Access the full text.
Sign up today, get DeepDyve free for 14 days.
V. Mehdipour, M. Memarianfard, F. Homayounfar (2017)
Application of Gene Expression Programming to water dissolved oxygen concentration prediction, 2
Bo-guang Sun, Linxiu Zhang, Lin-zhang Yang, Fusuo Zhang, D. Norse, Zhao-liang Zhu (2012)
Agricultural Non-Point Source Pollution in China: Causes and Mitigation MeasuresAMBIO, 41
Balraj Singh, Parveen Sihag, Karan Singh (2017)
Modelling of impact of water quality on infiltration rate of soil by random forest regressionModeling Earth Systems and Environment, 3
V. Atluri, C. Hung, T. Coleman (1999)
An artificial neural network for classifying and predicting soil moisture and temperature using Levenberg-Marquardt algorithmProceedings IEEE Southeastcon'99. Technology on the Brink of 2000 (Cat. No.99CH36300)
Sinkyu Kang, Sumi Kim, S. Oh, Dowon Lee (2000)
Predicting spatial and temporal patterns of soil temperature based on topography, surface cover and air temperatureForest Ecology and Management, 136
S. Mehdizadeh, J. Behmanesh, K. Khalili (2017)
Evaluating the performance of artificial intelligence methods for estimation of monthly mean soil temperature without using meteorological dataEnvironmental Earth Sciences, 76
Wei Wu, Xiao-Ping Tang, Nai-Jia Guo, Chao Yang, Hongbin Liu, Yue Shang (2013)
Spatiotemporal modeling of monthly soil temperature using artificial neural networksTheoretical and Applied Climatology, 113
Parveen Sihag (2018)
Prediction of unsaturated hydraulic conductivity using fuzzy logic and artificial neural networkModeling Earth Systems and Environment, 4
Parveen Sihag, N. Tiwari, S. Ranjan (2018)
Support vector regression-based modeling of cumulative infiltration of sandy soilISH Journal of Hydraulic Engineering, 26
P. Talaee (2014)
Daily soil temperature modeling using neuro-fuzzy approachTheoretical and Applied Climatology, 118
A. Parsaie, A. Haghiabi (2015)
Predicting the longitudinal dispersion coefficient by radial basis function neural networkModeling Earth Systems and Environment, 1
A. Parsaie, H. Yonesi, Shadi Najafian (2015)
Predictive modeling of discharge in compound open channel by support vector machine techniqueModeling Earth Systems and Environment, 1
J. Shiri, Ö. Kisi (2011)
Comparison of genetic programming with neuro-fuzzy systems for predicting short-term water table depth fluctuationsComput. Geosci., 37
M. Bilgili (2010)
Prediction of soil temperature using regression and artificial neural network modelsMeteorology and Atmospheric Physics, 110
W. McCulloch, W. Pitts (1990)
A logical calculus of the ideas immanent in nervous activityBulletin of Mathematical Biology, 52
O. Kisi, M. Tombul, M. Kermani (2015)
Modeling soil temperatures at different depths by using three different neural computing techniquesTheoretical and Applied Climatology, 121
H. Citakoglu (2015)
Comparison of artificial intelligence techniques via empirical equations for prediction of solar radiationComput. Electron. Agric., 118
E. Öztürk, E. Atsan, T. Polat, K. Kara (2011)
Variation in heavy metal concentrations of potato (Solanum tuberosum L.) cultivars.Journal of Animal and Plant Sciences, 21
Parveen Sihag, N. Tiwari, S. Ranjan (2019)
Prediction of unsaturated hydraulic conductivity using adaptive neuro- fuzzy inference system (ANFIS)ISH Journal of Hydraulic Engineering, 25
R. Schaetzl, Bruce Knapp, S. Isard (2005)
Modeling soil temperatures and the mesic-frigid boundary in the central Great Lakes region, 1951-2000Soil Science Society of America Journal, 69
Behnaz Nahvi, J. Habibi, Kasra Mohammadi, S. Shamshirband, Othman Razgan (2016)
Using self-adaptive evolutionary algorithm to improve the performance of an extreme learning machine for estimating soil temperatureComput. Electron. Agric., 124
R. Singh (2000)
Environmental consequences of agricultural development: a case study from the green revolution state of Haryana, India.Agriculture, Ecosystems & Environment, 82
L. Breiman (2001)
Random ForestsMachine Learning, 45
M. Ghorbani, R. Khatibi, A. Goel, M. Fazelifard, A. Azani (2016)
Modeling river discharge time series using support vector machine and artificial neural networksEnvironmental Earth Sciences, 75
T. Kätterer, O. Andrén (2009)
Predicting daily soil temperature profiles in arable soils in cold temperate regions from air temperature and leaf area indexActa Agriculturae Scandinavica, Section B — Soil & Plant Science, 59
H. Citakoglu (2017)
Comparison of artificial intelligence techniques for prediction of soil temperatures in TurkeyTheoretical and Applied Climatology, 130
M. Firat, M. Güngör (2009)
Generalized Regression Neural Networks and Feed Forward Neural Networks for prediction of scour depth around bridge piersAdv. Eng. Softw., 40
L. Breiman (1996)
Bagging PredictorsMachine Learning, 24
O. Kisi, H. Sanikhani, M. Cobaner (2017)
Soil temperature modeling at different depths using neuro-fuzzy, neural network, and genetic programming techniquesTheoretical and Applied Climatology, 129
Adalberto Siqueira, C. Tiba, N. Fraidenraich (2010)
Generation of daily solar irradiation by means of artificial neural net worksRenewable Energy, 35
Sungwon Kim, V. Singh (2014)
Modeling daily soil temperature using data-driven models and spatial distributionTheoretical and Applied Climatology, 118
H. Tabari, B. Somee, M. Zadeh (2011)
Testing for long-term trends in climatic variables in IranAtmospheric Research, 100
Parveen Sihag, Balraj Singh, Alireza Vand, V. Mehdipour (2020)
Modeling the infiltration process with soft computing techniquesISH Journal of Hydraulic Engineering, 26
N. Tiwari, Parveen Sihag, Sanjeev Kumar, S. Ranjan (2018)
Prediction of trapping efficiency of vortex tube ejectorISH Journal of Hydraulic Engineering, 26
V. Mehdipour, M. Memarianfard (2017)
Application of Support Vector Machine and Gene Expression Programming on Tropospheric ozone Prognosticating for Tehran MetropolitanCivil Engineering Journal, 3
M. Mohandes, S. Rehman, T. Halawani (1998)
Estimation of global solar radiation using artificial neural networksRenewable Energy, 14
Andy Liaw, M. Wiener (2007)
Classification and Regression by randomForest
GEOLOGY, ECOLOGY, AND LANDSCAPES 2020, VOL. 4, NO. 3, 203–215 INWASCON https://doi.org/10.1080/24749508.2019.1610841 RESEARCH ARTICLE Model-based soil temperature estimation using climatic parameters: the case of Azerbaijan Province, Iran a b c d Parveen Sihag , Fatemeh Esmaeilbeiki , Balraj Singh and Siraj Muhammed Pandhiani a b Civil Engineering Department, National Institute of Technology, Kurukshetra, India; Department of Soil Science and Engineering, University of Tabriz, Tabriz, Iran; Department of Civil Engineering, Panipat Institute of Engineering and Technology, Panipat, India; Department of General Studies, University College Jubail, Jubail, Saudi Arabia ABSTRACT ARTICLE HISTORY Received 9 July 2018 Estimating soil temperature (ST) proﬁle is identiﬁed as essential knowledge for plants, crop Accepted 20 April 2019 growth, and germination in all agriculture regions. In this study, daily soil temperature (DST) was modeled using Multilayer perceptron (MLP) model, Gaussian Process (GP), Random Forest KEYWORDS (RF), and the M5P model methods for estimating and comparing DST in arid regions. The data Daily soil temperature; selected to test the proposed models are obtained from two stations in Tabriz and Ahar, Multilayer perceptron located in the Azerbaijan province of Iran. Input dataset includes air temperature, relative model; Gaussian Process; humidity, wind speed, and sunshine as dependent parameters, whereas ST at depths of 5 cm Random Forest; M5P model was selected for the target in model development. The results show the MLP works better than GP-, RF-, and M5P-based models in estimating the DST, with excellent performance indicators such as the mean absolute error, root mean square error, and coeﬃcient of correlation. Results showed that the MLP model with RMSE = 3.2626°C was more suitable than other models in ST estimation 2 days ahead for Tabriz station. Also, in Ahar, MLP with RMSE = 6.3332°C was more suitable than GP-, RF-, and M5P-based models for estimating DST. As a conclusion, the developed MLP is recommended for estimating the DST proﬁles. 1. Introduction received much attention in recent years due to the development and usage of the latest modeling methods Climatic and weather conditions are dependent upon and software. Due to seldom availability of comprehen- several factors among which soil temperature (ST) is an sive ST data (Schaetzl, Knapp, & Isard, 2005)and diﬃ- essential parameter in exchanging the heat and energy culties in spatial measurement (Kang, Kim, Oh, & Lee, between soil surface and atmosphere. The information 2000) regarding the frequent collection of large-scale about ST is needed for many agricultural (agronomy, data, forecasting of ST needs to be estimated from other soil science) and engineering applications (geo-technol- informative metrological variables using theoretical, ogy, hydrology, meteorology, ecology, and environ- empirical, or soft computing models. mental studies). Moreover, it assists agronomists and Bilgili (2010) investigated the potential of regres- engineers to decide the proper plantation date, design sion and multilayer perceptron model (MLP) models drainage and irrigation systems, and optimum utiliza- in estimating ST using climate data (atmospheric tion of pesticides and fertilizers to reduce chemical pressure, solar radiation, and atmospheric tempera- pollution of soils and groundwater (Singh, 2000;Sun ture) in the Adana, Turkey and observed that MLP et al., 2012). Mutually with chemical and physical fea- model has higher performance than the regression tures of soil organic material, ST is one of the main models. Öztürk et al. (2011) choose meteorological parameters contributing to the biological activity of soil data of diﬀerent stations to estimate STs using ANN (Kätterer & Andrén, 2009), e.g., soil respiration, micro- model and found very satisfactory results. Wu et al. bial decomposition, organic matter storage, and miner- (2013) implemented the MLP and indicated MLP as a alization, etc. (Öztürk, Atsan, Polat, & Kara, 2011). Soil capable tool for estimation of STs. Kim and Singh thermal conditions largely aﬀect the plant growth and (2014) also suggested that adaptive neuro-fuzzy infer- agricultural crop yield, due to which, in many cases, ST ence system (ANFIS) and ANN models as an eﬃcient has relatively higher importance than surface air tem- tool for estimating STs at diﬀerent depths. perature (Hillel, 1998). Talaee (2014) estimated daily soil temperature (DST) Atluri, Hung, and Coleman (1999)designed an arti- using coactive neuro-fuzzy inference system (CANFIS) ﬁcial neural network (ANN) for the classiﬁcation of for semi-arid and arid areas. The DST data were collected soils and predicted soil moisture and temperature from two metrological places at six diﬀerent levels. The based on remotely sensed data. Forecasting and simula- satisfactory performance of the CANFIS is reported for tions related to temporal and spatial predictions of ST CONTACT Parveen Sihag parveen12sihag@gmail.com Civil Engineering Department, National Institute of Technology, Kurukshetra, India © 2019 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. 204 P. SIHAG ET AL. both areas. Kisi, Tombul, and Kermani (2015) compared Ranjan, 2018). Furthermore, the mentioned methods diﬀerent methods and models including generalized have been also used for water and air quality prognos- regression neural networks, radial basis neural networks, ticating (Mehdipour & Memarianfard, 2017a, 2017b; and MLP neural networks to estimate STs by using Mehdipour, Memarianfard, & Homayounfar, 2017)and meteorological data. Nahvi, Habibi, Mohammadi, the outcomes were very acceptable. As per the author’s Shamshirband, and Al Razgan (2016)haveforecasted best knowledge, no one has to implement GP-, M5P-, DST using self-adaptive evolutionary extreme learning andRF-basedmodelstoestimatethe STsatdiﬀerent soil machine for six diﬀerent soil depths. Data were collected depths. Our main objective in this study is to estimate the from two metrological stations (Kerman and Bandar ST at a depth of 5 cm. In this manner, it is possible to Abbas in Iran). They selected meteorological parameters estimate forthcoming STs by simply acquiring the cli- (global solar radiation, atmospheric pressure, and air matic data from the meteorological stations. This is temperature) as input and ST as output. Results of this particularly important for future planning of agriculture study conﬁrmed that the performance of self-adaptive and hydrological process. This study was carried out evolutionary extreme learning machine is better than using the climate data of 3 years (daily) statistical period extreme learning machine. The performance of self- of 2013–2015 for Tabriz and Ahar (East Azarbaijan adaptive evolutionary extreme learning machine is also Province). With the presence of meteorological informa- better than ANN and genetic programming. Citakoglu tion, the ST of the soil was estimated at a depth of 5 cm. (2017) used ANN, ANFIS, and multiple linear regression In this study, the performance of various data mining (MLR) models for estimating ST of Turkish, Turkey and techniques such as MLP model, GP, RF, and M5P model reported that ANFIS is better than ANN- and MLR- for estimating the ST was evaluated for estimating the based models. Kisi, Sanikhani, and Cobaner (2017)esti- DST at a depth of 5 cm in semi-arid regions such as Iran. mated ST at diﬀerent depths of soil using ANN, ANFIS, The most important reason to choose these depths is that and gene expression programming (GEP) techniques for the depth of 5 cm is important for seeds germination Adana and Mersin, Turkey. The authors found that GEP since seeds “waking up” process extremely relies on ST has better performance than ANFIS and ANN. and soil moisture (Miles & Brown, 2007). Mehdizadeh, Behmanesh, and Khalili (2017)alsoused ANN, ANFIS, and GEP models for the estimation of 2. Materials and methods monthly ST for 31 locations in Iran and found that the 2.1. Area of study performance of ANFIS model is superior to ANN- and GEP-based models. This study was carried out using data from East Last few years, soft computing techniques such as Azarbaijan province, weather stations of Tabriz and support vector machine, Gaussian Process (GP), Ahar (Figure 1). East Azarbaijan Province is the largest Random Forest (RF), ANNs, M5P tree, and ANFIS has and most populous province of northwestern Iran successfully been implemented to solve complex and (Azerbaijan) and its geographic location is in the range non-linear problems (Parsaie & Haghiabi, 2015; of 45°7′ to 48°20′ east longitude and 36°45′ ‘to 39°26′ Parsaie, Yonesi, & Najaﬁan, 2015;Shiri &KişI, 2011; north latitude. The capital of the East Azerbaijan pro- Sihag, 2018;Sihag,Tiwari, &Ranjan, 2018a, 2017; vince is Tabriz metropolis, which is about 1348–1561 m Singh, Sihag, & Singh, 2017; Tiwari, Sihag, Kumar, & above sea level. The average annual rainfall is 310 mm Figure 1. Location of the Tabriz and Ahar synoptic stations in Iran (East Azerbaijan Province). GEOLOGY, ECOLOGY, AND LANDSCAPES 205 and the number of freezing days is 104 days. The area of of input attributes. The splitting approach is applied at East Azarbaijan is 45 km . In terms of agriculture and in each node instead to gain the maximum information to terms of the number of wheat silos, East Azarbaijan is minimize the variation in the intra-subset class value ranked ﬁrst in Iran. Cultivation of cereals, especially down to each branch. The splitting process will be con- wheat and barley, is one of the most important crops of verged when there are diminutive variations among the the province. The most important cereals cultivated in class values of the instances or left only a few instances or this area are wheat, barley, rice, and corn. The best when the tree is pruned back. The developed tree shows potatos are grown in East Azarbaijan province. the very good structure and prediction accuracy due to Ahar is one of the major cities of the East showing more potential linearity at the leaf node. Azerbaijan province and the city center of Ahar in the north of the province. The area is about 13 km . 2.4. Random forests The city of Ahar is located at an altitude of 1360 m above sea level. This city is located in a mountainous RF is a highly ﬂexible assembly of decision trees that region. The average annual rainfall of Ahar is 350 mm. perform well for linear and nonlinear estimation by The maximum temperature of Ahar is 34°C and the adjusting variance and bias (Breiman, 1996, 2001). minimum temperature below 0°C. This assembly learning process is identiﬁed as “bag- ging” as it develops trees in which consecutive trees do not base on former trees. Every tree is separately esti- 2.2. Soft computing techniques mated using a bootstrap sample of the dataset and a simple commonly vote is considered for ﬁnal estima- 2.2.1. GP regression tion (Liaw & Wiener, 2002). RF model requires two GP regression relies upon the postulation that nearby speciﬁc standard parameters: quantity of input vari- study must share the knowledge mutually and it is an ables (m) implemented at every node to develop a tree approach of state of earlier straight over the function and the number of trees to be developed (k). Thus, the space. The generalization of Gaussian distribution is RF regression contains k number of trees, which is known as Gaussian regression. The matrix and vector searched through for the best split. of Gaussian distribution are articulated as covariance and mean in GP regression. Due to having earlier knowledge of functional dependence and data, the 2.5. Multilayer perceptron validation for generalization is not crucial. The GP McCulloch and Pitts (1943)proposedANNforthe ﬁrst regression models are capable to distinguish the fore- time. In general, the layered perceptron neural network cast distribution consequent to the input test data structure contains three layers: input, hidden, and the (Rasmussen & Williams, 2006). target layer. Each layer includes the number of neurons A GP is the choice of numbers of the random (Figure 2). Thenumberofneurons in the input and variable, any ﬁnite number of them has a collective output layers is determined by the nature of the problem multivariate Gaussian distribution. Assume p and q underconsideration, while thenumberofneurons in the stand for input and target domain accordingly, there- hidden layers, as well as the number of these layers, is upon x pairs (g , h ) are drawn liberally and equiva- i i determined by the trial and errors to reduce the amount lently distribution. For regression, it is assumed that h of errorby theuser(Firat&Gungor, 2009;Schalkoﬀ, ⊆ R then a GP on p is represented by the mean 1997). Each of the neurons in the input layer is weighted function V : p ! Re and covariance function whose value determines the eﬀect of each variable on the μ : p p ! Re. Readers are requested to follow the input layer performance. Each neuron consists of two Kuss (2006) to get the exhaustive details of GP. parts: in the ﬁrst part of it, the weighted sum of the input Pearson VII kernel function is used in this study values is computed and in the second part of the neuron, Pearson VII function kernelðÞ PUK ! ,"# ω the output of the ﬁrst part is located in a mathematical qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ðÞ 1=ω ¼ 1 1þ 2 x x 2 1 σ i j function and through which the output of the neuron is calculated. This mathematical function is referred to as Here Gaussian noise, σ and ω are the kernel con- the actuator function, the threshold function, or the straints of GP_PUK model. transfer function which functions as a nonlinear ﬁlter and makes the output of the neuron in a particular number range (Ghorbani, Khatibi, Goel, FazeliFard, & 2.3. M5P model (M5P) Azani, 2016). M5P tree, ﬁrst time introduced by Quinlan (1992), is used to develop a decision tree by engaging the linear 2.6. Data analysis regression function approach at nodes with an aim of constructing a model which proposes a correlation The dataset contains ST and other meteorological para- among the target value of the training cases and value meters including variables such as air temperature, 206 P. SIHAG ET AL. Figure 2. Simple conﬁguration of multilayer perceptron neural network. relative humidity, wind speed, sunshine hours, and ST are given below. Lower values of RMSE, MAE and at depths of 5 cm, which have been taken from Tabriz higherCC values indicate that developed model is and Ahar meteorological station of Iran. Dataset used in more perfect (Sihag, Singh, SepahVand, & this study was recorded between the years of 2013 and Mehdipour, 2018b), where m and n are observational 2015. The average air temperature lies between the and computational estimates of the ST at the time of −13°C and 32.2°C in Tabriz and −13.5°C and 28.2°C i, a number of data in Ahar. The range of the ST at 5 cm depth lies between −8.13°C and 39°C in Tabriz and −5.27°C to 35.93°C in MAE ¼ ð jj m n (1) Ahar. Table 1 represents the features of the dataset used i¼1 in this study. The data scale is the daily average. Table 2 sﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ represents the input variables and input combination X RMSE ¼ ð ðÞ m n (2) that are considered for modeling. 75% of the total i¼1 dataset was selected for the training the model and P P P rest 25% was used for the testing the developed models. a mn ð mÞð nÞ CC ¼ qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃqﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ P P P P 2 2 2 2 að m Þ ð mÞ að n Þð nÞ 3. Models evaluation (3) After analyzing data with a variety of methods and models, their performance needs to be evaluated. There are various ways to do this. The most impor- 4. Result and discussion tant of these methods is to compare the observational 4.1. Results of GP and computational values of models using evaluation criteria. In this study mean absolute error (MAE), The Pearson VII kernel function-based GP model was root mean square error (RMSE), and the coeﬃcient implemented using WEKA 3.9. The optimum values of of determination (CC) are used to evaluate the per- GP_PUK models with Gaussian noise = 0.01, ω =0.1, formance of developed models and their relationships and σ = 1 are kept constant for all models for the fair Table 1. Statistical data of climate variables of Tabriz and Ahar stations. Tabriz Variables Skewness Stretch index Average Min Max Standard deviation Average air temperature (°C) −0.21 −0.9 13.58 −13 32.2 10.26 Sunny hours average (h) −0.78 −0.45 8.01 0 13.5 3.74 Average relative humidity (%) 0.23 −1.03 51.18 14.12 94.37 18.21 −1 Average wind speed (km·h ) 0.73 0.012 3.7 0 10.62 1.71 Soil temperature at 5 cm depth (°C) 0.005 −1.31 17.42 −8.13 39 12.43 Ahar Variables Skewness Stretch index Average Max Min Standard deviation Average air temperature (°C) −0.29 −0.82 11.67 −13.5 28.2 8.62 Sunny hours average (h) −0.59 −0.77 7.42 0 13/6 3.91 Average relative humidity (%) 0.28 −0.48 58.21 22.62 97 15.28 −1 Average wind speed (km·h ) 1.21 1.24 3.25 0 15.75 1.71 Soil temperature at 5 cm depth (°C) 0.095 −1.36 14.88 −5.27 35.93 11.22 GEOLOGY, ECOLOGY, AND LANDSCAPES 207 Table 2. Selecting diﬀerent modes for modeling soil temperature estimates with a delay of 2 days for depth of 5 cm. Structure Entrance Output 1 T(t − 2) Soil temperature at a depth of 5 cm 2 T(t − 2) + RH(t − 2) Soil temperature at a depth of 5 cm 3 T(t − 2) + RH(t − 2) + SUN(t − 2) Soil temperature at a depth of 5 cm 4 T(t − 2) + RH(t − 2) + SUN(t − 2) + W(t − 2) Soil temperature at a depth of 5 cm 5 T(t−2) + RH(t − 2) + W(t − 2) Soil temperature at a depth of 5 cm Table 3. The results of modeling with Pearson VII kernel function-based GP models for Tabriz and Ahar at a depth of 5 cm. Tabriz Training Testing Models MAE (°C) RMSE (°C) CC MAE (°C) RMSE (°C) CC GP_PUK1 2.4368 3.1289 0.9650 3.0404 3.9284 0.9586 GP_PUK2 0.0631 0.3228 0.9996 3.9947 5.1962 0.9270 GP_PUK3 0.0019 0.0032 1.0000 2.8156 3.5798 0.9666 GP_PUK4 0.0009 0.0012 1.0000 2.7741 3.4742 0.9696 GP_PUK5 0.0020 0.0047 1.0000 2.9947 3.7844 0.9622 Ahar GP_PUK1 4.5014 5.6524 0.8524 6.0310 7.3045 0.8195 GP_PUK2 0.1044 0.4931 0.9990 7.3778 9.2538 0.6973 GP_PUK3 0.0034 0.0094 1.0000 5.9536 7.3060 0.8074 GP_PUK4 0.0019 0.0048 1.0000 5.4103 6.6202 0.8562 GP_PUK5 0.0039 0.0084 1.0000 6.0319 7.6266 0.8024 Bold values indicate the best performing model. comparison among models. Table 3 represents the average wind speed are provided in Table 2) performs results obtained (MAE, RMSE, and CC) by GP_PUK relatively better than other models with testing model for Tabriz and Ahar metrological stations respec- MAE = 2.7741°C, RMSE = 3.4742°C, and CC = 0.9696 tively. These assessment criteria values shows that model for Tabriz and MAE = 5.4103°C, RMSE = 6.6202°C, and GP_PUK4 (dependent variables of average air tempera- CC = 0.8562 for Ahar stations. ture, average sunny hours, average relative humidity, and Actual GP_PUK4 Tabriz -5 0 50 100 150 200 250 Time (days) Tabriz GP_PUK4 -5 -5 5 15253545 Actual Soil Temperature ( C) Figure 3. Comparison of actual and predicted soil temperature at 5 cm and agreement diagram for Tabriz metrological station – Pearson VII kernel function-based GP model, testing period. 0 0 Predicted Soil Temperature ( C) Soil Temperature ( C) 208 P. SIHAG ET AL. Actual GP_PUK4 Ahar -5 0 50 100 150 200 250 Time (days) GP_PUK4 Ahar -5 -5 5 15 25 35 45 Actual Soil Temperature ( C) Figure 4. Comparison of actual and predicted soil temperature at 5 cm and agreement diagram for Ahar metrological station – Pearson VII kernel function-based GP model, testing period. Figures 3 and 4 present the detail of actual and esti- 4.2. Results of the M5P model mated ST at 5 cm depth and their corresponding scatter The M5P model was implemented using WEKA 3.9. plot for the best GP_PUK model using testing dataset for The process of development of M5P model is same as Tabriz and Ahar stations respectively. It is clearly noted GP_PUK model trial and error process Table 4 pre- from Table 3 and Figures 4 and 5 that GP-PUK4 model is sents the results obtained (MAE, RMSE, and CC) by suitable for estimating the ST of both stations. GP_PUK model for Tabriz and Ahar metrological Actual M5P4 Tabriz -5 0 50 100 150 200 250 Time (days) M5P4 Tabriz -5 -5 5 15253545 Actual Soil Temperature ( C) Figure 5. Comparison of actual and predicted soil temperature at 5 cm and agreement diagram for Tabriz metrological station – M5P model, testing period. 0 0 Predicted Soil Temperature ( C) Soil Temperature ( C) Soil Temperature ( C) Predicted Soil Temperature ( C) GEOLOGY, ECOLOGY, AND LANDSCAPES 209 Table 4. The results of modeling with M5P models for Tabriz and Ahar at a depth of 5 cm. Tabriz Training Testing Models MAE (°C) RMSE (°C) CC MAE (°C) RMSE (°C) CC M5P1 2.7371 3.4992 0.9560 2.6340 3.2935 0.9728 M5P2 2.5042 3.2365 0.9625 2.7241 3.4834 0.9684 M5P3 2.4425 3.1665 0.9641 2.7697 3.5270 0.9679 M5P4 2.3647 3.1031 0.9655 2.7106 3.4563 0.9692 M5P5 2.4473 3.1962 0.9634 2.6871 3.4235 0.9696 Ahar M5P1 5.0539 6.2918 0.8131 5.3557 6.6469 0.8535 M5P2 4.9483 6.0115 0.8311 5.6708 7.0755 0.8354 M5P3 4.5374 5.6238 0.8541 5.5610 7.0044 0.8299 M5P4 4.7013 5.7717 0.8456 5.5907 6.9010 0.8448 M5P5 4.8876 5.9343 0.8358 5.7101 7.1277 0.8320 Bold values indicate the best performing model. Ahar Actual M5P4 -5 0 50 100 150 200 250 Time (days) M5P4 Ahar -5 -5 5 15 25 35 45 Actual Soil Temperature ( C) Figure 6. Comparison of actual and predicted soil temperature at 5 cm and agreement diagram for Ahar metrological station – M5P model, testing period. Table 5. The results of modeling with RF models for Tabriz and Ahar at a depth of 5 cm. Tabriz Training Testing Models MAE (°C) RMSE (°C) CC MAE (°C) RMSE (°C) CC RF1 2.1748 2.8435 0.9712 2.9012 3.7893 0.9633 RF2 1.0435 1.3652 0.9936 2.9360 3.8205 0.9624 RF3 1.0221 1.3259 0.9942 2.7948 3.5322 0.9690 RF4 1.0230 1.3335 0.9943 2.8455 3.4722 0.9728 RF5 1.0157 1.3195 0.9943 2.7874 3.5619 0.9690 Ahar RF1 4.0636 5.2105 0.8762 6.3293 7.8062 0.7926 RF2 2.0039 2.5155 0.9753 6.2698 7.6817 0.7950 RF3 1.7278 2.2008 0.9817 5.7205 6.9710 0.8324 RF4 1.6979 2.1423 0.9835 5.5501 6.8549 0.8436 RF5 1.8673 2.3385 0.9799 5.8844 7.3743 0.8193 Bold values indicate the best performing model. 0 0 Predicted Soil Temperature ( C) Soil Temperature ( C) 210 P. SIHAG ET AL. Actual RF4 Tabriz -5 0 50 100 150 200 250 Time (days) RF4 Tabriz -5 -5 5 15 25 35 45 Actual Soil Temperature ( C) Figure 7. Comparison of actual and predicted soil temperature at 5 cm and agreement diagram for Tabriz metrological station – RF model, testing period. Ahar Actual RF4 -5 0 50 100 150 200 250 Time (days) RF4 Ahar -5 -5 5 15 25 35 45 Actual Soil Temperature ( C) Figure 8. Comparison of actual and predicted soil temperature at 5 cm and agreement diagram for Ahar metrological station – RF model, testing period. stations respectively. These performance evaluation RMSE = 6.9010°C, and CC = 0.8448 for Ahar criteria values show that model M5P4 performs stations. quite better than other models with testing Figures 5 and 6 display the detail of actual and MAE = 2.7106°C, RMSE = 3.4563°C, and estimated ST at 5 cm depth and their correspond- CC = 0.9692 for Tabriz and MAE = 5.5907°C, ing scatter plot for the best M5P model using 0 0 Predicted Soil Temperature ( C) 0 Soil Temperature ( C) Soil Temperature ( C) Predicted Soil Temperature ( C) GEOLOGY, ECOLOGY, AND LANDSCAPES 211 Table 6. Structure of the MLP-based model. reduced scatter. In particular, there is variation in the predicted peak values compared with the actual Tabriz values for both stations. Models Structure of the MLP Results of MLP: The MLP modeling was implemen- MLP1 1-8-1 ted using WEKA 3.9 software. In this study, single MLP2 2-1-1 hidden layer is used for developing models with learn- MLP3 3-1-1 MLP4 4-13-1 ing rate 0.2, momentum 0.1, and iteration 1500. The MLP5 3-5-1 optimum number of neurons was identiﬁed using trial Ahar and error method by changing the number of neurons MLP1 1-25-1 in the hidden layer from 1 to 25. The optimum num- MLP2 2-8-1 ber of the neurons in the hidden layer for each com- MLP3 3-14-1 bination is listed in Table 6. The performance of MLP- MLP4 4-9-1 MLP5 3-23-1 based models for Tabriz and Ahar stations is listed in Table 7 in terms of MAE, RMSE, and CC. These performance evaluation criteria values show that testing dataset for Tabriz and Ahar stations respec- model MLP4 (4-13-1) performs quite better than tively. It is clearly noted from Table 4 and Figures other models with testing MAE = 2.5525°C, 5 and 6 that M5P4 model is suitable than other RMSE = 3.2626°C, and CC = 0.9718 for Tabriz and models for estimating the ST of both Tabriz and MLP4 (4-9-1) model with MAE = 5.1261°C, Ahar stations. RMSE = 6.3332°C and CC = 0.8619 for Ahar stations. Figure 9 and 10 display the detail of actual and 4.3. Results of the RF model predicted ST at 5 cm depth and their corresponding scatter plot for the best MLP model using the testing The RF model is developed as GP_PUK and M5P dataset for Tabriz and Ahar stations respectively. It is models. In this study, the RF model with parameters clearly noted from Table 7 and Figures 9 and 10 that (k =1, m =1, I = 100) was used for ST modeling; the MLP4 model is suitable for estimating the ST of both accuracy of RF models depends on these parameters. Tabriz and Ahar stations. The results of the RF model for every data model deﬁnitions (see Table 2) are listed in Table 5 in terms of MAE, RMSE, and CC. The assessment criteria 5. Discussion indicate that the RF4 model (dependent variables of T (t − 2), RH(t − 2), SUN(t − 2), W(t − 2), see Table 2)is This section presents the comparison of prediction suitable and works better than the other models for ability of MLP-, GP-, M5P-, and RF-based models. both Tabriz and Ahar stations. The comparison of the best-selected model among all Figure 7 and 8 display the detail of actual and the four techniques was done by using the standard predicted ST at 5 cm depth and their corresponding statistical performance evaluation parameters (MAE, scatter plot for the best RF model using the testing RMSE, and CC). dataset for Tabriz and Ahar stations respectively. It is The details of the best-selected models with MLP, clearly noted from Table 5 and Figures 7 and 8 that GP, M5P, and RF are listed in Table 8 and Figure 11 the RF model is suitable for estimating the ST of both for Tabriz station. Table 8 and Figure 11 suggest that Tabriz and Ahar stations. The results show further MLP4 models work better than GP_PUK-, M5P-, and improvements in the prediction of ST in terms of RF-based models. Model 4 (dependent variables of Table 7. The results of modeling with MLP models for Tabriz and Ahar at a depth of 5 cm. Tabriz Training Testing Models MAE (°C) RMSE (°C) CC MAE (°C) RMSE (°C) CC MLP1 2.7504 3.4995 0.9560 2.6422 3.2876 0.9729 MLP2 2.7602 3.5385 0.9550 2.6912 3.4151 0.9703 MLP3 2.7237 3.5218 0.9569 2.5965 3.2724 0.9728 MLP4 2.3703 3.0819 0.9664 2.5525 3.2626 0.9718 MLP5 2.6849 3.4821 0.9564 2.5862 3.3040 0.9719 Ahar MLP1 5.2196 6.3906 0.8098 5.2436 6.3446 0.8653 MLP2 5.1136 6.1346 0.8275 5.3283 6.4985 0.8532 MLP3 4.5904 5.6653 0.8518 5.2046 6.3427 0.8618 MLP4 4.4826 5.5747 0.8570 5.1261 6.3332 0.8619 MLP5 4.7300 5.8143 0.8431 5.2867 6.7915 0.8521 Bold values indicate the best performing model. 212 P. SIHAG ET AL. Actual MLP4 Tabriz -5 0 50 100 150 200 250 Time (days) MLP4 Tabriz -5 -5 5 15253545 Actual Soil Temperature ( C) Figure 9. Comparison of actual and predicted soil temperature at 5 cm and agreement diagram for Tabriz metrological station – MLP model, testing period. Actual MLP4 Ahar -5 0 50 100 150 200 250 Time (days) Ahar MLP4 -5 -5 5 15253545 Actual Soil Temperature ( C) Figure 10. Comparison of actual and predicted soil temperature at 5 cm and agreement diagram for Ahar metrological station – MLP model, testing period. average air temperature, average sunny hours, aver- closer to the line of the perfect agreement for Tabriz age relative humidity, and average wind speed (Table station. 2)) performs better than other combinations. In Table 9 and Figure 12 suggest that MLP4 models Figure 11, predicted values of the ST at 5 cm are work better than GP_PUK-, M5P-, and RF-based 0 0 Soil Temperature ( C) 0 Soil Temperature ( C) Predicted Soil Temperature ( C) Predicted Soil Temperature ( C) GEOLOGY, ECOLOGY, AND LANDSCAPES 213 Table 8. The results of diﬀerent modeling approaches for Tabriz at a depth of 5 cm. Tabriz Training Testing Models MAE (°C) RMSE (°C) CC MAE (°C) RMSE (°C) CC GP_PUK4 0.0009 0.0012 1.0000 2.7741 3.4742 0.9696 M5P4 2.3647 3.1031 0.9655 2.7106 3.4563 0.9692 RF4 1.0230 1.3335 0.9943 2.8455 3.4722 0.9728 MLP4 2.3703 3.0819 0.9664 2.5525 3.2626 0.9718 Bold values indicate the best performing model. MLP4 Tabriz M5P4 RF4 GP_PUK4 -5 -5 5 15 25 35 45 Actual Soil Temperature ( C) Figure 11. Agreement plot of actual and predicted values of soil temperature at 5 cm for Tabriz metrological station – using MLP, M5P, RF, and GP-based models model, testing period. Table 9. The results of diﬀerent modeling approaches for Ahar at a depth of 5 cm. Ahar Training Testing Models MAE (°C) RMSE (°C) CC MAE (°C) RMSE (°C) CC GP_PUK4 0.0019 0.0048 1.0000 5.4103 6.6202 0.8562 M5P4 4.7013 5.7717 0.8456 5.5907 6.9010 0.8448 RF4 1.6979 2.1423 0.9835 5.5501 6.8549 0.8436 MLP4 4.4826 5.5747 0.8570 5.1261 6.3332 0.8619 Bold values indicate the best performing model. MLP4 Ahar M5P4 RF4 GP_PUK4 -5 -5 5 15 25 35 45 Actual Soil Temperature ( C) Figure 12. Agreement plot of actual and predicted values of soil temperature at 5 cm for Ahar metrological station – using MLP, M5P, RF, and GP-based models model, testing period. Predicted Soil Temperature ( C) Predicted Soil Temperature ( C) 214 P. SIHAG ET AL. models. Model 4 (dependent variables of average air ORCID temperature, average sunny hours, average relative Parveen Sihag http://orcid.org/0000-0002-7761-0603 humidity, and average wind speed (Table 2)) per- Balraj Singh http://orcid.org/0000-0002-0381-4363 forms better than other combinations. In Figure 12, predicted values of the ST at 5 cm are closer to the line of the perfect agreement for Ahar station. References Öztürk et al. (2011) developed ANN-based models Atluri, V., Hung, C. C., & Coleman, T. L. (1999). An for forecasting the STs using eight diﬀerent variables, artiﬁcial neural network for classifying and predicting which are latitude, longitude, altitude, month number, soil moisture and temperature using Levenberg- solar radiation, sunshine duration, temperature, and Marquardt algorithm. In Proceedings IEEE year. It was concluded that the estimated values using Southeastcon'99. Technology on the Brink of 2000 (Cat. the ANN-based model lie very close to actual values for No.99CH36300) (pp. 10–13). Lexington, KY: IEEE. Bilgili, M. (2010). Prediction of soil temperature using all depths. Numerous researchers implemented latitude, regression and artiﬁcial neural network models. longitude, and altitude to estimate solar radiation Meteorology and Atmospheric Physics, 110(1–2), 59–70. (Citakoglu, 2015; Mohandes, Rehman, & Halawani, Breiman, L. (1996). Bagging predictors. Machine Learning, 1998; Siqueira, Tiba, & Fraidenraich, 2010). Tabari, 24(2), 123–140. Somee, and Zadeh (2011) used precipitation, relative Breiman, L. (2001). Random forests. Machine Learning, 45 humidity, air temperature, and solar radiation data of (1), 5–32. Citakoglu, H. (2015). Comparison of artiﬁcial intelligence MLR- and ANN-based models to estimate STs at var- techniques via empirical equations for prediction of ious soil depths and reported that ANN is more appro- solar radiation. Computers and Electronics in priate than MLR. In the present research, ST was Agriculture, 118,28–37. estimated with average air temperature, sunshine Citakoglu, H. (2017). Comparison of artiﬁcial intelli- hours, average relative humidity, and average wind gence techniques for prediction of soil temperatures in Turkey. Theoretical and Applied Climatology, 130 using MLP, GP, RF, and M5P models and results of (1–2), 545–556. this study indicate that MLP works better than other Firat, M., & Gungor, M. (2009). Generalized regression techniques. Kisi et al. (2015) used only Mersin station to neural networks and feed forward neural networks for estimate STs and found that air temperature is the most prediction of scour depth around bridge piers. Advances important variable in modeling ST for Mersin, Turkey. in Engineering Software, 40(8), 731–737. Ghorbani, M. A., Khatibi, R., Goel, A., FazeliFard, M. H., & Azani, A. (2016). Modeling river discharge time series using support vector machine and artiﬁcial neural net- 6. Conclusion works. Environmental Earth Sciences, 75(8), 685. Hillel, D. (1998). Environmental Soil Physics (pp. 771). San In this study, the Pearson VII kernel function-based GP, Diego,CA: Academic Press. M5P, RF, and MLP models were used to estimate the Kang, S., Kim, S., Oh, S., & Lee, D. (2000). Predicting DST. In this regard, diﬀerent input structures (Table 2) spatial and temporal patterns of soil temperature based on topography, surface cover, and air temperature. were used. Finally, the accuracy of each model and the Forest Ecology and Management, 136(1–3), 173–184. accuracy of each structure in ST estimation were inves- Kätterer, T., & Andrén, O. (2009). Predicting daily soil tigated. The statistical period of the data was 3 years. temperature proﬁles in arable soils in cold temperate Comparison of the performance of these models in regions from air temperature and leaf area index. estimating ST showed that MLP4 with much higher ActaAgriculturaeScandinavica Section B–Soil and Plant accuracy and less error than other models estimates Science, 59(1), 77–86. Kim, S., & Singh, V. P. (2014). Modeling daily soil tempera- the ST. MLP4 model includes the combination of aver- ture using data-driven models and spatial distribution. age air temperature, average sunny hours, average rela- Theoretical and Applied Climatology, 118(3), 465–479. tive humidity, and average wind speed which shows Kisi, O., Sanikhani, H., & Cobaner, M. (2017). Soil tem- superior performance than other input combination- perature modeling at diﬀerent depths using neuro-fuzzy, based models with all approaches. neural network, and genetic programming techniques. Theoretical and Applied Climatology, 129(3–4), 833–848. DuetothebetterperformanceoftheMLP Kisi, O., Tombul, M., & Kermani, M. Z. (2015). Modeling model for DST modeling in East Azerbaijan pro- soil temperatures at diﬀerent depths by using three dif- vince of Iran, it is recommended to use the MLP ferent neural computing techniques. Theoretical and model at other stations. Also, this model could be Applied Climatology, 121(1–2), 377–387. used in other time-dependent applications based Kuss, M., 2006. Gaussian process models for robust regres- on the availability of diﬀerent parameters. sion, classiﬁcation, and reinforcement learning (Doctoral dissertation, TechnischeUniversität). Liaw, A., & Wiener, M. (2002). Classiﬁcation and regres- sion by randomForest. R News, 2(3), 18–22. Disclosure statement McCulloch, W. S., & Pitts, W. (1943). A logical calculus of the ideas immanent in nervous activity. The Bulletin of No potential conﬂict of interest was reported by the Mathematical Biophysics, 5(4), 115–133. authors. GEOLOGY, ECOLOGY, AND LANDSCAPES 215 Mehdipour, V., & Memarianfard, M. (2017a). Application Shiri, J., & KişI, Ö. (2011). Comparison of genetic pro- of Support Vector Machine and Gene Expression gramming with neuro-fuzzy systems for predicting Programming on Tropospheric ozone Prognosticating short-term water table depth ﬂuctuations. Computers & for Tehran Metropolitan. Civil Engineering Journal, 3 Geosciences, 37(10), 1692–1701. (8), 557–567. Sihag, P. (2018). Prediction of unsaturated hydraulic con- Mehdipour, V., Memarianfard, M., & Homayounfar, F. ductivity using fuzzy logic and artiﬁcial neural network. (2017). Application of Gene Expression Programming Modeling Earth Systems and Environment, 4(1), 189–198. to water dissolved oxygen concentration prediction. Sihag, P., Singh, B., SepahVand, A., & Mehdipour, V. International Journal of Human Capital in Urban (2018b). Modeling the inﬁltration process with soft Management, 2(1), 39–48. computing techniques. ISH Journal of Hydraulic Mehdipour, V., & Memarianfardb, M. (2017b). Civil Engineering,1–15. doi:10.1080/09715010.2018.1464408 Engineering Journal. Civil Engineering, 3(8), 557–567. Sihag, P., Tiwari, N. K., & Ranjan, S. (2017). Prediction of Mehdizadeh, S., Behmanesh, J., & Khalili, K. (2017). unsaturated hydraulic conductivity using adaptive Evaluating the performance of artiﬁcial intelligence neuro-fuzzy inference system (ANFIS). ISH Journal of methods for estimation of monthly mean soil tempera- Hydraulic Engineering,1–11. doi:10.1080/ ture without using meteorological data. Environmental 09715010.2017.1381861 Earth Sciences, 76(8), 325. Sihag, P.,Tiwari, N.K.,& Ranjan,S. (2018a). Support Miles, A., & Brown, M. (2007). Teaching Organic Farming vector regression-based modeling of cumulative inﬁl- and Gardening: Resources for Instructors. Santa Cruz: tration of sandy soil. ISH Journal of Hydraulic University of California Farm and Garden. Engineering,1–7. doi:10.1080/09715010.2018.1439776 Mohandes, M., Rehman, S., & Halawani, T. O. (1998). Singh, B., Sihag, P., & Singh, K. (2017). Modelling of Estimation of global solar radiation using artiﬁcial impact of water quality on inﬁltration rate of soil by neural networks. Renewable Energy, 14(1–4), 179–184. random forest regression. Modeling Earth Systems and Nahvi, B., Habibi, J., Mohammadi, K., Shamshirband, S., & Environment, 3(3), 999–1004. Al Razgan, O. S. (2016). Using self-adaptive evolutionary Singh, R. B. (2000). Environmental consequences of agri- algorithm to improve the performance of an extreme cultural development: A case study from the Green learning machine for estimating soil temperature. Revolution state of Haryana, India. Agriculture, Computers and Electronics in Agriculture, 124, 150–160. Ecosystems & Environment, 82(1–3), 97–103. Öztürk, E., Atsan, E., Polat, T., & Kara, K. (2011). Variation Siqueira,A.N., Tiba,C.,&Fraidenraich,N.(2010). in heavy metal concentrations of potato Generation of daily solar irradiation by means of arti- (Solanumtuberosum L.) cultivars. Journal of Animal ﬁcial neural net works. Renewable Energy, 35(11), Plant Science, 21(2), 235–239. 2406–2414. Parsaie, A., & Haghiabi, A. H. (2015). Predicting the long- Sun, B., Zhang, L., Yang, L., Zhang, F., Norse, D., & Zhu, Z. itudinal dispersion coeﬃcient by radial basis function (2012). Agricultural non-point source pollution in neural network. Modeling Earth Systems and China: Causes and mitigation measures. Ambio, 41(4), Environment, 1(4), 34. 370–379. Parsaie, A., Yonesi, H. A., & Najaﬁan, S. (2015). Predictive Tabari, H., Somee, B. S., & Zadeh, M. R. (2011). Testing for modeling of discharge in compound open channel by long-term trends in climatic variables in Iran. support vector machine technique. Modeling Earth Atmospheric Research, 100(1), 132–140. Systems and Environment, 1(1–2), 1. Talaee, P. H. (2014). Daily soil temperature modeling using Quinlan, J. R., 1992, November. Learning with continuous neuro-fuzzy approach. Theoretical and Applied classes. In 5th Australian joint conference on artiﬁcial intel- Climatology, 118(3), 481–489. ligence (Vol.92, pp. 343–348). Singapore: World Scientiﬁc. Tiwari, N. K., Sihag, P., Kumar, S., & Ranjan, S. (2018). Rasmussen, C. E., & Williams, C. K. (2006). Gaussian process Prediction of trapping eﬃciency of vortex tube ejector. for machine learning. Cambridge, England: MIT Press. ISH Journal of Hydraulic Engineering,1–9. doi:10.1080/ Schaetzl, R. J., Knapp, B. D., & Isard, S. A. (2005). 09715010.2018.1441752 Modeling soil temperatures and the mesic-frigid bound- Wu, W., Tang, X. P., Guo, N. J., Yang, C., Liu, H. B., & ary in the central Great Lakes region, 1951–2000. Soil Shang, Y. F. (2013). Spatiotemporal modeling of Science Society of America Journal, 69(6), 2033–2040. monthly soil temperature using artiﬁcial neural net- Schalkoﬀ,R.J. (1997). Artiﬁcial neural networks (Vol. 1). works. Theoretical and Applied Climatology, 113(3–4), New York: McGraw-Hill. 481–494.
Geology Ecology and Landscapes – Taylor & Francis
Published: Jul 2, 2020
Keywords: Daily soil temperature; Multilayer perceptron model; Gaussian Process; Random Forest; M5P model
You can share this free article with as many people as you like with the url below! We hope you enjoy this feature!
Read and print from thousands of top scholarly journals.
Already have an account? Log in
Bookmark this article. You can see your Bookmarks on your DeepDyve Library.
To save an article, log in first, or sign up for a DeepDyve account if you don’t already have one.
Copy and paste the desired citation format or use the link below to download a file formatted for EndNote
Access the full text.
Sign up today, get DeepDyve free for 14 days.
All DeepDyve websites use cookies to improve your online experience. They were placed on your computer when you launched this website. You can change your cookie settings through your browser.