terms and Conditions Privacy Policy
ISSN: 2077-0472
DOI: 10.3390/agriculture13050927
Publisher site: See Article on Publisher Site

Abstract

agriculture Article Research on Provincial-Level Soil Moisture Prediction Based on Extreme Gradient Boosting Model 1 2 3 , Yifang Ren , Fenghua Ling and Yong Wang * Jiangsu Provincial Climate Center, Nanjing 210008, China; renyifang2023@gmail.com Institute for Climate and Application Research (ICAR)/CIC-FEMD/KLME/ILCEC, School of Atmospheric Sciences, Nanjing University of Information Science and Technology, Nanjing 210044, China; 20211101018@nuist.edu.cn School of Applied Meteorology, Nanjing University of Information Science and Technology, Nanjing 210044, China * Correspondence: ywang@nuist.edu.cn Abstract: As one of the physical quantities concerned in agricultural production, soil moisture can effectively guide ﬁeld irrigation and evaluate the distribution of water resources for crop growth in various regions. However, the spatial variability of soil moisture is dramatic, and its time series data are highly noisy, nonlinear, and nonstationary, and thus hard to predict accurately. In this study, taking Jiangsu Province in China as an example, the data of 70 meteorological and soil moisture automatic observation stations from 2014 to 2022 were used to establish prediction models of 0–10 cm soil relative humidity (RH ) via the extreme gradient boosting (XGBoost) algorithm. Before constructing the s10cm model, according to the measured soil physical characteristics, the soil moisture observation data were divided into three categories: sandy soil, loam soil, and clay soil. Based on the impacts of various factors on the soil water budget balance, 14 predictors were chosen for constructing the model, among which atmospheric and soil factors accounted for 10 and 4, respectively. Considering the differences in soil physical characteristics and the lagged effects of environmental impacts, the best inﬂuence times of the predictors for different soil types were determined through correlation analysis to improve the rationality of the model construction. To better evaluate the importance of soil factors, two sets of models (Model and Model ) were designed by taking soil factors as optional predictors _atmo _soil&atmo put into the XGBoost model. Meanwhile, the contributions of predictors to the prediction results Citation: Ren, Y.; Ling, F.; Wang, Y. were analyzed with Shapley additive explanation (SHAP). Six prediction effect indicators, as well as Research on Provincial-Level Soil a typical drought process that happened in 2022, were analyzed to evaluate the prediction accuracy. Moisture Prediction Based on The results show that the time with the highest correlations between environmental predictors and Extreme Gradient Boosting Model. RH varied but was similar between soil types. Among these predictors, the contribution rates of s10cm Agriculture 2023, 13, 927. https:// maximum air temperature (T ), cumulative precipitation (P ), and air relative humidity (RH ) amax sum a doi.org/10.3390/agriculture13050927 in atmospheric factors, which functioned as a critical factor affecting the variation in soil moisture, Academic Editors: Tarendra are relatively high in both models. In addition, adding soil factors could improve the accuracy of soil Lakhankar and Gerard Arbat moisture prediction. To a certain extent, the XGBoost model performed better when compared with artiﬁcial neural networks (ANNs), random forests (RFs), and support vector machines (SVMs). The Received: 28 February 2023 values of the correlation coefﬁcient (R), root mean square error (RMSE), mean absolute error (MAE), Revised: 14 April 2023 Accepted: 17 April 2023 mean absolute relative error (MARE), Nash–Sutcliffe efﬁciency coefﬁcient (NSE), and accuracy (ACC) Published: 24 April 2023 of Model were 0.69, 11.11, 4.87, 0.12, 0.50, and 88%, respectively. This study veriﬁed that _soil&atmo the XGBoost model is applicable to the prediction of soil moisture at the provincial level, as it could reasonably predict the development processes of the typical drought event. Copyright: © 2023 by the authors. Keywords: soil moisture; prediction; XGBoost algorithm; SHAP Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons 1. Introduction Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ Soil moisture is a critical climate variable that regulates climate change by facilitating 4.0/). the exchange and distribution of water and energy in land–air interaction. Additionally, Agriculture 2023, 13, 927. https://doi.org/10.3390/agriculture13050927 https://www.mdpi.com/journal/agriculture Agriculture 2023, 13, 927 2 of 17 soil moisture plays a signiﬁcant role in agricultural production, as deﬁcits or overﬂows of soil moisture during critical periods can impact crop growth and yields [1]. Integrating information on available soil moisture and crop water demands can help the development of timely and appropriate irrigation schedules [2], which is particularly important in areas with poor water conditions. The variations and differences in soil moisture across regions are determined by its budget balance, which is inﬂuenced by several factors. Soil moisture is sourced from atmospheric precipitation and artiﬁcial irrigation, and its expenditure depends on physical processes such as evapotranspiration and runoff, which are inﬂuenced by local weather conditions, soil characteristics, land cover, and other factors [3]. Usually, soil moisture can be expressed using physical variables such as relative humidity, weight water content, and volume water content. Among these variables, relative humidity, calculated as the percentage of soil water content and ﬁeld capacity, can comprehensively reﬂect the soil moisture status and surface hydrological processes [4,5]. Consequently, soil relative humid- ity is an essential reference in irrigation, enabling an analysis of soil moisture differences between regions. Soil moisture prediction based on relative humidity can enhance the defense against waterlogging and drought in farmland. Numerous studies have investigated soil moisture prediction using various methods. Traditional approaches include the water balance method [6–8], statistical empirical for- mula method [9], time series method [10,11], and physical models based on hydrological processes [12]. These methods typically consider the soil water budget balance principle, relationships between soil water and environmental factors, change characteristics of soil water over time, and land–air interaction. They use model building or time series analysis to forecast soil moisture. With advances in information technology, various applications of machine learning (ML) in agricultural production have been widely developed, including predictions of the crop growth period, yield, and soil moisture [13–16]. ML technologies such as artiﬁcial neural networks (ANNs) [17], support vector machines (SVMs) [18], and gradient boosting regression trees (GBRTs) [19] offer a novel perspective for soil moisture prediction due to their advantages of having a low computational cost, strong self-learning ability, high prediction accuracy, and wide suitability [20–22]. For instance, a GA-BP neural network regression model was tested to perform well in predicting the soil moisture of high side slopes [23]. A proposed novel encoder–decoder model with residual learning played an excellent role in solving the nonlinear problem of soil moisture prediction, which was tested using data from 13 FLUXNET sites with varying plant function types and climatic characteristics [24]. In the research of soil moisture prediction based on machine learning, besides ﬁnding suitable prediction models [25], selecting the appropriate input factors for the prediction model is crucial. Many studies have selected meteorological factors directly related to soil moisture, such as precipitation, transpiration, sunshine, and surface temperature [26]. For instance, Xu et al. (2010) developed and tested an integrated soil moisture prediction model based on artiﬁcial neural networks (ANNs) with meteorological data in the semi-arid region of eastern China, and the model performed well at basin scales [27]. Li et al. (2018) applied the adaptive genetic ANN method to improve the quality of soil moisture prediction using atmospheric forcing data, which include air temperature, relative humidity, wind speed, radiation, and precipitation, as well as soil forcing data, such as soil temperature at 5 cm depth and lagged soil moisture at 0–10 cm [28]. Moreover, with the advancement of remote sensing technology, remote sensing monitoring indexes based on multi-source data, including optical, thermal infrared, microwave, and other data, have also been widely used for soil moisture monitoring and prediction [29–31]. However, current research on soil moisture prediction has some limitations, including discontinuity in remote sensing images, an inadequate use of data from automatic observa- tion stations, and unclear inﬂuencing factors of soil moisture [24,32]. Therefore, this study utilized the soil moisture data and corresponding meteorological data from 70 automatic stations in Jiangsu Province, determined the optimal inﬂuence times of the input factors Agriculture 2023, 13, x FOR PEER REVIEW 3 of 18 observation stations, and unclear inﬂuencing factors of soil moisture [24,32]. Therefore, Agriculture 2023, 13, 927 3 of 17 this study utilized the soil moisture data and corresponding meteorological data from 70 automatic stations in Jiangsu Province, determined the optimal inﬂuence times of the in- put factors for prediction models using a correlation analysis method, and applied ex- treme gradient boosting (XGBoost) to establish two sets of soil relative humidity predic- for prediction models using a correlation analysis method, and applied extreme gradient tion models (i.e., Model_soil&atmo and Model_atmo). To bett er interpret the inﬂuences of the boosting (XGBoost) to establish two sets of soil relative humidity prediction models (i.e., input factors on these two models and evaluate their performance, Shapley additive ex- Model and Model ). To better interpret the inﬂuences of the input factors on _atmo _soil&atmo planation (SHAP) was applied, and six metrics were utilized as the predicting eﬀect indi- these two models and evaluate their performance, Shapley additive explanation (SHAP) cators to compare the models’ (e.g., ANN, RF, and SVM) prediction accuracy. Further- was applied, and six metrics were utilized as the predicting effect indicators to compare more, a typical drought development process in August 2022 in Jiangsu Province was an- the models’ (e.g., ANN, RF, and SVM) prediction accuracy. Furthermore, a typical drought alyzed in depth. This study aimed to establish a provincial-level and understandable soil development process in August 2022 in Jiangsu Province was analyzed in depth. This m study oistuaimed re pred to ict establish ion mode al b pr yovincial-level applying a mand achine understandable learning algorit soil hm, moistur which co e pruld ediction pro- vid model e a ca by se stu applying dy for other a machine region learning s. algorithm, which could provide a case study for other regions. 2. Materials and Methods 2. Materials and Methods 2.1. Study Area 2.1. Study Area Jiangsu Province (see Figure 1) is located on the east coast of China, in the mid-lati- Jiangsu Province (see Figure 1) is located on the east coast of China, in the mid-latitude tude zone, with a geographical location between 30°46’–35°07’ N and 116°22’–121°55’ E. 0 0 0 0 zone, with a geographical location between 30 46 –35 07 N and 116 22 –121 55 E. It It lies in the climate transition zone between the subtropical and warm temperate zones lies in the climate transition zone between the subtropical and warm temperate zones and belongs to the East Asian monsoon climate zone. The average annual temperature, and belongs to the East Asian monsoon climate zone. The average annual temperature, precipitation, and sunshine hours in Jiangsu Province are between 13.6–16.1 °C, 704–1250 precipitation, and sunshine hours in Jiangsu Province are between 13.6–16.1 C, 704–1250 mm, and 1816–2503 h, respectively [33]. The terrain is generally ﬂat, with the Taihu Plain, mm, and 1816–2503 h, respectively [33]. The terrain is generally ﬂat, with the Taihu Plain, Yanjiang, and Lixia River areas being low-lying and having dense water networks. The Yanjiang, and Lixia River areas being low-lying and having dense water networks. The low mountains and hills account for only 14.33% and are mainly distributed in the west low mountains and hills account for only 14.33% and are mainly distributed in the west and north regions. There are various soil types in Jiangsu, including zonal soils such as and north regions. There are various soil types in Jiangsu, including zonal soils such as cinnamon, brown soil, yellow-brown soil, and yellow soil, and non-zonal soils such as cinnamon, brown soil, yellow-brown soil, and yellow soil, and non-zonal soils such as saline soil, meadow soil, and marsh soil. With a long history of agriculture, natural soil in saline soil, meadow soil, and marsh soil. With a long history of agriculture, natural soil in Jiangsu has evolved into various types of farming soil with diﬀerent soil textures under Jiangsu has evolved into various types of farming soil with different soil textures under the the inﬂuence of diﬀerent farming systems and utilization methods [34]. inﬂuence of different farming systems and utilization methods [34]. Fi Figure gure 1. 1. Ov Overview erview o of f th the e s study tudy area area of of J Jiangsu iangsu Prov Province, ince, China, a China, and nd its geographic its geographical al d distribution istribution map of soil moisture observation stations. map of soil moisture observation stations. 2.2. Data Source 2.2. Data Source Automatic moisture observation instruments have been gradually incorporated into Automatic moisture observation instruments have been gradually incorporated into the meteorological operational observation system since 2010, resulting in the availability the meteorological operational observation system since 2010, resulting in the availability of high regional density and continuous soil moisture observation data across Chinese of high regional density and continuous soil moisture observation data across Chinese provinces [35]. Consequently, daily 0–10 cm soil relative humidity data, measured by provinces [35]. Consequently, daily 0–10 cm soil relative humidity data, measured by 70 70 automatic soil moisture observation stations in Jiangsu Province from 2014 to 2022, along automatic soil moisture observation stations in Jiangsu Province from 2014 to 2022, along with with meteorological data collected by automatic weather stations and soil temperature data meteorological data collected by automatic weather stations and soil temperature data meas- measured by soil temperature instruments at the corresponding 70 soil moisture station ured by soil temperature instruments at the corresponding 70 soil moisture station locations, locations, were used for predicting 0–10 cm soil relative humidity. These atmospheric and soil observation data were obtained from the Jiangsu Meteorological Information Center. Based on the principle of soil water budget balance and considering the inﬂuence of various factors on the 0–10 cm soil relative humidity (RH , %), the predictive factors s10cm were divided into two categories: atmospheric and soil factors. There are ten atmospheric factors, including the mean air temperature (T , C), minimum air temperature (T , C), a amin Agriculture 2023, 13, 927 4 of 17 maximum air temperature (T , C), air relative humidity (RH , %), precipitation (P, mm), amax a sunshine hours (S, h), wind speed (W, ms ), atmospheric pressure (P , hPa), water vapor pressure (e, hPa), and potential evapotranspiration (ET , mm). Additionally, there are four soil factors, including the mean surface temperature (T , C), maximum soil surface temperature (T , C), minimum soil surface temperature (T , C), and 0–10 cm soil smax smin temperature (T , C). s10cm 2.3. Data Classiﬁcation Soil textures and hydrological constants varied signiﬁcantly in Jiangsu Province. Even when weather conditions are identical, different regions may exhibit distinct soil water dynamics due to the differences in soil physical properties [36]. Therefore, it is necessary to consider regional soil characteristics and hydrological constants when predicting soil moisture. To this end, according to the soil hydrological and physical characteristics measured by 70 automatic soil moisture observation stations in Jiangsu Province, the soil moisture observation data were classiﬁed into three categories: sandy soil, loam soil, and clay soil. The statistics of physical parameters corresponding to the different soil types are shown in Table 1. Table 1. Classiﬁcation results and corresponding soil physical characteristics of soil moisture obser- vation data. Soil Bulk Density Field Water Capacity Withering Humidity Soil Type Samples (gcm ) (%) (%) Sand 1.43 25.46 4.04 40,880 Loam 1.40 26.50 5.29 75,920 Clay 1.36 26.62 5.72 87,600 2.4. Methodology Description 2.4.1. Selection of Predictive Factors Soil relative humidity changes are mainly affected by previous and current weather conditions and the state of the soil itself. By distinguishing different soil types, we correlated RH with the averaged or accumulated value (including precipitation and sunshine s10cm hours) of 14 predictor factors on the same day as the soil moisture observed, and 1–10 days in the previous period, to determine the maximum impact time of each predictor (see Table 2). We used the time with the largest correlation coefﬁcient of each predictor as its maximum impact time on RH . The corresponding sample numbers for each soil type s10cm used to take correlation analysis are shown in Table 1. Table 2. List of predictor factors of 0–10 days prior, which are used for correlation analysis with RH . s10cm Names Units Descriptions Range Sunshine hours h Accumulated sunshine hours 0–128.6 Precipitation mm Cumulative precipitation 0–595.4 Evapotranspiration mm Averaged potential evapotranspiration 0.1–10.2 Wind speed ms Averaged wind speed 0–15.9 Relative humidity % Averaged mean air relative humidity 19–100 Averaged water vapor pressure 0.6–42.0 Pressure hPa Averaged atmospheric pressure 983.5–1042.4 Agriculture 2023, 13, 927 5 of 17 Table 2. Cont. Names Units Descriptions Range Averaged mean air temperature 11.1–36.0 Averaged minimum air temperature 15.6–31.9 Averaged maximum air temperature 7.2–40.9 Temperature Averaged mean soil surface temperature 7.0–45.8 Averaged minimum soil surface temperature 14.7–31.2 Averaged maximum soil surface temperature 0.9–70.2 Averaged 0–10 cm mean soil temperature 2.7–39.0 2.4.2. XGBoost Model The XGBoost is an ensemble learning method based on boosting [37]. The boosting technique combines multiple decision trees and aggregates their predictions to obtain a ﬁnal prediction that is more accurate than any individual tree. XGBoost is designed to prevent over-ﬁtting. The XGBoost model builds multiple trees sequentially, with each subsequent tree intended to reduce the errors of the previous tree. As the training proceeds iteratively, new trees are added to predict the error of the prior tree. Such a ﬁtting process is repeated several times until a stopping criterion is met, such as when the root mean square error (RMSE) reaches an asymptotic value. The ultimate prediction of the model is the sum of the predictions from all of the trees. The formula for the prediction at the step t and site location i can be deﬁned as follows [37]: (t1) y ˆ = f (x ) = y ˆ + f (x ) (1) å k i t i k=1 i (t1) where f (x ) is the tree model at step t, y ˆ and y ˆ are the predictions at steps t and t i i i t 1, and x are the predictor variables. The parameters of the model f (x ) are selected i i by optimizing the objective function, and the objective function is deﬁned by root mean square error. Additionally, XGBoost offers several other advanced features [37] that can further enhance the model’s performance. For instance, early stopping allows the training process to be stopped early if the performance on a validation set stops improving. This advanced feature prevents the model from overﬁtting to the training data and can improve its ability to generalize to new data. Cross-validation is another useful technique that can estimate the model’s generalization performance and help to select the optimal hyperparameters. By incorporating these and other advanced features, XGBoost has emerged as one of the most popular and inﬂuential machine learning models. The ﬂow chart depicting the XGBoost model is presented in Figure 2. 2.4.3. The Key Parameters of XGBoost Model In this study, we focused on optimizing several crucial parameters of the XGBoost algorithm, including the number of boost rounds, maximum depth, minimum weight in a child, and learning rate. The number of boost rounds determines the maximum number of boosting iterations, while the maximum depth sets the maximum depth of an individual tree. The minimum weight in a child parameter is utilized to prevent overﬁtting, and the learning rate parameter controls the model’s shrinkage at every step (i.e., a lower learning rate indicates more steps used to achieve the optimum) (see Figure 2). To optimize these parameters, we applied a tuning technique called grid search [38]. This approach computes the optimal values of hyperparameters by exhaustively searching over a range of possible parameter values. We utilized third-fold cross-validation [39] to evaluate the performance of different parameter combinations. In total, we searched through 1500 combinations of parameter values. Ultimately, our XGBoost model achieved Agriculture 2023, 13, 927 6 of 17 the best performance with the maximum depth, minimum weight needed in a child, and learning rate equal to 15, 10, and 0.02, respectively. In addition, we set the maximum number of boosting rounds to 5000 during training and used the early stop technique to Agriculture 2023, 13, x FOR PEER REVIEW 6 of 18 stop the training. The ﬁnal number of iterations was 4218 when the loss on the validation set no longer decreased. Figure 2. The ﬂowchart of the XGBoost model. Figure 2. The ﬂowchart of the XGBoost model. 2.4.4. Shapley Additive Explanations (SHAPs) 2.4.3. The Key Parameters of XGBoost Model SHAP is a local attribution method that is based on the use of Shapley values. The In this study, we focused on optimizing several crucial parameters of the XGBoost Shapley values originate from the ﬁeld of cooperative game theory and represent each algorithm, including the number of boost rounds, maximum depth, minimum weight in play’s average expected marginal contribution in a cooperative game after all possible a child, and learning rate. The number of boost rounds determines the maximum number combinations of players have been considered. It can be formulated as follows [40]: of boosting iterations, while the maximum depth sets the maximum depth of an individ- jSj!(FjSj 1)! ual tree. The minimum weight in a child parameter is utilized to prevent overﬁtt ing, and f = [ f (S[fig) f (S)] (2) i å x x F! the learning rate para Sm Fentfe igr controls the model’s shrinkage at every step (i.e., a lower learning rate indicates more steps used to achieve the optimum) (see Figure 2). where f is the weighted average of all marginal contributions of the predictor i, F is the total To optimize these parameters, we applied a tuning technique called grid search [38]. number of features, S is the subset of predictors from all predictors except for predictor i, This approach computes the optimal values of hyperparameters by exhaustively search- jSj!(FjSj1)! and is the weighting factor counting the number of permutations of the subset F! ing over a range of possible parameter values. We utilized third-fold cross-validation [39] S. f (S) is the expected output given the predictors subset S. [ f (S[fig) f (S)] is the x x x to evaluate the performance of diﬀerent parameter combinations. In total, we searched difference made by the predictor i. through 1500 combinations of parameter values. Ultimately, our XGBoost model achieved 2.4.5. Model Construction and Application the best performance with the maximum depth, minimum weight needed in a child, and learni This ng ra study te eq aimed ual to to15, develop 10, ana d soil 0.02 moistur , respec e tiv prediction ely. In ad model dition for , w dif e fer set t ent he m soila types ximum num- using relevant atmospheric and soil factors. To achieve this, 14 most related factors were ber of boosting rounds to 5000 during training and used the early stop technique to stop obtained by calculating the correlation. Additionally, to account for the different impacts of the training. The ﬁnal number of iterations was 4218 when the loss on the validation set soil types, the variable St was included in the model, with values of 1, 2, and 3 repre- ﬂag no longer decreased. senting sandy, loam, and clay soils, respectively. To further evaluate the importance of soil factors in predicting 0–10 cm soil relative 2.4.4. Shapley Additive Explanations (SHAPs) humidity, two sets of data used as the model’s independent variables were constructed using S 14 HAP optimal is a pr loc edictors al att ribu (including tion metatmospheric hod that is ba and sed soil on variabl the ues) se o and f Sh 10 apl optimal ey values. The predictors (including atmospheric variables only) from 70 stations in Jiangsu Province Shapley values originate from the ﬁeld of cooperative game theory and represent each between 2014 and 2021. Before prediction, missing values in these two data sets were com- play’s average expected marginal contribution in a cooperative game after all possible pleted with the mean values, and the dataset was normalized. A tri-fold cross-validation combinations of players have been considered. It can be formulated as follows [40]: approach [39] was employed to train, validate, and evaluate the model. The data were SF !( −−S 1)! φ=∪ [(fS {i})− f(S)] (2) ix  x F! SF ⊆ \{}i where φ is the weighted average of all marginal contributions of the predictor i , F is the total number of features, S is the subset of predictors from all predictors except for SF !( −−S 1)! predictor i , and is the weighting factor counting the number of permuta- F! tions of the subset S . is the expected output given the predictors subset S . f () S is the diﬀerence made by the predictor i . [( f Si∪− {}) f(S)] xx Agriculture 2023, 13, 927 7 of 17 randomly divided into three sets: 80% (163,520 samples) as the model training dataset, 10% (20,440 samples) as the model validation dataset for parameter optimization, and the remaining 10% (20,440 samples) as the model prediction evaluating dataset. 2.4.6. Model Prediction Effect Interpretation and Veriﬁcation After building the prediction model, the SHAP method was applied to obtain each predictive factor ’s positive and negative effects separately for both Model and _soil&atmo Model . In addition, six metrics were used on the evaluating dataset to evaluate the _atmo performance of XGBoost and other state-of-the-art predictive models, including correlation coefﬁcient (R), root mean square error (RMSE), mean absolute error (MAE), mean absolute relative error (MARE), Nash–Sutcliffe efﬁciency coefﬁcient (NSE), and accuracy (ACC). These indicators are calculated as follows [41]: (y y )(y ˆ y ˆ ) i i i i i=1 R = (3) n n (y y ) (y ˆ y ˆ ) å å i i i i i=1 i=1 u (y y ˆ ) i i i=1 RMSE = (4) MAE = j(y y ˆ )j (5) å i i i=1 1 (y y ˆ ) i i MARE = (6) n y i=1 (y y ˆ ) i i i=1 NSE = 1 (7) n i i=1 (y ˆ ) i=1 1 (y y ˆ ) i i ACC = 1 100% (8) n y i=1 where y is the observed value, y is the predicted value, n is the number of samples, y is i i the mean of observations, and y ˆ is the mean of the prediction. To further verify the prediction capabilities of Model and Model based _atmo _soil&atmo on XGBoost, we compared these models with three state-of-the-art machine learning models (i.e., ANN [42], RF [43], and SVM [44]) for soil moisture prediction over 70 sites in Jiangsu. The comparison was based on the values of these above metrics and the scatter distributions of predicted and observed soil moisture values. Furthermore, we evaluated the performance of Model and Model during a typical drought in August _atmo _soil&atmo 2022 in Jiangsu Province. The ﬂow chart depicting the establishment, interpretation, and evaluation of the prediction models for soil moisture is presented in Figure 3. Agriculture 2023, 13, x FOR PEER REVIEW 8 of 18 y yˆ where is the observed value, is the predicted value, n is the number of samples, i i y is the mean of observations, and y is the mean of the prediction. i i To further verify the prediction capabilities of Model_soil&atmo and Model_atmo based on XGBoost, we compared these models with three state-of-the-art machine learning models (i.e., ANN [42], RF [43], and SVM [44]) for soil moisture prediction over 70 sites in Jiangsu. The comparison was based on the values of these above metrics and the scatt er distribu- tions of predicted and observed soil moisture values. Furthermore, we evaluated the per- formance of Model_soil&atmo and Model_atmo during a typical drought in August 2022 in Agriculture 2023, 13, 927 8 of 17 Jiangsu Province. The ﬂow chart depicting the establishment, interpretation, and evalua- tion of the prediction models for soil moisture is presented in Figure 3. Figure 3. Flow chart of establishing, interpreting, and evaluating soil moisture models. Figure 3. Flow chart of establishing, interpreting, and evaluating soil moisture models. 3. Results 3. Results 3.1. Correlation Analysis between Soil Moisture and Predictive Factors 3.1. Correlation Analysis between Soil Moisture and Predictive Factors After analyzing the correlations between 0–10 cm soil relative humidity (RH ) and s10cm After analyzing the correlations between 0–10 cm soil relative humidity (RHs10cm) and various predictors for different soil types with different advance days (See Figure 4), it was various predictors for diﬀerent soil types with diﬀerent advance days (See Figure 4), it observed that, among the atmospheric factors, RH had a high positive correlation with s10cm was observed that, among the atmospheric factors, RHs10cm had a high positive correlation the mean air relative humidity (RH ) and cumulative precipitation (P ). The correlation a sum with the mean air relative humidity (RHa) and cumulative precipitation (Psum). The corre- coefﬁcients were between 0.17–0.33 and 0.13–0.26, respectively, and their absolute values gradually lation coeincr ﬃcients w eased with ere betw the leading een 0.time, 17–0.3 peaking 3 and 0.13 8–10 –0. days 26, respect prior. Additionally ively, and their ab , RH solute s10cm had a high negative correlation with the mean water vapor pressure (e) and accumulated values gradually increased with the leading time, peaking 8–10 days prior. Additionally, sunshine hours (S ). The absolute correlation coefﬁcients were between 0.24–0.33 and sum RHs10cm had a high negative correlation with the mean water vapor pressure (e) and accu- 0.15–0.33, respectively. The absolute values also increased with the leading time, reaching mulated sunshine hours (Ssum). The absolute correlation coeﬃcients were between 0.24– the maximum at 8 and 10 days prior, respectively. Among the soil factors, RH had a s10cm 0.33 and 0.15–0.33, respectively. The absolute values also increased with the leading time, high negative correlation with the mean maximum surface temperature (T ), with its Agriculture 2023, 13, x FOR PEER REVIEW smax 9 of 18 reaching the maximum at 8 and 10 days prior, respectively. Among the soil factors, RHs10cm maximum absolute value appearing 4–5 days prior. The correlations between RH and s10cm had a high negative correlation with the mean maximum surface temperature (Tsmax), with other factors were relatively low, but all passed the signiﬁcance test of p = 0.01. its maximum absolute value appearing 4–5 days prior. The correlations between RHs10cm and other factors were relatively low, but all passed the signiﬁcance test of . p =0.01 Figure 4. Correlation coefﬁcients between 0–10cm soil relative humidity and various predictive Figure 4. Correlation coeﬃcients between 0–10cm soil relative humidity and various predictive fac- Commented [M1]: 虚线，M 是否需要解释 factors of different soil types, which are (a) sandy soil, (b) loam, and (c) clay, respectively. tors of diﬀerent soil types, which are (a) sandy soil, (b) loam, and (c) clay, respectively. Overall, the correlations between RH and various predictor factors, as well as s10cm Overall, the correlations between RHs10cm and various predictor factors, as well as their change rules with the days advanced, were relatively consistent among different soil their change rules with the days advanced, were relatively consistent among diﬀerent soil types, with the times taken to reach the maximum value being similar (see Figure 4a–c). types, with the times taken to reach the maximum value being similar (see Figure 4a–c). The variabilities of positive–negative correlation with RH were mainly reﬂected in the s10cm factors of the minimum surface temperature and wind speed. Thus, a ﬁxed optimal impact The variabilities of positive–negative correlation with RHs10cm were mainly reﬂected in the factors of the minimum surface temperature and wind speed. Thus, a ﬁxed optimal impact time was set for each predictor factor as the model input, and its corresponding diﬀerences in the impact times between diﬀerent soil types were no longer distinguished. 3.2. Interpretability of Model We analyzed the relationships between the predictor variables and the soil moisture using the XGBoost model and presented the results through SHAP summary plots for each variable. In Figure 3, for each predictor variable displayed on the y-axis, each colored point represents a value of this variable in the dataset and the SHAP values displayed on the x-axis denoting the contributions of that predictor variable, which can be a positive or negative eﬀect on the prediction of soil moisture. The gradient color of each point indi- cates the value of the predictor variable, ranging from low (blue) to high (red), providing a visual representation of the relationships between the predictors and soil moisture. From the SHAP summary chart of Model_soil&atmo in Figure 5a, we observed that Tsmax, Ts10cm, and Tamax had a signiﬁcant negative contribution to the model prediction, consider- ing both atmospheric and soil variables. Conversely, the eﬀects of other factors on the prediction results were either opposite or insigniﬁcant. Among them, Psum had the most considerable positive contribution to the model prediction, followed by RHa. According to the importance of each predictor, the order of the top ﬁve predictors was Tsmax > Psum > Ts10cm > RHa > Ts. Figure 5. SHAP summary chart of (a) Model_soil&atmo and (b) Model_atmo. From the SHAP summary chart of Model_atmo in Figure 5b, we found that the greater value of Tamax, e, and W had a greater negative contribution to the model prediction, con- sidering only atmospheric variables. In contrast, other factors have opposite effects on the Agriculture 2023, 13, x FOR PEER REVIEW 9 of 18 Figure 4. Correlation coeﬃcients between 0–10cm soil relative humidity and various predictive fac- Commented [M1]: 虚线，M 是否需要解释 tors of diﬀerent soil types, which are (a) sandy soil, (b) loam, and (c) clay, respectively. Overall, the correlations between RHs10cm and various predictor factors, as well as their change rules with the days advanced, were relatively consistent among diﬀerent soil types, with the times taken to reach the maximum value being similar (see Figure 4a–c). The variabilities of positive–negative correlation with RHs10cm were mainly reﬂected in the factors of the minimum surface temperature and wind speed. Thus, a ﬁxed optimal impact time was set for each predictor factor as the model input, and its corresponding diﬀerences Agriculture 2023, 13, 927 9 of 17 in the impact times between diﬀerent soil types were no longer distinguished. 3.2. timeIwas nterpret set for abieach lity of Mode predictor l factor as the model input, and its corresponding differences in the impact times between different soil types were no longer distinguished. We analyzed the relationships between the predictor variables and the soil moisture using the XGBoost model and presented the results through SHAP summary plots for 3.2. Interpretability of Model each variable. In Figure 3, for each predictor variable displayed on the y-axis, each colored We analyzed the relationships between the predictor variables and the soil moisture point represents a value of this variable in the dataset and the SHAP values displayed on using the XGBoost model and presented the results through SHAP summary plots for th each e x- variable. axis deno In ti Figur ng t eh3 e , for con each tribu pr ti edictor ons ofvariable that pre displayed dictor va on ria the bley , w -axis, hiceach h can color be ed a positive or point represents a value of this variable in the dataset and the SHAP values displayed on negative eﬀect on the prediction of soil moisture. The gradient color of each point indi- the x-axis denoting the contributions of that predictor variable, which can be a positive or cates the value of the predictor variable, ranging from low (blue) to high (red), providing negative effect on the prediction of soil moisture. The gradient color of each point indicates a visual representation of the relationships between the predictors and soil moisture. the value of the predictor variable, ranging from low (blue) to high (red), providing a visual From the SHAP summary chart of Model_soil&atmo in Figure 5a, we observed that Tsmax, representation of the relationships between the predictors and soil moisture. Ts10cm, and Tamax had a signiﬁcant negative contribution to the model prediction, consider- From the SHAP summary chart of Model in Figure 5a, we observed that _soil&atmo ing both atmospheric and soil variables. Conversely, the eﬀects of other factors on the T , T , and T had a signiﬁcant negative contribution to the model prediction, smax amax s10cm considering both atmospheric and soil variables. Conversely, the effects of other factors prediction results were either opposite or insigniﬁcant. Among them, Psum had the most on the prediction results were either opposite or insigniﬁcant. Among them, P had sum considerable positive contribution to the model prediction, followed by RHa. According the most considerable positive contribution to the model prediction, followed by RH . to the importance of each predictor, the order of the top ﬁve predictors was Tsmax > Psum > According to the importance of each predictor, the order of the top ﬁve predictors was Ts10cm > RHa > Ts. T > P > T > RH > T . smax sum a s s10cm Figure 5. SHAP summary chart of (a) Model_ and (b) Model . soil&atmo _atmo Figure 5. SHAP summary chart of (a) Model_soil&atmo and (b) Model_atmo. From the SHAP summary chart of Model in Figure 5b, we found that the greater _atmo value of T , e, and W had a greater negative contribution to the model prediction, con- From the SHAP summary chart of Model_atmo in Figure 5b, we found that the greater amax sidering only atmospheric variables. In contrast, other factors have opposite effects on the value of Tamax, e, and W had a greater negative contribution to the model prediction, con- prediction results, or their positive–negative characteristics were insignificant. Among them, sidering only atmospheric variables. In contrast, other factors have opposite effects on the P had the most significant positive contribution to the model prediction, followed by RH , sum a which was consistent with the results of Model . According to the importance of each _soil&atm predictor, the order of the top five predictors was P > T > RH > e > W. sum amax a 3.3. Model Prediction Evaluation 3.3.1. Analysis of Model Prediction Accuracy To further verify the prediction capabilities of Model and Model based _soil&atmo _atmo on XGBoost, we compared them with three other state-of-the-art machine learning models (i.e., ANN, RF, and SVM) based on the scatter distributions of the predicted and observed values of soil moisture, and the values of six metrics (i.e., R, RMSE, MAE, MARE, MSE, and ACC). The scatter distributions of the model predictions based on XGBoost and the ac- tual observations of the 0–10 cm soil relative humidity are presented in Figure 6a1,a2. Model and Model showed an even distribution of predicted and observed _soil&atmo _atmo values around the 1:1 diagonal, with Model exhibiting a slightly more clustered _soil&atmo distribution. The mean and standard deviation of Model ’s predictions (79.28% and _soil&atmo Agriculture 2023, 13, x FOR PEER REVIEW 10 of 18 prediction results, or their positive–negative characteristics were insignificant. Among them, Psum had the most significant positive contribution to the model prediction, followed by RHa, which was consistent with the results of Model_soil&atm. According to the importance of each predictor, the order of the top five predictors was Psum > Tamax > RHa > e > W. 3.3. Model Prediction Evaluation 3.3.1. Analysis of Model Prediction Accuracy To further verify the prediction capabilities of Model_soil&atmo and Model_atmo based on XGBoost, we compared them with three other state-of-the-art machine learning models (i.e., ANN, RF, and SVM) based on the scatter distributions of the predicted and observed values of soil moisture, and the values of six metrics (i.e., R, RMSE, MAE, MARE, MSE, and ACC). The scatt er distributions of the model predictions based on XGBoost and the actual observations of the 0–10 cm soil relative humidity are presented in Figure 6a1,a2. Agriculture 2023, 13, 927 10 of 17 Model_soil&atmo and Model_atmo showed an even distribution of predicted and observed val- ues around the 1:1 diagonal, with Model_soil&atmo exhibiting a slightly more clustered dis- tribution. The mean and standard deviation of Model_soil&atmo ’s predictions (79.28% and 10.32%, respectively) were similar to those of the observations (79.30% and 15.77%, re- 10.32%, respectively) were similar to those of the observations (79.30% and 15.77%, respec- spectively). Model_atmo ’s prediction results were comparable to those of Model_soil&atmo, tively). Model ’s prediction results were comparable to those of Model , with _atmo _soil&atmo with only minor diﬀerences. However, overall, the prediction performance of only minor differences. However, overall, the prediction performance of Model _soil&atmo Model_soil&atmo was slightly bett er than that of Model_atmo. was slightly better than that of Model . _atmo Figure 6. Scatter plot of soil moisture observations and predictions of Model and Model _atmo Figure 6. Scatt er plot of soil moisture observations and predictions of Model_soil&_soil&atmo atmo and Model_atmo based on (a1,a2) XGBoost, (b1,b2) ANN, (c1,c2) RF, and (d1,d2) SVM. (The 1:1 diagonal is shown by based on (a1,a2) XGBoost, (b1,b2) ANN, (c1,c2) RF, and (d1,d2) SVM. (The 1:1 diagonal is shown by the gray dashed line, the regression line is shown by the red solid line, and the observed and predicted means and standard deviations are shown by the red dots and dashed boxes, respectively). After comparing the scatter distributions of observations with model predictions based on XGBoost, ANN, RF, and SVM (see Figure 6), it was observed that the lines between the predicted and observed soil moisture for XGBoost were much closer to the ideal line (y = x) than those for the other predictive models. Additionally, the prediction results of the other models presented a relatively smaller standard deviation. Table 3 shows the comprehensive predictive performances of XGBoost, ANN, RF, and SVM over 70 sites in Jiangsu Province. The values of R, RMSE, MAE, MARE, NSE, and ACC for Model and Model based on XGBoost were 0.69, 11.11, 4.87, 0.12, _soil&atmo _atmo 0.50, and 88%, as well as 0.66, 11.49, 4.96, 0.14, 0.47, and 86%, respectively. Comparing the values of the six evaluated indexes of other LM models, it was found that models based on XGBoost always had the lowest RMSE, MAE, and MARE, as well as the highest R, NSE, and ACC. In addition, for XGBoost, compared with Model having an average prediction _atmo accuracy of 86%, Model had better precision, with an average accuracy of 88%. _soil&atmo Notably, Model ’s prediction effects were always slightly better than those of _soil&atmo Model , which was also evident from the prediction results of other models, whether _atmo from the scatter charts or metrics. Agriculture 2023, 13, 927 11 of 17 Table 3. Comparison of XGBoost, ANN, RF, and SVM performances in soil moisture prediction using two data sets as the model’s input. ML Models R RMSE MAE MARE NSE ACC (%) Model 0.69 11.11 4.87 0.12 0.50 88% _soil&atmo XGBoost Model 0.66 11.49 4.96 0.14 0.47 86% _atmo Model 0.59 12.85 6.55 0.16 0.27 84% _soil&atmo ANN Model 0.56 13.19 6.71 0.17 0.23 83% _atmo Model 0.64 12.08 6.07 0.15 0.36 85% _soil&atmo RF Model 0.63 12.25 6.19 0.16 0.34 84% _atmo Model 0.54 13.68 7.56 0.17 0.19 83% _soil&atmo SVM Model 0.51 13.58 6.86 0.18 0.18 82% _atmo Furthermore, the spatial distribution map of the model evaluation indexes (i.e., R and MAE) showed that both Model and Model based on XGBoost had a _atmo _soil&atmo high accuracy in soil moisture prediction, and their spatial distribution patterns were very similar, with differences only at individual stations (see Figure 7). Stations with relatively small correlation coefﬁcients and large average absolute errors of predictions Agriculture 2023, 13, x FOR PEER REVIEW 12 of 18 and observations of both models were mainly concentrated along the northern area of the Yangtze River and in the northeastern area of Jiangsu Province. Figure Figure 7. 7. Spatial Spatial distribution distribution of of predic prediction tion ac accuracy curacy ev evaluation aluation indicators indicators ofof (a ()aModel ) Model_soil&atmo and and _soil&atmo (b) Model_atmo. (b) Model . _atmo In addition, we found that the prediction accuracy of both models varied greatly 3.3.2. Analysis of Typical Drought Process between sites from the spatial distribution maps. According to the statistical analysis, for During 2–23 August 2022, a third round of persistent high temperature occurred in Model , the R between the predicted and measured values ranged from 0.34 to _soil&atmo Jiangsu Province, with the ﬁrst two rounds taking place on 16–22 June and 8–15 July, re- 0.87, with a mean value of 0.69, and the MAE ranged from 0.12% to 14.52%, with a mean spectively. The south of Huaihe region experienced 14–19 days of a maximum tempera- value of 4.87%. The number of sites with R > 0.60 reached 58, accounting for more than ture ≥ 37 °C, with the average temperature between 32–33.7 °C. Compared to the same 82%, and the number of sites with MAE < 5% reached 40, accounting for more than 57%. period in a normal year, the temperature in 2022 was approximately 4 °C higher and the For Model , the R between the predicted and measured values ranged from 0.34 to 0.85, _atmo precipitation was less than 90%. In particular, southern Jiangsu faced widespread high with an average value of 0.66, and the MAE ranged from 0.05% to 13.96%, with an average temperatures above 40 °C from 12–15 August, resulting in a rapid expansion of drought value of 5.04%. The number of sites with R > 0.60 reached 53, accounting for more than across the province. By 15 August, most of the southern Huaihe Basin experienced mod- 75%, and the number of sites with MAE < 5% reached 38, accounting for more than 50%. erate or above meteorological drought, with some areas suﬀering from severe drought. However, the high temperature gradually receded from 24 August, and the precipitation 3.3.2. Analysis of Typical Drought Process gradually increased, mainly in the Huaibei and Sunan areas. As a result, the moisture During 2–23 August 2022, a third round of persistent high temperature occurred conditions across the province improved eﬀectively, and the moisture content reached an in Jiangsu Province, with the ﬁrst two rounds taking place on 16–22 June and 8–15 July, appropriate level. respectively. The south of Huaihe region experienced 14–19 days of a maximum tempera- According to the distribution of a 0–10 cm soil relative humidity on 1, 15, and 30 ture 37 C, with the average temperature between 32–33.7 C. Compared to the same August, which was interpolated from the measurement of the automatic soil moisture sta- period in a normal year, the temperature in 2022 was approximately 4 C higher and the tion (see Figure 8a1–a3), we found on 1 August, aﬀected by antecedent precipitation, the precipitation was less than 90%. In particular, southern Jiangsu faced widespread high soil moisture in most areas of northern Jiangsu was saturated, and the ﬁeld humidity was relatively high, while the 0–10 cm soil relative humidity in some areas of southern Jiangsu was less than 60%. By August 15th, there was a severe soil water shortage in most of the southern Huaihe Basin. The 0–10 cm soil relative humidity was only 40% to 50%, which had reached moderate drought, and was even less than 40% in some regions, reaching severe drought. Aﬀected by precipitation, by 30 August, the ﬁeld soil humidity in some areas of Huaibei was relatively high, and the 0–10 cm soil relative humidity in most south- ern Huaihe Basin had generally improved to more than 60%, with only sporadic areas still suﬀering from the drought. Thus, it can be seen that the variation in farmland drought perfectly corresponds with the beginning, aggravation, and extinction of the entire high- temperature process. Agriculture 2023, 13, 927 12 of 17 temperatures above 40 C from 12–15 August, resulting in a rapid expansion of drought across the province. By 15 August, most of the southern Huaihe Basin experienced mod- erate or above meteorological drought, with some areas suffering from severe drought. However, the high temperature gradually receded from 24 August, and the precipitation gradually increased, mainly in the Huaibei and Sunan areas. As a result, the moisture conditions across the province improved effectively, and the moisture content reached an appropriate level. According to the distribution of a 0–10 cm soil relative humidity on 1, 15, and 30 Au- gust, which was interpolated from the measurement of the automatic soil moisture station (see Figure 8a1–a3), we found on 1 August, affected by antecedent precipitation, the soil moisture in most areas of northern Jiangsu was saturated, and the ﬁeld humidity was relatively high, while the 0–10 cm soil relative humidity in some areas of southern Jiangsu was less than 60%. By 15th August, there was a severe soil water shortage in most of the southern Huaihe Basin. The 0–10 cm soil relative humidity was only 40% to 50%, which had reached moderate drought, and was even less than 40% in some regions, reach- ing severe drought. Affected by precipitation, by 30 August, the ﬁeld soil humidity in some areas of Huaibei was relatively high, and the 0–10 cm soil relative humidity in most southern Huaihe Basin had generally improved to more than 60%, with only sporadic Agriculture 2023, 13, x FOR PEER REVIE areas W still suffering from the drought. Thus, it can be seen that the variation in farmland 13 of 18 drought perfectly corresponds with the beginning, aggravation, and extinction of the entire high-temperature process. Figure 8. Relative humidity of 10 cm soil relative humidity of (a1–a3) observations, (b1–b3) Figure 8. Relative humidity of 10 cm soil relative humidity of (a1–a3) observations, (b1–b3) Model_soil&atmo predictions, and (c1–c3) Model_atmo predictions on 1, 15, and 30 August 2022. Model predictions, and (c1–c3) Model predictions on 1, 15, and 30 August 2022. _soil&atmo _atmo The spatial distribution patt erns of the corresponding prediction results of the mod- The spatial distribution patterns of the corresponding prediction results of the models els agreed with the observation results. The prediction results reﬂected not only the de- agreed with the observation results. The prediction results reﬂected not only the devel- velopment process of drought but also the distribution areas of diﬀerent levels of farm- opment process of drought but also the distribution areas of different levels of farmland dr lan ought. d droug However ht. Howe , v the er, the predicted predicte dr d dr ought ough situation t situatiowas n war s elatively relativelyweak weak c compar ompared ed t to o the observation results. Overall, the diﬀerences in the distribution patt ern and numeric the observation results. Overall, the differences in the distribution pattern and numeric value between the predictions and observations of Model_soil&atmo were less than those of Model_atmo (see Figure 8b1–b3 and Figure 8c1–c3, respectively). 4. Discussion Based on the observation, soil types, and meteorological data, this study adopted XGBoost to predict soil moisture variations. Diﬀerent atmospheric and soil factor combi- nations were selected as input variables to establish two sets of prediction models (Model_soil&atmo and Model_atmo) for RHs10cm. At the same time, the contributions of the pre- dictive factors were discussed using SHAP. The prediction accuracy was evaluated by comparing six evaluated indexes with other popular ML methods and analyzing a typical drought process in 2022. The variation in soil moisture is a complex coupling system that exhibits high noise, nonlinearity, and unstable random time series data [22]. Compared to traditional statisti- cal models, machine learning algorithms use multiple processing layers consisting of com- plex structures or multiple nonlinear transformations to highly abstract data, which could overcome the inﬂuence of white noise on the prediction accuracy and eﬀectively improve the simulation accuracy [25]. However, diﬀerent ML methods have diﬀerent applicabili- ties for the same dataset. For example, in a study predicting soil moisture based on three diﬀerent datasets, machine learning techniques such as multiple linear regression (MLR), support vector regression (SVR), and recurrent neural networks (RNNs) were compared, and MLR was found to have a bett er performance than the others. Our study used auto- matic soil moisture observations to compare the prediction accuracies of two models Agriculture 2023, 13, 927 13 of 17 value between the predictions and observations of Model were less than those of _soil&atmo Model (see Figure 8b1–b3 and Figure 8c1–c3, respectively). _atmo 4. Discussion Based on the observation, soil types, and meteorological data, this study adopted XG- Boost to predict soil moisture variations. Different atmospheric and soil factor combinations were selected as input variables to establish two sets of prediction models (Model _soil&atmo and Model ) for RH . At the same time, the contributions of the predictive factors _atmo s10cm were discussed using SHAP. The prediction accuracy was evaluated by comparing six evaluated indexes with other popular ML methods and analyzing a typical drought process in 2022. The variation in soil moisture is a complex coupling system that exhibits high noise, nonlinearity, and unstable random time series data [22]. Compared to traditional statistical models, machine learning algorithms use multiple processing layers consisting of complex structures or multiple nonlinear transformations to highly abstract data, which could overcome the inﬂuence of white noise on the prediction accuracy and effectively improve the simulation accuracy [25]. However, different ML methods have different applicabilities for the same dataset. For example, in a study predicting soil moisture based on three different datasets, machine learning techniques such as multiple linear regression (MLR), support vector regression (SVR), and recurrent neural networks (RNNs) were compared, and MLR was found to have a better performance than the others. Our study used automatic soil moisture observations to compare the prediction accuracies of two models based on XGBoost with ANN, RF, and SVM. It showed that Model based on _soil&atm XGBoost was superior, providing the lowest RMSE (11.11), MAE (4.87), and MARE (0.12), and highest R (0.69), NSE (0.50), and ACC (88%). Due to different research and application purposes, the dataset applied in soil moisture prediction studies based on machine learning algorithms is varied, including in situ sites [45], remote sensing [46], reanalysis [47], and ﬂux stations [24]. These datasets usually belong to diverse regions with different spatial and temporal resolutions, so it is still challenging to make direct comparisons even if the same method is applied. The analysis of a typical drought process showed that the XGBoost model based on site data had a good performance and was a feasible method for soil water content prediction, as it could capture a reasonable spatial distribution of the soil moisture. In addition, several advantages were considered for choosing the data observed from the automatic observation stations. Firstly, for a speciﬁc site, the data of the automatic observation station have lower errors than the data obtained by remote sensing instruments and reanalysis data, where the problems of insufﬁcient time resolution and delayed acquisition also exist [47]. Hence, we can more accurately explore the relationship between soil moisture and environmental parameters. Secondly, soil moisture and its related meteorological or soil data are commonly available with the exact temporal resolution, so abundant data could be provided for training the predictive model. It is important to note that the predictivity of soil moisture depends on the data’s time steps and spatial resolutions due to their different distribution and variation [24,48]. Moreover, the wideness of the application of soil moisture prediction usually depends on its spatial representativeness. Therefore, as more automatic weather stations are installed, the proposed model based on site data could be helpful for the operational studies on soil moisture prediction over larger regions and could provide information for timely and optimal irrigation scheduling. However, considering the spatial variability of soil moisture, in-depth future research is still needed, using situ data, remote sensing, and reanalysis data. The appropriate selection of model input factors could promote the accuracy of the prediction model [49]. In this research, we correlated the RH with 14 predictors s10cm 1–10 days before to determine each predictor ’s maximum impact time. The selected predictors were taken as inputs for the model, which would make the model establishment more reasonable, but still needs to be tested in the future. In addition, the contributions of Agriculture 2023, 13, 927 14 of 17 each predictor on the modeling results of two sets of models were discussed via SHAP. The analysis revealed that soil factors in Model played a positive role in the prediction _soil&atmo of soil moisture. Overall, the prediction accuracy of Model was higher than that of _soil&atmo Model . Therefore, introducing soil factors such as T , T , and T could improve _atmo smax s s10cm the prediction accuracy of soil moisture to some extent. For atmospheric factors, T , amax P , and RH are crucial for improving the soil moisture prediction accuracy. These sum a results are consistent with the view that temperature and precipitation are the main factors affecting the variations in soil moisture by adjusting the water budget balance [50,51]. This study aimed to predict the 0–10cm soil relative humidity, which is a crucial parameter for drought and waterlogging prevention, as well as farmland fertilization and irrigation. Generally, the cultivation layer of crops is 0–20 cm, and the water condition of this layer has a good characterization of crop drought. However, compared with the deep soil layer, the 0–10 cm soil layer is more directly affected by meteorological conditions such as precipitation and temperature. When the temperature is high and the amount of evapotranspiration increases, the lack of moisture in crop ﬁelds appears gradually from top to bottom. The moisture deﬁcit in surface soil is easily detected and can serve as the evaluation index for preventing and controlling crop drought. In addition, there is an excellent linear correlation between the soil relative moisture at different levels of depth [52], and hence the surface soil moisture condition is a good indicator of deep soil moisture conditions. This study deeply integrated the XGBoost with meteorological data to establish a provincial-level soil moisture prediction model, which can provide a reference for soil moisture prediction research in other regions. The model can be used to analyze historical soil water change rules and typical drought and ﬂood cases during the period lacking soil moisture observation while high-density meteorological observation is available (mainly from the 1960s to 2010s). However, there are some deﬁciencies and uncertainties in this study. For instance, only four frequently used machine learning algorithms were used in the study. In the future, multiple machine learning algorithms or other methods [53–55] could be used to conduct soil moisture prediction research to analyze the advantages and disadvantages of different methods and applicable conditions. Based on the XGBoost algorithm, the positive and negative contributions of most factors in the Model and _soil&atmo Model for soil moisture prediction analyzed by SHAP were consistent and conformed _atmo to the actual physical meaning. However, there were some cases where the same factor had the opposite contribution to the prediction results, which needs further investigation. 5. Conclusions Soil moisture is the characterization of farmland drought and ﬂood and the basis for irrigation schemes. The prediction of soil relative humidity was achieved based on the XGBoost model using continuous daily atmospheric and soil observation data from automatic stations. The methods of correlation analysis and SHAP were applied to select model predictors and evaluate the contribution of model factors. In addition, six effect indicators and a typical drought process were analyzed to compare the predictive accuracy of the XGBoost model with the other three machine learning models (i.e., ANN, RF, and SVM) to assess the predictive power of the model. Through correlative analysis, we found that the time with the highest correlations between environmental predictors and RH varied but was similar between soil types. s10cm Among atmospheric factors, the mean RH and P exhibited strong positive correlations a sum with RH , with correlation coefﬁcients ranging from 0.17 to 0.33 and 0.13 to 0.26. The s10cm correlation gradually increased over time, reaching the maximum 8~10 days ago. On the other hand, the mean e and S displayed strong negative correlations with RH , sum s10cm with correlation coefﬁcients ranging from 0.24 to 0.33 and from 0.15 to 0.33. Their absolute values also gradually increased over time, peaking at the time of 8 days ago and 10 days ago, respectively. Among the soil factors, the mean T showed a strong smax negative correlation with RH , and its maximum absolute value appeared 4~5 days s10cm Agriculture 2023, 13, 927 15 of 17 ago. Furthermore, via SHAP analysis, it showed that the contributions and impacts of the predictors on the modeling results for Model and Model were different. _atmo _soil&atmo According to the importance of each predictor, the orders of the top ﬁve predictors of these two models were T > P > T > RH > T and P > T > RH > e > W, smax sum s10cm a s sum amax a respectively. Overall, among the predictors, the contribution rates of T , P , and RH amax sum a in atmospheric factors, which functioned as a critical factor affecting the variation in soil moisture, were relatively high in both models. The overall performances of Model and Model based on XGBoost ex- _soil&atmo _atmo hibited lower error values when compared to ANN, RF, and SVM, thereby verifying the prediction capabilities of the XGBoost model. The values of R, RMSE, MAE, MARE, NSE, and ACC for Model and Model based on XGBoost were 0.69, 11.11, 4.87, 0.12, _atmo _soil&atmo 0.50, and 88%, and 0.66, 11.49, 4.96, 0.14, 0.47, and 86%, respectively. Both Model _soil&atmo and Model using XGBoost outperformed the other machine learning models in the _atmo scatter distribution of the predicted and measured values. In addition, by integrating the re- sults of SHAP analysis and comparisons of Model and Model , it showed that _soil&atmo _atmo Model ’s prediction effects were always slightly better than those of Model . _atmo _soil&atmo Hence, it is worth noting that introducing soil factors (e.g., T , T , and T ) can smax s s10cm positively improve the soil moisture prediction accuracy. Furthermore, the XGBoost model was applicable for provincial-level soil moisture prediction as it captured the spatial distribution characteristics of different levels of drought and effectively predicted the dynamic change process of the “occurrence–development– termination” of a speciﬁc drought event. Therefore, the excellent establishment of a soil moisture prediction model based on automatic observation stations, which effectively overcomes the temporary discontinuity of remote sensing inversion and the problem of a low prediction accuracy, could not only effectively guide farmland irrigation but also validly compensate for the insufﬁcient historical observation of soil moisture stations. Author Contributions: Conceptualization, Y.R. and Y.W.; methodology, F.L.; software, Y.R. and F.L.; validation, Y.R. and F.L.; formal analysis, F.L.; investigation, Y.R. and Y.W.; resources, Y.R.; data curation, Y.R.; writing—original draft preparation, Y.R. and F.L.; writing—review and editing: Y.W.; visualization: Y.R. and F.L.; supervision: Y.W.; project administration: Y.W. All authors have read and agreed to the published version of the manuscript. Funding: This research was funded by the National Natural Science Foundation of China Project (41805049). Institutional Review Board Statement: Not applicable. Data Availability Statement: The model prediction results presented in this study are available upon request from the corresponding author. The original observations are not publicly available due to the privacy policy. Acknowledgments: We thank the editors and reviewers for their comments to improve our manuscript. Conﬂicts of Interest: The authors declare no conﬂict of interest. References 1. Ahmad, N.; Malagoli, M.; Wirtz, M.; Hell, R. Drought stress in maize causes differential acclimation responses of glutathione and sulfur metabolism in leaves and roots. BMC Plant Biol. 2016, 16, 247. [CrossRef] [PubMed] 2. Isabel Ferreira, M.; Valancogne, C. Experimental Study of a Stress Coefﬁcient: Application on a Simple Model for Irrigation Scheduling and Daily Evapotranspiration Estimation. IFAC Proc. Vol. 1997, 30, 33–38. [CrossRef] 3. Dai, Y.; Zeng, X.; Dickinson, R.E.; Baker, I.; Bonan, G.B.; Bosilovich, M.G.; Denning, A.S.; Dirmeyer, P.A.; Houser, P.R.; Niu, G.; et al. The Common Land Model. Bull. Am. Meteorol. Soc. 2003, 84, 1013–1024. [CrossRef] 4. Kunstmann, H.; Jung, G.; Wagner, S.; Clottey, H. Integration of atmospheric sciences and hydrology for the development of decision support systems in sustainable water management. Phys. Chem. Earth Parts A/B/C 2008, 33, 165–174. [CrossRef] 5. Dan, B.; Zheng, X.; Wu, G. Assimilating Shallow Soil Moisture Observations into Land Models with a Water Budget Constraint. Hydrol. Earth Syst.Sci. 2020, 24, 5187–5201. [CrossRef] 6. Robinson, J.M.; Hubbard, K.G. Soil Water Assessment Model for Several Crops in the High Plains. Agron. J. 1990, 82, 1141–1148. [CrossRef] Agriculture 2023, 13, 927 16 of 17 7. Mahmood, R.; Hubbard, K.G. An Analysis of Simulated Long-Term Soil Moisture Data for Three Land Uses under Contrasting Hydroclimatic Conditions in the Northern Great Plains. J. Hydrometeorol. 2004, 5, 160–179. [CrossRef] 8. Zhang, X.; Ma, Y.H.; Anlauf, R. Forecast and Analysis of Soil Moisture Based on SIMPEL model. J. Agric. Sci. Technol. 2013, 14, 490–493. 9. Holland, J.E.; Biswas, A. Predicting the mobile water content of vineyard soils in New South Wales, Australia. Agric. Water Manag. 2015, 148, 34–42. [CrossRef] 10. Hu, W.; Si, B.C. Soil water prediction based on its scale-speciﬁc control using multivariate empirical mode decomposition. Geoderma 2013, 193–194, 180–188. [CrossRef] 11. Prasad, R.; Ravinesh, C.; Li, Y.; Maraseni, T. Weekly soil moisture forecasting with multivariate sequential, ensemble empirical mode decomposition and Boruta-random forest hybridizer algorithm approach. Catena 2019, 177, 149–166. [CrossRef] 12. Shoaib, M.; Shamseldin, A.Y.; Melville, B.W.; Khan, M.M. A comparison between wavelet based static and dynamic neural network approaches for runoff prediction. J. Hydrol. 2016, 535, 211–225. [CrossRef] 13. Kamilaris, A.; Francesc, X. Deep learning in agriculture: A survey. Comput. Electron. Agric. 2018, 147, 70–90. [CrossRef] 14. Yalcin, H. An Approximation for A Relative Crop Yield Estimate from Field Images Using Deep Learning. In Proceedings of the International Conference on Agro-Geoinformatics (Agro-Geoinformatics), Istanbul, Turkey, 16–19 July 2019. 15. Yu, J.; Tang, S.; Zhangzhong, L.; Zheng, W.; Xu, L. A Deep Learning Approach for Multi-Depth Soil Water Content Prediction in Summer Maize Growth Period. IEEE Access 2020, 8, 199097–199110. [CrossRef] 16. Fathi, M.T.; Ezziyyani, M.; Ezziyyani, M.; Mamoune, S.E. Crop Yield Prediction Using Deep Learning in Mediterranean Region. In Proceedings of the Advanced Intelligent Systems for Sustainable Development (AI2SD’2019), Marrakech, Morocco, 8–11 July 2019. 17. Ji, R.; Li, X.; Zhang, S.; Zheng, L. Prediction of soil moisture in multiple depth based on time delay neural network. Trans. Chin. Soc. Agric. Eng. 2017, 33, 132–136. 18. Gill, M.K.; Asefa, T.; Kemblowski, M.W.; McKee, M. Soil moisture predition using support vector machines. J. Am. Water Resour. Assoc. 2006, 42, 1033–1046. [CrossRef] 19. Pan, J.; Shangguan, W.; Li, L.; Yuan, H.; Zhang, S.; Lu, X.; Wei, N.; Dai, Y. Using data-driven methods to explore the predictability of surface soil moisture with FLUXNET site data. Hydrol. Process. 2019, 33, 2978–2996. [CrossRef] 20. Tharani, P.P.; Baranidharan, B. An Analysis on Application of Deep Learning Techniques for Precision Agriculture. In Pro- ceedings of the International Conference on Inventive Research in Computing Applications (ICIRCA), Coimbatore, India, 2–4 September 2021. 21. Gumiere, S.J.; Camporese, M.; Botto, A.; Lafond, J.A.; Paniconi, C.; Gallichand, J.; Rousseau, A.N. Machine Learning vs. Physics-Based Modeling for Real-Time Irrigation Management. Front. Water 2020, 2, 8. [CrossRef] 22. Li, P.; Zha, Y.; Shi, L.; Tso, C.-H.; Zhang, Y.; Zeng, W. Comparison of the use of a physical-based model with data assimilation and machine learning methods for simulating soil water dynamics. J. Hydrol. 2020, 584, 124692. [CrossRef] 23. Liu, D.; Liu, C.; Tang, Y.; Gong, C. A GA-BP Neural Network Regression Model for Predicting Soil Moisture in Slope Ecological Protection. Sustainability 2022, 14, 1386. [CrossRef] 24. Li, Q.; Li, Z.; Shangguan, W.; Wang, X.; Li, L.; Yu, F. Improving soil moisture prediction using a novel encoder-decoder model with residual learning. Comput. Electron. Agric. 2022, 195, 106816. [CrossRef] 25. Prakash, S.; Sharma, A.; Sahu, S.S. Soil Moisture Prediction Using Machine Learning. In Proceedings of the Second International Conference on Inventive Communication and Computational Technologies (ICICCT), Coimbatore, India, 20–21 April 2018. 26. Adeyemi, O.; Grove, I.; Peets, S.; Domun, Y.; Norton, T. Dynamic Neural Network Modelling of Soil Moisture Content for Predictive Irrigation Scheduling. Sensors 2018, 18, 3408. [CrossRef] [PubMed] 27. Xu, J.W.; Zhao, J.F.; Zhang, W.C.; Xu, X.X. A Novel Soil Moisture Predicting Method Based on Artiﬁcial Neural Network and Xinanjiang Model. Adv. Mater. Res. 2010, 121–122, 1028–1032. [CrossRef] 28. Li, N.; Zhang, Q.; Yang, F.X.; Deng, Z.L. Research of adaptive genetic neural network algorithm in soil moisture prediction. Comput. Eng. Appl. 2018, 54, 54–59+69. 29. Notarnicola, C.; Angiulli, M.; Posa, F. Soil moisture retrieval from remotely sensed data: Neural network approach versus Bayesian method. IEEE Trans. Geosci. Remote Sens. 2008, 46, 547–557. [CrossRef] 30. Wei, W.; Zhang, J.; Zhou, L.; Xie, B.; Zhou, J.; Li, C. Comparative evaluation of drought indices for monitoring drought based on remote sensing data. Environ. Sci. Pollut. Res. 2021, 28, 20408–20425. [CrossRef] 31. Sandholt, I.; Rasmussen, K.; Andersen, J. A simple interpretation of the surface temperature/vegetation index space for assessment of surface moisture status. Remote Sens. Environ. 2002, 79, 213–224. [CrossRef] 32. Zheng, W.; Zhangzhong, L.; Zhang, X.; Wang, C.; Zhang, S.; Sun, S.; Niu, H. A Review on the Soil Moisture Prediction Model and Its Application in the Information System. In Proceedings of the Computer and Computing Technologies in Agriculture XI, Jilin, China, 12–15 August 2017. 33. Jiang, A.J.; Peng, H.Y.; Wang, B.M. The analyses of Jiangsu climate variety in forty years. J. Meteorol. Sci. 2006, 26, 525–529. 34. Qi, Y.; Darilek, J.L.; Huang, B.; Zhao, Y.; Sun, W.; Gu, Z. Evaluating soil quality indices in an agricultural region of Jiangsu Province, China. Geoderma 2009, 149, 325–334. [CrossRef] 35. Wang, J.Q.; Zhao, Y.F.; Ren, Z.H.; Gao, J. Design and Veriﬁcation of Quality Control Methods for Automatic Soil Moisture Observation Data in China. Meteorology 2018, 44, 244–257. Agriculture 2023, 13, 927 17 of 17 36. Wang, S.; Fu, G. Modelling soil moisture using climate data and normalized difference vegetation index based on nine algorithms in alpine grasslands. Front. Environ. Sci. 2023, 11, 1130448. [CrossRef] 37. Chen, T.; Guestrin, C. XGBoost: A Scalable Tree Boosting System. In Proceedings of the ACM, San Francisco, CA, USA, 13–17 August 2016. 38. Bergstra, J.; Bengio, Y. Random search for hyper-parameter optimization. J. Mach. Learn. Res. 2012, 13, 281–305. 39. Kohavi, R. A study of cross-validation and bootstrap for accuracy estimation and model selection. Int. Jt. Conf. Artif. Intell. 1995, 14, 1137–1145. 40. Eisenman, R.L. A proﬁt-sharing interpretation of shapley value for n-person games. Syst. Res. Behav. Sci. 1967, 12, 396–398. [CrossRef] 41. Niazkar, M. Assessment of artiﬁcial intelligence models for calculating optimum properties of lined channels. J. Hydroinform. 2020, 22, 1410–1423. [CrossRef] 42. Agatonovic-Kustrin, S.; Beresford, R. Basic concepts of artiﬁcial neural network (ANN) modeling and its application in pharma- ceutical research. J. Pharm. Biomed. Anal. 2000, 22, 717–727. [CrossRef] 43. Biau, G. Analysis of a random forests model. J. Mach. Learn. Res. 2012, 13, 1063–1095. 44. Cherkassky, V.; Ma, Y. Practical selection of SVM parameters and noise estimation for SVM regression. Neural Netw. 2004, 17, 113–126. [CrossRef] 45. Matei, O.; Rusu, T.; Petrovan, A.; Mihu¸ t, G. A Data Mining System for Real Time Soil Moisture Prediction. Procedia Eng. 2017, 181, 837–844. [CrossRef] 46. Nguyen, T.T.; Ngo, H.H.; Guo, W.; Chang, S.W.; Nguyen, D.D.; Nguyen, C.T.; Zhang, J.; Liang, S.; Bui, X.T.; Hoang, N.B. A low-cost approach for soil moisture prediction using multi-sensor data and machine learning algorithm. Sci. Total Environ. 2022, 833, 155066. [CrossRef] 47. Filipovi, N.; Brdar, S.; Mimi, G.; Marko, O.; Crnojevi, V. Regional soil moisture prediction system based on long short-term memory network. Biosyst. Eng. 2022, 213, 30–38. [CrossRef] 48. Li, Q.; Zhu, Y.; Shangguan, W.; Wang, X.; Li, L.; Yu, F. An attention-aware LSTM model for soil moisture and soil temperature prediction. Geoderma 2022, 409, 115651. [CrossRef] 49. Cai, Y.; Zheng, W.; Zhang, X.; Zhangzhong, L.; Xue, X. Research on soil moisture prediction model based on deep learning. PLoS ONE 2019, 14, e0214508. [CrossRef] [PubMed] 50. Bell, J.E.; Sherry, R.; Luo, Y. Changes in soil water dynamics due to variation in precipitation and temperature: An ecohydrological analysis in a tallgrass prairie. Water Resour. Res. 2010, 46, W03523. [CrossRef] 51. Feng, H.; Liu, Y. Combined effects of precipitation and air temperature on soil moisture in different land covers in a humid basin. J. Hydrol. 2015, 531, 1129–1140. [CrossRef] 52. Ragab, R. Towards a continuous operational system to estimate the root-zone soil moisture from intermittent remotely sensed surface moisture. J. Hydrol. 1995, 173, 1–25. [CrossRef] 53. Yan, H.; Dechant, C.; Hamid, M. Improving Soil Moisture Proﬁle Prediction with the Particle Filter-Markov Chain Monte Carlo Method. IEEE Trans. Geosci. Remote Sens. 2015, 53, 6134–6147. [CrossRef] 54. Huang, Y.; Jiang, H.; Wang, W.F.; Wang, W.; Sun, D. Soil moisture content prediction model for tea plantations based on SVM optimised by the bald eagle search algorithm. Cogn. Comput. Syst. 2021, 3, 351–360. [CrossRef] 55. Wang, X.; Lv, J.; Wang, C.; Xie, D. Soil moisture content prediction using wavelet transform and support vector machine with genetic algorithm optimization. ICIC Express Lett. Part B Appl. 2014, 5, 1141–1148. Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Journal

Agriculture – Multidisciplinary Digital Publishing Institute

Published: Apr 24, 2023

Keywords: soil moisture; prediction; XGBoost algorithm; SHAP

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Research on Provincial-Level Soil Moisture Prediction Based on Extreme Gradient Boosting Model

Research on Provincial-Level Soil Moisture Prediction Based on Extreme Gradient Boosting Model

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Research on Provincial-Level Soil Moisture Prediction Based on Extreme Gradient Boosting Model

Research on Provincial-Level Soil Moisture Prediction Based on Extreme Gradient Boosting Model

References (50)

Abstract

Journal

Recommended Articles

There are no references for this article.

Our policy towards the use of cookies