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

Learn More →

Methodological evaluation of vegetation indexes in land use and land cover (LULC) classification

Methodological evaluation of vegetation indexes in land use and land cover (LULC) classification GEOLOGY, ECOLOGY, AND LANDSCAPES 2020, VOL. 4, NO. 2, 159–169 INWASCON https://doi.org/10.1080/24749508.2019.1608409 RESEARCH ARTICLE Methodological evaluation of vegetation indexes in land use and land cover (LULC) classification a a a a Vanessa Sousa da Silva , Gabriela Salami , Marília Isabelle Oliveira da Silva , Emanuel Araújo Silva , a b José Jorge Monteiro Junior and Elisiane Alba a b Department of Forest engineering, University Federal Rural of Pernambuco, Recife, Brazil; Forest engineering, University Federal of Santa Maria, University City, Brazil ABSTRACT ARTICLE HISTORY Received 16 October 2018 Vegetation indices are intended to emphasize the vegetation spectral behavior in relation to Accepted 13 April 2019 the soil and other terrestrial surface targets. The objective of this study was to evaluate the vegetation cover types present in the municipality of Campo Belo do Sul, Brazil, using data KEYWORDS from five vegetation indices obtained through satellite images. In order to do so, calculations Remote sensing; image of the Normalized Difference Vegetation Index (NDVI), Soil-adjusted Vegetation Index (SAVI), processing; Landscape Leaf Area Index (LAI), Enhanced Vegetation Index (EVI) and Normalized Difference Water Index (NDWI) were performed using Quantum Gis software. The generated maps allowed the detection of the different vegetation cover classes, thus the results indicated that there is no specific vegetation index that best represents all the evaluated classes in the study, however, NDVI, EVI, and SAVI had good adjustments in the majority of the thematic classes. Introduction Barros, Faria, & Adami, 2010; Ramirez, Zullo Junior, Assad, & Pinto, 2006; Silva & Costa, 2015). In countries of great extension such as Brazil, earth The reflectance of surface targets combinations at observation through satellite systems based on medium- two or more wavelengths, especially in the visible and resolution multispectral scanners is one of the most infrared region, generates dimensionless radiometric efficient and economical ways to obtain relevant infor- measurements called vegetation indices (VIs). The mation about terrestrial natural resources and the vege- objective of the VIs is to highlight a particular vegeta- tation conditions (Mallmann, Prado, & Pereira Filho, tion characteristic such as leaf area index (LAI), the 2015). Remote sensing can generate useful spectral reflec- percentage of green cover, chlorophyll content, green tance data that provides rapid means for monitoring and biomass and absorbed photosynthetically active radia- managing natural resources. In addition, through visual tion (Jensen, 2009). Those are often used in ecological and digital images processing it is possible to extract research, ecosystems modeling, biophysical parameters biophysical information from the vegetation cover, of vegetation estimation, as well as for monitoring the which is considered crucial to elucidate processes related terrestrial surface (Robinson et al., 2017). VIs are math- to forest distribution, human activities, biodiversity con- ematical models used in studies conducted since the servation, as well as socioeconomic processes (Mancino, 1960s developed to evaluate, monitor and analyze the Nolè, Ripullone, & Ferrara, 2014). vegetation cover and relating the spectral signature and The supervised classification of orbital images is measurable parameters in the field both quantitative a usual method for mapping and evaluation studies of and qualitatively (Mallmann et al., 2015). land use changes and occupation (Antunes, There are several VIs, with their specificapplicabil- Mercante, Esquerdo, Lamparelli, & Rocha, 2012; ity, used in the different representative phytophysiog- Canetti, Garrastazu, Mattos, Braz, & Pellico Netto, nomies of the world biomes for vegetation mapping and 2018; Churches, Wampler, Sun, & Smith, 2014; monitoring on the terrestrial surface. The vegetation Ghebrezgabher, Yang, Yang, Wang, & Khan, 2016; indices allow to obtain information from the spectral Liesenberg & Gloaguen, 2013; Oliveira, Fernandes response of the targets, and thus, to diagnose different Filho, Soares, & Souza, 2013). Among the supervised biophysical parameters such as leaf area index, biomass, classification methods, the algorithm of maximum the percentage of land cover, photosynthetic activity likelihood (Maxver) is one of the most applied meth- and productivity (Ponzoni, Shimabukuro, & Kuplich, odologies to characterization and monitoring studies 2012). To Xue and Su (2017) these indices are simple of forested and agricultural areas (Moreira, Rudorff, and effective algorithms to evaluate the vigor and CONTACT Emanuel Araújo Silva emanuelmadster@gmail.com Department of Forest engineering, University Federal Rural of Pernambuco, Recife, Brazil © 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. 160 V. S. DA SILVA ET AL. dynamics of terrestrial vegetation. It is emphasized that suitable for annual crops with some obstacles from such indices have particularities regarding their sensi- the undulating and gently undulating soils, more tivity towards targets since this relation is influenced by suitable for crops; being predominant the latosols factors inherent to the target. and tropohumult in the less steep areas and lithosols This study aimed to evaluate the efficiency of differ- and neosols in more rugged areas (Embrapa, 2004). ent vegetation indices as NDVI, SAVI, LAI, EVI, and The study area presents in rural activity its main NDWI; in the classification of land use and occupation, socioeconomic activity; most of the properties are in order to identify the index that best represents the composed of natural pastures, temporary/permanent current coverage and more closely resembles the classi- crops and planted forests. According to Silva (2015), fication performed by the MaxVer algorithm. the municipality of Campo Belo do Sul is known for having its economy based on agrosilvopastoral sys- tems crops, with emphasis on forestry from the forest Materials and method farm Gateados with the largest reforested area in the south of the country. The studied area is located in the municipality of The image processed in this work was obtained Campo Belo do Sul, southwest region of the state of from the Landsat 8 Operational Land Imager (OLI) Santa Catarina, Brazil (Figure 1), between the coordi- sensor made available by the USGS, with a passage on nates 27º53’57“ south latitude and 50º45’39” west 11 March 2017, orbit/point 221/79, with a spatial longitude, covering an area of approximately resolution of 30 m for visible bands, near and mid- 1.027,65 km with a population of 7.483 inhabitants, infrared and 15 m for the panchromatic band. The 16 with population density of 7,28 inhab./km ; with an days temporal resolution enables continuous moni- equivalence between urban and rural population toring projects (Explorer, 2016). The geoprocessing of (IBGE, 2011). the cartographic data was performed by the software According to the Köppen classification (1948), the Quantum Gis 2.18.6, with the help of the Semi- climatic type of the Santa Catarina Plateau region is automatic Classification Plugin (SCPA) tool, which a transition between Cfa (wet mesothermic, with no allowed the processing of information from the defined dry season, hot summers, with rare occur- Landsat 8 image. rence of frost in winter) and Cfb (mesothermic wet, For the elaboration of the image chart and analysis with no defined dry season, fresh summers, with of the land cover, the pseudo-color composition was severe and frequent frosts occurring in winter), with used, with the bands stacking 4–5–6 (RGB). The temperatures varying from 13°C to 25° C and rain- conversion of digital levels to surface reflectance was falls distributed throughout the year, totaling an aver- performed through the SCP Pre-Processing options, age of 1841 mm annually (Oliveira, Bertol, Barbosa, which uses the DOS 1 method to correct atmospheric Campos, & Mecabô Junior, 2015). As for the soil, effects. The atmospheric correction is done in order they are shallow and stony, with low fertility and little Figure 1. Location of the municipality of Campo Belo do Sul-SC, BR. GEOLOGY, ECOLOGY, AND LANDSCAPES 161 to eliminate the imperfections that may damage the the occurrences of each class with reference points, information (Maranhão, Pereira, Costa, & Anjos, generating a contingency matrix (matrix of errors). 2017), thus enabling physical surface reflectance The reference points were randomly collected and values to be obtained without the effects of atmo- labeled by visual interpretation. The contingency spheric interference. Regarding the vector data, the matrix allows validating the supervised classification cartographic bases pertinent to the municipal bound- by estimating the overall accuracy of the mapping, as ary, obtained from the Brazilian Geological Service well as the Kappa coefficient (Cohen, 1960), which (CPRM, 2017) were used. quantifies the agreement between classification and The vegetation indices of this study were selected reference data, ranging from zero to one, being more based on those usually applied in studies of this accurate those data that have the value closer to one, nature (Xue & Su, 2017), such as the Normalized while it will have a doubtful veracity the closer the Difference Vegetation Index (NDVI), Soil-Adjusted value is to zero (Silva, 2011). Vegetation Index (SAVI) Leaf area index (LAI), Through the error matrix it was also possible to Enhanced Vegetation Index (EVI) and Normalized extract information regarding the producer accuracy Difference Water Index (NDWI) (Table 1). (relative to the omission error, calculated from the Using the Semi-automatic Classification Plugin reference data, indicates the probability of a reference (SCP) plugin in Quantum Gis 2.18.6 software, the data – field truth, to be correctly classified) and the algebraic operations were performed on the compo- user accuracy (relative to the commission error, cal- nents of the indices mentioned above, from the con- culated from the classification data, indicates the verted bands to apparent reflectance values, and then probability that a classified pixel will effectively repre- the classified images by each index were generated sent the category in the field) (Furtado, 2013) and and then compared. with this information it was possible to evaluate if the The mapping of land use and the cover was classification was effective. obtained by a supervised classification, using the A 100 random samples were collected from each maximum likelihood classifier algorithm (MaxVer), land use and cover class and later extracted classifica- which according to Meneses and Sano (2012), con- tion and indices using the Point Sample Tools. The siders the distance weighting between the means of data were compiled and statistically analyzed in the the digital levels based on statistical parameters. For software R Statistical 3.3.1 through the Rcmdr pack- the study area, the following thematic classes were age and then the contribution of each vegetation proposed: Water bodies (water slide, dams, rivers, index in the identification of the proposed thematic lagoons, artificial lakes); Native forests (areas occu- classes was evaluated. pied by different native forest formations, including Correlation analysis was performed between the vege- permanent preservation areas); Planted forests (estab- tation indices and the land use and occupation classes, lished monocultures occupied with Eucalyptus and determining the linear dependence degree between Pinus genus plantations); built areas (includes urban them. A total of five independent variables were analyzed area, established rural areas, roads and other con- using the original indices. For the variables selection, it structions and infrastructures); agricultural areas was used the Forward method, and the indices that (cultivated areas with undefined crop types); under contributed the most to the identification of land use fallow areas (newly harvested areas, from farming or and occupation classes were used, as it allows to examine forestry, and areas prepared for the next planting); the contribution of each independent index to the regres- and native fields (areas where there is no forest pre- sion model. The models were evaluated based on the sence and characteristic vegetation is a natural pas- adjusted coefficient of determination (R aj), standard ture, planted pastures were also included). error of the estimate (Syx) and coefficient of variation After obtaining the map of land use and cover, in (CV%), where the fitting level of the selected models for order to estimate the accuracy of the classification each class of land use and occupation was determined by data, we performed statistical analyzes that relate the distribution of the residuals (Alba et al., 2017)and by the sum of the statistical scores proposed by Thiersch (1997). It was assigned the lowest weight (one) for the best statistical results of each evaluated index, the best Table 1. Vegetation Indices analyzed in the present study. model was designated by the sum of the scores, values Vegetation index Formula Autors from one to N, where the lowest sum of the scores ρnir ρred NDVI Rouse, Haas, Schell, and NDVI ¼ indicates the selection of the best equation. ρnirþ ρred Deering (1974) ρnir ρred SAVI Huete (1988) SAVI ¼ðÞ 1 þ L ρnirþ ρredþL 0;69SAVI Ln LAI ðÞ Allen, Tasumi, and 0;59 LAI ¼ Results 0;91 Trezza (2002) ρnir ρred EVI Justice et al. (1998) EVI ¼ 2; 5 ðÞ ρnirþ6ρred7;5ρblueþ1 Through the land use and cover mapping, it was ρnir ρswir NDWI Gao (1996) NDWI ¼ ρnirþ ρswir possible to identify seven different thematic classes 162 V. S. DA SILVA ET AL. in the study area (Figure 2). The visual analysis Table 2. Area occupied by thematic classes expressed in hectares and percentage. allowed to identify the predominance of native forests Classes Area (ha) Area (%) in the southern region, associated to watercourses. Native Forests 39,059.37 38.01 The mapping quantitative analysis (Table 2) Native Fields 34,950.87 34.01 expresses the area occupied by the thematic classes, Monocultures 11,945.34 11.63 Agricultural areas 7,510.41 7.31 which corresponds to the original classification. It Under fallow areas 6,181.2 6.01 should be noted that the native forests, native fields, Built areas 1,786.5 1.74 Water 1,314.09 1.28 and established monocultures are the classes that Total area 102,747.78 100 comprise the highest percentages, occupying the lar- gest areas in the municipality. On the other hand, Table 3. Percentage and producer accuracy. built areas and watercourses are less expressive in Classes Producer Accuracy User Accuracy Campo Belo do Sul. Native Forests 89.46 97,48 The Kappa index, calculated by the contingency Native Fields 94.73 92,24 matrix of the thematic map and the reference points, Monocultures 93.52 77,76 Agricultural areas 98.95 97,59 presented a value of 0.88. Thus, the classification was Under fallow areas 89.87 83,14 considered as very good according to the evaluation Built areas 74.60 81,31 criteria of Galparsoro and Fernández (1999), as well Water 99.89 98,43 Global Accuracy 91.59 as its overall accuracy, which expressed a value of approximately 92% (Table 3). The vegetation indices resulting from the maps with each other, except for the native forests class, algebra are shown in Figure 3, demonstrating the which showed little correlation in most of the vari- variations occurring in the classification after applica- ables, showing a high correlation only in relation to tion of the indices when compared to the initial the LAI to the SAVI. classification made by the MaxVer classifier algo- It was also observed that for the native forest, the rithm (Figure 2). relationship between NDVI and NDWI expressed It is inferred that there are variations in the classi- a correlation coefficient of 0.68, representing one of fication of the land cover according to each index, the highest correlations for this class. It should be which are corroborated by observing the different noted that EVI and SAVI were strongly correlated class intervals obtained for each vegetation index when the native field, planted forest, agriculture, and tested in the study (Table 4). urbanization were evaluated, showing the highest The Pearson correlation analysis performed in the correlation coefficients. However, presenting vegetation indices obtained in each use class is a higher correlation does not mean that both indices demonstrated in Figure 4, in which 100% correlated are adequate for the classes, as can be seen in Table 5. variables expressed the value one. It is possible to From the linear regression models, based on the analyze that for the different land uses and cover, scores of each class of land use (Table 5), it can be the vegetation indices showed a high correlation inferred that for native forests the most adequate Figure 2. Map of land use and classification of the municipality of Campo Belo do Sul-SC, Brazil. GEOLOGY, ECOLOGY, AND LANDSCAPES 163 Figure 3. Land use and cover mapping based on vegetation indices NDVI (a), EVI (b), NDWI (c), LAI (d) e SAVI (e). Discussion Table 4. Class intervals for each vegetation index analyzed for the municipality of Campo Belo do Sul, Santa Catarina, Brazil. According to the initial analysis of land use and cover Class Interval of the municipality of Campo Belo do Sul, it is note- Vegetation Index Minimum Value Maximum Value worthy that native compositions of the region, repre- EVI −0.235 −0.054 NDWI −0.246 0.143 sented by forests and native fields, occupy LAI 0.745 4.09 approximately 70% of the municipality. This result SAVI 0.357 0.964 NDVI −0.603 0.936 corroborates with the percentage of the classes related to the anthropic action being less expressive, allowing to infer that the municipality environmental degrada- tion is not an expressive practice. The native fields are indices were NDVI and EVI; for planted forests, present in 34% of the area of the municipality, SAVI, followed by EVI and NDVI; for the class of a percentage below only the native vegetation, agricultural areas were SAVI and EVI, the same which can be explained due to the extensive livestock indices that were more adequate for the native fields; practice in this territory, in which the native fields for the fallow areas class the best adjustments were (natural pastoral ecosystem) offer the support for the with the NDWI and NDVI indices; NDWI was also development of the activity (Kaibara, 2014). the best performing index to determine constructed It is also observed that approximately 50% of the areas; and, for the waterbodies, NDVI and SAVI were municipal territory is covered by shrub-tree vegeta- more indicated. Making it possible to infer that there tion, adding native forests (38%) and established was not an index that excelled in all classes. 164 V. S. DA SILVA ET AL. Figure 4. Correlation of vegetation indices for different land uses and cover: Native forests (a), Native fields (b), Monocultures (c), Agricultural areas (d), Under fallow areas (e) and Built areas (g). monocultures (12%) that are composed of Eucalyptus for the development of agriculture and livestock pro- and Pinus tree individuals. However, in 8% of the duction in the region. Agricultural areas (7%) and municipality, there is no vegetation cover, analyzing watercourses (1%) were also present in the identified areas under fallow and also built areas; the areas classes, being less expressive. under fallow have its soil exposed, which are The variation of accuracy among the classes eval- a result of the intense soil exploitation carried out uated, can be associated with common remote GEOLOGY, ECOLOGY, AND LANDSCAPES 165 Table 5. Linear regression models resulting from the Forward selection method for estimating land use and land cover. Regression Model F R aj Syx CV(%) Scores Native Forests Y= −0,3532 + 0,9790 (EVI) 56.88*** 0.3608 0.0008 1.55 8 Y = 0,5543–0,4937 (NDVI) 528.3*** 0.8419 0.0023 2.78 6 Y = 0,2046–0,1567 (NDWI) 311.3*** 0.7581 0.0070 16.58 10 n.s 2.32 0.01316 0.0048 2.65 - n.s 2.342 0.01338 0.0005 0.88 - Monocultures Y = 0,00937 + 0,23398(EVI) 417.1*** 0.8078 0.0055 10.91 10 Y= −0,1086 + 0,2782(NDVI) 17.37*** 0.1419 0.0020 2.36 10 Y = 0,09230 + 0,06745(NDWI) 5.715* 0.04546 0.0050 9.44 16 Y = 0,07622 + 0,02715(IAF) 279.3*** 0.7376 0.0456 23.88 16 Y= −0,04891 + 0,30,626(SAVI) 328.2*** 0.7701 0.0041 7.14 8 Agricultural Areas Y = 0,02528 + 0,29251(EVI) 911.1*** 0.9019 0.0029 3.14 10 Y= −0,4443 + 0,8072(NDVI) 129.3*** 0.5644 0.0008 0.91 12 Y = 0,008275 + 0,695017(NDWI) 213.6*** 0.6823 0.0011 2.57 13 Y = 0,35177–0,03338(IAF) 229.3*** 0.6975 0.0225 13.79 16 Y= −0,1577 + 0,5512(SAVI) 624.9*** 0.863 0.0015 1.83 9 Under Fallow Areas n.s 0.1003 0.009171 0.0082 25.16 - Y = 0,22976–0,08759(NDVI) 10.66** 0.08893 0.0114 20.43 6 Y = 0,1926–0,1492(NDWI) 54.31*** 0.35 0.0128 162.27 6 n.s 0.2263 0.007876 0.0306 38.69 - n.s 0.4321 0.005769 0.0085 21.65 - Built Areas n.s 0.001082 0.01019 0.0096 38.66 - n.s 2.035 0.01035 0.0135 35.03 - Y = 0,20189–0,05121(NDWI) 5.595* 0.04435 0.0090 233.07 4 n.s 0.2402 0.007734 0.0335 72.95 - n.s 0.4388 0.005701 0.0103 36.07 - Native Fields Y = 0,07465 + 0,22079(EVI) 65.92*** 0.396 0.0064 12.21 6 n.s 0.06224 0.009563 0.0045 5.89 - n.s 1.325 0.003276 0.0063 24.94 - Y = 0,1537 + 0,0191(IAF) 38.5*** 0.2747 0.0617 32.57 12 Y = 0,05102 + 0,24281(SAVI) 39.95*** 0.2823 0.0049 8.60 6 Water n.s 0.4378 0.005711 0.0002 2.38 - Y = 0,05341 + 0,06213(NDVI) 524.8*** 0.841 0.0023 5.32 9 Y = 0,03615–0,08531(NDWI) 162.6*** 0.6202 0.0015 12.44 11 Y = 0,07703 0,15110(IAF) 15.22*** 0.1256 0.0004 1.14 11 Y = 0,04929 + 0,20745(SAVI) 15.25*** 0.1258 0.0003 2.50 9 n.s. Signif, codes: 0 ‘***‘; 0,001 ‘**‘; 0,01 ‘*‘; not significant ‘ ‘ sensing limitations, among them the distinction of When comparing the classification of land use by targets with partially similar spectral responses, such MaxVer (initial classification) with the vegetation as: native and planted forests, native fields, agricul- indices, it was noticed variations in the classification tural areas and areas under fallow. Another possible of the municipality. Among vegetation indices, this confounding factor in the automatic classification is difference is also expected, since they are based on the characteristic relief, because the sun-earth-sensor different algorithms to classify the images, each index alignment prevents the electromagnetic radiation having its specificities and different levels of detail. from being reflected by the targets located on steep The correlogram within each thematic class slopes in full shape, detecting only shadows. This showed that some indices have a high correlation factor interferes in the spectral response of targets between them for most classes. Macedo, Sousa, located on slopes and consequently in the quality of Gonçalves, Silva, and Rodrigues (2017), seeking to the automatic classification (Caixeta, Edmundo, develop allometric functions to estimate total and Rodrigues, Moreira, & Medeiros, 2012). commercial volume, using as a dependent variable The understanding of the land use and occupation a vegetation index, infers that the high correlations forms, with modeling purposes and representing obtained for vegetation indices are justified mainly large areas with good levels of detail, can be consid- due to their compositions, which essentially use the erably improved using geoprocessing tools. In addi- spectral bands in the region of the electromagnetic tion to reducing fieldwork, the extraction of spectrum related to red and infrared. Thus, such information in medium-resolution digital images, behaviors observed in the present study are expected, such as the Landsat 8 satellite, allows larger areas to since in order to determine some indices there is be studied in a shorter time. dependence on others. 166 V. S. DA SILVA ET AL. From the regression analysis, it was possible to noting that both NDWI and NDVI are prominent in notice that in each thematic class the indices are identifying vegetation and water coverages consider- presented in a different way, some being more ade- ing their individual restrictions. quate than others in the determination of the differ- With another approach, Choung and Jo (2016) ent coverages. used NDWI to monitor changes in water resources In determining the native forests class, NDVI was in South Korea using Landsat multitemporal images; the best index, also suitable for determining waterbo- their results showed significant differences in water- dies (best fit), areas under fallow (second best fit) and body sizes throughout the study years. Sarp and planted forests (third best fit). NDVI is a relevant index Ozcelik (2018) also used NDWI to assess changes in for areas of medium to high vegetation density, since it water resources in southwestern Turkey, and the is less susceptible to the soil and to the effects of the results effectively showed the detection of change in atmosphere. According to Ponzoni et al. (2012), this the water surface between specified time intervals. On index has been used to detect the effects of seasonality, the other hand, aiming to determine indicators for phenological stage of vegetation, duration of the growth assessing the vulnerability of forests and fires in period, green peak, physiological changes of leaves, Indonesia, Nurdiana and Risdiyanto (2015) used senescence periods and other important vegetation NDVI and NDWI as key parameters for the identifi- related situations. However, it is not suitable for areas cation of fire outbreaks, where both indices were able with low vegetation coverage (Karimi et al., 2018). The to provide the highest contribution to the vulnerabil- index values range from −1to +1and isbased on the ity level. high reflection of the healthy plants in the wavelengths Efforts around the world have been observed to in the infrared band and its low reflection in the pre- justify the use of this index (associated or not to sence of the red band of the electromagnetic spectrum others) to identify dry areas. Gu, Brown, Verdin, (Hesketh, Sánchez-Azofeifa, & Azofeifa, 2014), there- and Wardlow (2007) evaluated the results of the use fore, healthy forest cover usually has higher NDVI of satellite-derived indices (NDVI and NDWI) to values. monitor vegetation drought through soil moisture This is the most commonly used vegetation index, observations in the United States. They found that which minimizes topographic effects by producing there is a strong relationship between the vegetation a linear scale of measurement, in which the closer to indices with the heterogeneity of the land cover, soil 1 the greater the vegetation cover density and, in its type and humidity, and for areas of homogeneous turn, the 0 would be the approximate value for the forest cover, both were sensitive to changes in soil absence of vegetation, that is, represents non-vegetated moisture and are strongly related to vegetation surfaces (Rosendo, 2005). In this regard, it is justified drought conditions; suggesting that both indices are that this index had good results to determine water- appropriate to monitor water stress in vegetation. bodies and areas under fallow, being an index that Szabó, Gácsi, and Balázs (2016) investigated ranges presents comprehensive values of class intervals. of three spectral indices for land cover types by com- In the determination of areas under fallow and paring the spectral ratios by land cover types and areas with buildings, as well as waterbodies (after assessed their efficiency in discriminating land cover NDVI), the best fit index was NDWI. This is classes. As well as the multiple functions that the amodification of the NDVI, that allows to highlight authors suggest for NDWI, the results of this study water features and minimize the rest of the targets. foster the idea that their use can be indicated to The NDWI is highly correlated with the water content evaluate various land coverages. in the vegetation cover, making it possible to measure It was from the use of the abovementioned indices biomass changes and to evaluate vegetation water that the need for an index that considered soil stress, through mathematical operations using near- response, which could be dominant on the vegetation infrared and medium infrared bands (Jensen, 2009). response, was found, depending on the coverage per- However, some authors have used NDWI for centage. Thus, in order to mitigate this soil effect, numerous purposes. When investigating phenological SAVI (Huete, 1988) was created, which is based on metrics in dry and dormancy periods in semi-arid the principle that the vegetation curve tends to pastures, Ding, Liu, Huan, Li, and Zou (2017) have approach the soil curve for low vegetation densities, found results that suggest that phenological studies passing through a mixture of spectral responses to using NDWI can expand the understanding of the average densities and almost no influence of the soil terrestrial surface phenology; furthermore, they sug- to high vegetation densities (Sousa & Ponzoni, 1998). gest that the combination of this index with climatic In this study, SAVI was the best fit index to deter- variability could contribute to the study of ecosystem mine planted forests, agricultural areas, native fields, processes in semi-arid pastures. Ahmed and Akter and waterbodies. According to Huete (1988), the (2017) studied changes in land use and cover follow- SAVI adjusted to different soil conditions that may ing regular flooding in coastal areas of Bangladesh, exert considerable influence on the canopy, thus in GEOLOGY, ECOLOGY, AND LANDSCAPES 167 areas where there are considerable variations in the Disclosure statement soil brightness resulting from differences in moisture, No potential conflict of interest was reported by the variations in roughness, shade or differences in authors. organic matter, there are soil-induced influences on vegetation index values. For this reason, this index reduces the soil effects and it is probable that in this Funding study, its good adjustment in the different coverages This work was supported by the Conselho Nacional de was provoked precisely because these areas have soils Desenvolvimento Científico e Tecnológico [1]; where the reflectance is favored. Coordenação de Aperfeiçoamento de Pessoal de Nível Corroborating with the results obtained in the Superior [3];Fundação de Amparo à Ciência e Tecnologia present study, Liaqata et al. (2017) found that SAVI do Estado de Pernambuco [1]. was the index that best adjusted for estimated agri- cultural production in irrigated areas in Pakistan; González-Dugo and Mateos (2008) observed that References SAVI also excels when used in irrigated cotton and Ahmed, K. R., & Akter, S. (2017). Analysis of landcover beet crops in southern Spain. For planted forests change in southwest Bengal delta due to floods by areas, Alba et al. (2017) demonstrated that SAVI NDVI, NDWI and K-means cluster with landsat was the index that had the best correlation with the multi-spectral surface reflectance satellite data. Remote volume estimation for a Pinus elliottii forest. In Sensing Applications Society and Environment, 8(1), 168–181. a study developed by Cassol (2013), Maciel (2002) Alba, E., Mello, E., Marchesan, J., Silva, E. A., and Bernardes (1998), the SAVI index also stood out Tramontina, J., & Pereira, R. S. (2017). Spectral charac- as one of the spectral variables with the greatest terization of forest plantations with Landsat 8/OLI relation to the forest biomass in a Mixed images for forest planning and management. Pesquisa Ombrophilous Forest, which has a high population agropecuária brasileira, 52(11), 1072–1079. density. These authors reinforce that soil brightness Allen, R., Tasumi, M., & Trezza, R. (2002). SEBAL (Surface Energy Balance Algorithms for Land) – advanced training influences the spectral response, even in closed cano- and users manual – Idaho implementation, version 1.0. pies and with high individual density.b Antunes, J. F. G., Mercante, E., Esquerdo, J. C. D. M., The EVI was the second-best index for the four land Lamparelli, R. A. C., & Rocha, J. V. (2012). Estimativa uses, being the most sensitive to the canopy structure de área de soja por classificação de imagens normalizada (Huete et al., 2002), although it is more influenced by pela matriz de erros. Pesquisa Agropecuária brasileira, 47 (9), 1288–1294. the solar zenith angle (Galvão et al., 2011) under cer- Bernardes, S. (1998). Índices de vegetação e valores de tain conditions. This index has higher sensitivity in proporção na caracterização de floresta tropical regions with high biomass production and reduces primária e estádios sucessionais na área de influência da the atmospheric and soil influences (Jiang, Huete, Floresta Nacional do Tapajós-Estado do Pará. Didan, & Miura, 2008). Risso et al. (2012)evaluated Dissertação (Mestrado em Sensoriamento Remoto): the performance of EVI and NDVI to discriminate Instituto Nacional de Pesquisas Espaciais. 94f. Caixeta, P. R. C., Edmundo, I. S. B., Rodrigues, A. F. S., soybean areas from sugarcane, pasture, forested and Moreira, G. F., & Medeiros, N. G. (2012). Processamento Cerrado areas, they found that EVI excelled in the de Imagens de Alta Resolução RapidEye para image classification, quite possibly due to this charac- Mapeamento do Meio Físico. IV Simpósio Brasileiro de teristic of a greater sensitivity to higher biomass values. Ciências Geodésicas e Tecnologias da Geoinformação (pp.001–008). Recife: PE-BR. Canetti, A., Garrastazu, M. C., Mattos, P. P., Braz, E. M., & Pellico Netto, S. (2018). Understanding multi-temporal Conclusion urban forest cover using high resolution images. Urban There is no specific vegetation index that best repre- Forestry and Urban Greening, 29(1), 106–112. Cassol, H. L. G. (2013). Estimativa de biomassa e estoque de sents all the classes evaluated in the study; however, carbono em um fragmento de floresta ombrófila mista NDVI, EVI, and SAVI had good adjustments in most com uso de dados ópticos de sensores remotos of the thematic classes. The choice of the best vegeta- (Dissertação (Mestrado em Sensoriamento Remoto)). tion index for land use and occupation classification Universidade Federal do Rio Grande do Sul, 143. depends, particularly on the predominant soil use. Choung, Y. J., & Jo, M. H. (2016). Monitoring water resource changes using multi-temporal landsat imagery In general, it is possible to suggest that the infor- taken in changnyeong, South Korea. Procedia mation spectral mixture and the determination of Engineering, 154(1), 348–352. several thematic classes favored the non-definition Churches, C. E., Wampler, P. J., Sun, W., & Smith, A. J. of the best vegetation index, and thus it can be (2014). Evaluation of forest cover estimates for Haiti inferred that depending on the objective and the using supervised classification of Landsat data. type of predominant use, more adequate indexes International Journal of Applied Earth Observation and Geoinformation, 30(1), 203–216. can be indicated. 168 V. S. DA SILVA ET AL. Cohen, J. (1960). A coefficient of agreement for nominal Land remote sensing for global change research. IEEE scales. Educational and Psychological Measurement, 20 Transactions on Geosciences and Remote Sensing, 36(4), (1), 37–46. 1228–1249. CPRM. (2017) Serviço Geológico Brasileiro. Geobank. Retrieved Kaibara, M. A. (2014). Rendimento animal em uma pasta- from http://geobank.sa.cprm.gov.br/ gem nativa melhorada no planalto sul catarinense (pp. Ding, H., Liu, X., Huan, F., Li, Y., & Zou, X. (2017). Onset 60). Monografia, Florianópolis (SC): UFSC – of drying and dormancy in relation to water dynamics of Universidade Federal de Santa Catarina. semi-arid grasslands from MODIS NDWI. Agricultural Karimi, A., Abdollahi, S., Balajadeh, H. R. K., Ostad-Ali- and Forest Meteorology, 234–235,22–30. Askari, K., Eslamian, S., & Singh, V. P. (2018). The use EMBRAPA - Brazilian Agricultural Research Corporation. of remote sensing techniques in detecting and predicting (2004). Solos do Estado de Santa Catarina. Rio de Janeiro: forest vegetation change using MODIS satellite data, Embrapa Solos. Boletim de Pesquisa e Desenvolvimento, Golestan, Iran. American Journal of Engineering and 46,745. Applied Sciences, 11(1), 387–396. Explorer, U. E. (2016). Retrieved from http://earthexplorer. Köppen, W. (1948). Climatologia: Um studio do los climas usgs. gov/. de la tierra (pp. 478). Mexico (DF): Fondo del Cultura Furtado, L. F. A., Francisco, C. N., & Almeida, C. M. Economica. (2013). Análise de imagem baseada em objeto para Liaqata, M. U., Cheemaa, M. J. M., Huangc, W., classificação das fisionomias da vegetação em imagens Mahmoodd, T., Zamana, M., & Khan, M. M. (2017). de alta resolução espacial. Geociências, 32(3), 441–451. Evaluation of MODIS and Landsat multiband vegetation Galparsoro, L. U., & Fernández, S. P. (1999). Medidas de indices used for wheat yield estimation in irrigated Indus concordancia: El índice de Kappa. Cad Aten Primaria, 6, Basin. Computers and Electronics in Agriculture, 138, 169–171. 39–47. Galvão, L. S., Santos, J. R., Roberts, D. A., Breunig, F. M., Liesenberg, V., & Gloaguen, R. (2013). Evaluating SAR Toomey, M., & Moura, Y. M. (2011). On intra-annual polarization modes at L-band for forest classification pur- EVI variability in the dry season of tropical forest: A case poses in Eastern Amazon, Brazil. International Journal of study with MODIS and hyperspectral data. Remote Applied Earth Observation and Geoinformation, 21, Sensing of Environment, 115(9), 2350–2359. 122–135. Gao, B. C. (1996). NDWI – A Normalized Difference Macedo, F. L., Sousa, A. M. O., Gonçalves, A. C., Water Index for remote sensing of vegetation liquid Silva, H. R., & Rodrigues, R. A. F. (2017). Estimativa water from space. Remote Sensing of Environment, 58 do volume de madeira para Eucalyptus sp. com imagens (3), 257–266. de satélite de alta resolução espacial. Scientias Forestalis, Ghebrezgabher, M. G., Yang, T., Yang, X., Wang, X., & 45(114), 237–247. Khan, M. (2016). Extracting and analyzing forest and Maciel, M. N. M. (2002). Estimativa de parâmetros estru- woodland cover change in Eritrea based on landsat data turais de uma floresta primária na Amazônia Oriental using supervised classification. The Egyptian Journal of através de dados orbitais. Tese (Doutorado em Ciências Remote Sensing and Space Science, 19(1), 37–47. Florestais): Universidade Federal do Paraná. 144f. González-Dugo, M. P., & Mateos, L. (2008). Spectral vege- Mallmann, C. L., Prado, D. A., & Pereira Filho, W. (2015). tation indices for benchmarking water productivity of Índice de vegetação por diferença normalizada para irrigated cotton and sugarbeet crops. Agricultural Water caracterização da dinâmica florestal no parque estadual Management, 95(1), 48–58. Quarta Colônia, estado do Rio Grande do Sul - Brasil. Gu, Y., Brown, J. F., Verdin, J. P., & Wardlow, B. (2007). A Revista Brasileira de Geografia Física, 8(5), 1454–1469. five-year analysis of MODIS NDVI and NDWI for grass- Mancino, G., Nolè, A., Ripullone, F., & Ferrara, A. (2014). land drought assessment over the central great plains of the Landsat TM imagery and NDVI differencing to detect United States. Geophysical Research Letters, 34(6), 1–6. vegetation change: Assessing natural forest expansion in Hesketh, M., Sánchez-Azofeifa, A., & Azofeifa, A. (2014). Basilicata, southern Italy. iForest, 7,75–84. A review of remote sensing of tropical dry forests. In Maranhão, D. D. C., Pereira, M. G., Costa, E. M., & Tropical dry forests in the Americas (pp. 95–112). Boca Anjos, L. H. C. (2017). Correção de imagens Raton: CRC Press. e caracterização do uso da terra no município de Huete, A. R. (1988). A soil-adjusted vegetation index. Pinheiral, estado do Rio de Janeiro, Brasil. Caminhos Remote Sensing of Environment, 25(3), 295–309. de Geografia, 18(62), 174–188. Huete, A. R., Didan, K., Miura, T., Rodriguez, E. P., Meneses, P. R., & Sano, E. E. (2012). Classificação pixel Gao, X., & Ferreira, L. G. (2002). Overview of the radio- a pixel de imagens. In MENESES, P. R.; ALMEIDA, metric and biophysical performance of the MODIS vege- T. Introdução ao Processamento de imagens de tation indices. Remote Sensing of Environment, 83(1–2), Sensoriamento Remoto. Brasília: UNB/CNPQ. 195–213. Moreira, M. A., Rudorff, B. F., Barros, M. A., Faria, V. G., IBGE - Brazilian Institute of Geography and Statistics. & Adami, M. (2010). Geotecnologias para mapear (2011) Demographic census 2010. Rio de Janeiro. lavouras de café nos estados de Minas Gerais e São Jensen, J. R. (2009). Remote sensing of the environment: An Paulo. Engenharia Agrícola, 30(6), 1123–1135. earth resource perspective 2/e (pp. 671). New Delhi: Nurdiana, A., & Risdiyanto, I. (2015). Indicator determina- Pearson Education India. tion of forest and land fires vulnerability using landsat-5 Jiang, Z., Huete, A. R., Didan, K., & Miura, T. (2008). TM data (Case Study: Jambi Province). Procedia Development of a two-band enhanced vegetation index Environmental Sciences, 24, 141–151. without a blue band. Remote Sensing of Environment, Oliveira, F. P., Fernandes Filho, E. I., Soares, V. P., & 112(10), 3833–3845. Souza, A. L. (2013). Mapeamento de fragmentos florestais Justice, C. O., Vermote, E., Townshend, J. R., Defries, R., com monodominância de aroeira a partir da classificação Roy, D. P., Hall, D. K., . . . Lucht, W. (1998). The supervisionada de imagens Rapideye. Revista Árvore, 37 Moderate Resolution Spectroradiometer (MODIS): (1), 151–161. GEOLOGY, ECOLOGY, AND LANDSCAPES 169 Oliveira, L. C., Bertol, I., Barbosa, F. T., Campos, M. L., & Plains with ERTS. Third Symposium of ERTS. (pp. SP- Mecabô Junior, J. (2015). Perdas de solo, água 351. 1,301–317). Greenbelt, Maryland, USA: NASA. e nutrientes por erosão hídrica em uma estrada florestal Sarp, G., & Ozcelik, M. (2018). Water body extraction and na Serra Catarinense. Ciência Florestal, 25(3), 655–665. change detection using time series: A case study of Lake Ponzoni, F. J., Shimabukuro, Y. E., & Kuplich, T. M. Burdur, Turkey. Journal of Taibah University for Science, (2012). Sensoriamento remoto aplicado ao estudo da 11(3), 381–391. vegetação (2 ed., pp. 160). São José Dos Campos: Saraiva. Silva, D. E. M., & Costa, D. F. S. (2015). A caracterização Ramirez, G. M., Zullo Junior, J., Assad, E. D., & Pinto, H. S. do uso e ocupação do solo de um sistema flúvio-lagunar (2006). Comparação de dados dos satélites Ikonos-II no litoral semiárido potiguar: Uso da classificação super- e Landsat/ETM+ no estudo de áreas cafeeiras. Pesquisa visionada maxver. Revista CERES, 1(2), 196–200. Agropecuária Brasileira, 41(4), 661–666. Silva, E. A. (2011). Mapeamento do uso e cobertura vegetal Risso, J., Rizzi, R., Rudorff, B. F. T., Adami, M., do município de Floresta - PE ((Dissertação em Ciências Shimabukuro, Y. E., Formaggio, A. R., & Florestais)). Universidade Federal Rural de Pernambuco, Epiphanio, R. D. V. (2012). Índices de vegetação Modis Recife-PE. 56. aplicados na discriminação de áreas de soja. Pesquisa Silva, W. S. (2015). Histórico - Município de Campo Belo Agropecuária Brasileira, 47(9), 1317–1326. do Sul. Retrieved from http://www.campobelodosul.sc. Robinson, N. P., Allred, B. W., Jones, M. O., Moreno, A., gov.br/cms/pagina/ver/codMapaItem/15697. Kimball, J. S., Naugle, D. E., . . . Richardson, A. D. Szabó, S., Gácsi, Z., & Balázs, B. (2016). Specificfeatures of (2017). A dynamic landsat derived Normalized NDVI, NDWI AND MNDWI as reflected in land cover Difference Vegetation Index (NDVI) product for the categories. Landscape and Environment, 10(3–4), conterminous United States. Remote Sensing, 9(8), 863. 194–202. Rosendo, J. S. R. (2005). índices de vegetação Thiersch, A. (1997). Aeficiência das distribuições diamétricas e monitoramento do uso do solo e cobertura vegetal na para prognose da produção de Eucalyptus camaldulensis bacia do Rio Araguari – MG – utilizando dados do sensor (Dissertação (Mestrado em Engenharia Florestal)). MODIS. (Dissertação (Mestrado em Geografia)). Universidade Federal de Lavras, Lavras-MG. 155. Universidade Federal de Uberlândia, Uberlândia. 130. Xue, J., & Su, B. (2017). Significant remote sensing vegeta- Rouse, J., Jr, Haas, R. H., Schell, J. A., & Deering, D. W. tion indices: A review of developments and applications. (1974). Monitoring vegetation systems in the Great Journal of Sensors, 2017(1353691), 1–17. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Geology Ecology and Landscapes Taylor & Francis

Methodological evaluation of vegetation indexes in land use and land cover (LULC) classification

Download PDF
11 pages

Loading next page...
 
/lp/taylor-francis/methodological-evaluation-of-vegetation-indexes-in-land-use-and-land-xztiPYMINV

References (71)

Publisher
Taylor & Francis
Copyright
© 2019 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group.
ISSN
2474-9508
DOI
10.1080/24749508.2019.1608409
Publisher site
See Article on Publisher Site

Abstract

GEOLOGY, ECOLOGY, AND LANDSCAPES 2020, VOL. 4, NO. 2, 159–169 INWASCON https://doi.org/10.1080/24749508.2019.1608409 RESEARCH ARTICLE Methodological evaluation of vegetation indexes in land use and land cover (LULC) classification a a a a Vanessa Sousa da Silva , Gabriela Salami , Marília Isabelle Oliveira da Silva , Emanuel Araújo Silva , a b José Jorge Monteiro Junior and Elisiane Alba a b Department of Forest engineering, University Federal Rural of Pernambuco, Recife, Brazil; Forest engineering, University Federal of Santa Maria, University City, Brazil ABSTRACT ARTICLE HISTORY Received 16 October 2018 Vegetation indices are intended to emphasize the vegetation spectral behavior in relation to Accepted 13 April 2019 the soil and other terrestrial surface targets. The objective of this study was to evaluate the vegetation cover types present in the municipality of Campo Belo do Sul, Brazil, using data KEYWORDS from five vegetation indices obtained through satellite images. In order to do so, calculations Remote sensing; image of the Normalized Difference Vegetation Index (NDVI), Soil-adjusted Vegetation Index (SAVI), processing; Landscape Leaf Area Index (LAI), Enhanced Vegetation Index (EVI) and Normalized Difference Water Index (NDWI) were performed using Quantum Gis software. The generated maps allowed the detection of the different vegetation cover classes, thus the results indicated that there is no specific vegetation index that best represents all the evaluated classes in the study, however, NDVI, EVI, and SAVI had good adjustments in the majority of the thematic classes. Introduction Barros, Faria, & Adami, 2010; Ramirez, Zullo Junior, Assad, & Pinto, 2006; Silva & Costa, 2015). In countries of great extension such as Brazil, earth The reflectance of surface targets combinations at observation through satellite systems based on medium- two or more wavelengths, especially in the visible and resolution multispectral scanners is one of the most infrared region, generates dimensionless radiometric efficient and economical ways to obtain relevant infor- measurements called vegetation indices (VIs). The mation about terrestrial natural resources and the vege- objective of the VIs is to highlight a particular vegeta- tation conditions (Mallmann, Prado, & Pereira Filho, tion characteristic such as leaf area index (LAI), the 2015). Remote sensing can generate useful spectral reflec- percentage of green cover, chlorophyll content, green tance data that provides rapid means for monitoring and biomass and absorbed photosynthetically active radia- managing natural resources. In addition, through visual tion (Jensen, 2009). Those are often used in ecological and digital images processing it is possible to extract research, ecosystems modeling, biophysical parameters biophysical information from the vegetation cover, of vegetation estimation, as well as for monitoring the which is considered crucial to elucidate processes related terrestrial surface (Robinson et al., 2017). VIs are math- to forest distribution, human activities, biodiversity con- ematical models used in studies conducted since the servation, as well as socioeconomic processes (Mancino, 1960s developed to evaluate, monitor and analyze the Nolè, Ripullone, & Ferrara, 2014). vegetation cover and relating the spectral signature and The supervised classification of orbital images is measurable parameters in the field both quantitative a usual method for mapping and evaluation studies of and qualitatively (Mallmann et al., 2015). land use changes and occupation (Antunes, There are several VIs, with their specificapplicabil- Mercante, Esquerdo, Lamparelli, & Rocha, 2012; ity, used in the different representative phytophysiog- Canetti, Garrastazu, Mattos, Braz, & Pellico Netto, nomies of the world biomes for vegetation mapping and 2018; Churches, Wampler, Sun, & Smith, 2014; monitoring on the terrestrial surface. The vegetation Ghebrezgabher, Yang, Yang, Wang, & Khan, 2016; indices allow to obtain information from the spectral Liesenberg & Gloaguen, 2013; Oliveira, Fernandes response of the targets, and thus, to diagnose different Filho, Soares, & Souza, 2013). Among the supervised biophysical parameters such as leaf area index, biomass, classification methods, the algorithm of maximum the percentage of land cover, photosynthetic activity likelihood (Maxver) is one of the most applied meth- and productivity (Ponzoni, Shimabukuro, & Kuplich, odologies to characterization and monitoring studies 2012). To Xue and Su (2017) these indices are simple of forested and agricultural areas (Moreira, Rudorff, and effective algorithms to evaluate the vigor and CONTACT Emanuel Araújo Silva emanuelmadster@gmail.com Department of Forest engineering, University Federal Rural of Pernambuco, Recife, Brazil © 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. 160 V. S. DA SILVA ET AL. dynamics of terrestrial vegetation. It is emphasized that suitable for annual crops with some obstacles from such indices have particularities regarding their sensi- the undulating and gently undulating soils, more tivity towards targets since this relation is influenced by suitable for crops; being predominant the latosols factors inherent to the target. and tropohumult in the less steep areas and lithosols This study aimed to evaluate the efficiency of differ- and neosols in more rugged areas (Embrapa, 2004). ent vegetation indices as NDVI, SAVI, LAI, EVI, and The study area presents in rural activity its main NDWI; in the classification of land use and occupation, socioeconomic activity; most of the properties are in order to identify the index that best represents the composed of natural pastures, temporary/permanent current coverage and more closely resembles the classi- crops and planted forests. According to Silva (2015), fication performed by the MaxVer algorithm. the municipality of Campo Belo do Sul is known for having its economy based on agrosilvopastoral sys- tems crops, with emphasis on forestry from the forest Materials and method farm Gateados with the largest reforested area in the south of the country. The studied area is located in the municipality of The image processed in this work was obtained Campo Belo do Sul, southwest region of the state of from the Landsat 8 Operational Land Imager (OLI) Santa Catarina, Brazil (Figure 1), between the coordi- sensor made available by the USGS, with a passage on nates 27º53’57“ south latitude and 50º45’39” west 11 March 2017, orbit/point 221/79, with a spatial longitude, covering an area of approximately resolution of 30 m for visible bands, near and mid- 1.027,65 km with a population of 7.483 inhabitants, infrared and 15 m for the panchromatic band. The 16 with population density of 7,28 inhab./km ; with an days temporal resolution enables continuous moni- equivalence between urban and rural population toring projects (Explorer, 2016). The geoprocessing of (IBGE, 2011). the cartographic data was performed by the software According to the Köppen classification (1948), the Quantum Gis 2.18.6, with the help of the Semi- climatic type of the Santa Catarina Plateau region is automatic Classification Plugin (SCPA) tool, which a transition between Cfa (wet mesothermic, with no allowed the processing of information from the defined dry season, hot summers, with rare occur- Landsat 8 image. rence of frost in winter) and Cfb (mesothermic wet, For the elaboration of the image chart and analysis with no defined dry season, fresh summers, with of the land cover, the pseudo-color composition was severe and frequent frosts occurring in winter), with used, with the bands stacking 4–5–6 (RGB). The temperatures varying from 13°C to 25° C and rain- conversion of digital levels to surface reflectance was falls distributed throughout the year, totaling an aver- performed through the SCP Pre-Processing options, age of 1841 mm annually (Oliveira, Bertol, Barbosa, which uses the DOS 1 method to correct atmospheric Campos, & Mecabô Junior, 2015). As for the soil, effects. The atmospheric correction is done in order they are shallow and stony, with low fertility and little Figure 1. Location of the municipality of Campo Belo do Sul-SC, BR. GEOLOGY, ECOLOGY, AND LANDSCAPES 161 to eliminate the imperfections that may damage the the occurrences of each class with reference points, information (Maranhão, Pereira, Costa, & Anjos, generating a contingency matrix (matrix of errors). 2017), thus enabling physical surface reflectance The reference points were randomly collected and values to be obtained without the effects of atmo- labeled by visual interpretation. The contingency spheric interference. Regarding the vector data, the matrix allows validating the supervised classification cartographic bases pertinent to the municipal bound- by estimating the overall accuracy of the mapping, as ary, obtained from the Brazilian Geological Service well as the Kappa coefficient (Cohen, 1960), which (CPRM, 2017) were used. quantifies the agreement between classification and The vegetation indices of this study were selected reference data, ranging from zero to one, being more based on those usually applied in studies of this accurate those data that have the value closer to one, nature (Xue & Su, 2017), such as the Normalized while it will have a doubtful veracity the closer the Difference Vegetation Index (NDVI), Soil-Adjusted value is to zero (Silva, 2011). Vegetation Index (SAVI) Leaf area index (LAI), Through the error matrix it was also possible to Enhanced Vegetation Index (EVI) and Normalized extract information regarding the producer accuracy Difference Water Index (NDWI) (Table 1). (relative to the omission error, calculated from the Using the Semi-automatic Classification Plugin reference data, indicates the probability of a reference (SCP) plugin in Quantum Gis 2.18.6 software, the data – field truth, to be correctly classified) and the algebraic operations were performed on the compo- user accuracy (relative to the commission error, cal- nents of the indices mentioned above, from the con- culated from the classification data, indicates the verted bands to apparent reflectance values, and then probability that a classified pixel will effectively repre- the classified images by each index were generated sent the category in the field) (Furtado, 2013) and and then compared. with this information it was possible to evaluate if the The mapping of land use and the cover was classification was effective. obtained by a supervised classification, using the A 100 random samples were collected from each maximum likelihood classifier algorithm (MaxVer), land use and cover class and later extracted classifica- which according to Meneses and Sano (2012), con- tion and indices using the Point Sample Tools. The siders the distance weighting between the means of data were compiled and statistically analyzed in the the digital levels based on statistical parameters. For software R Statistical 3.3.1 through the Rcmdr pack- the study area, the following thematic classes were age and then the contribution of each vegetation proposed: Water bodies (water slide, dams, rivers, index in the identification of the proposed thematic lagoons, artificial lakes); Native forests (areas occu- classes was evaluated. pied by different native forest formations, including Correlation analysis was performed between the vege- permanent preservation areas); Planted forests (estab- tation indices and the land use and occupation classes, lished monocultures occupied with Eucalyptus and determining the linear dependence degree between Pinus genus plantations); built areas (includes urban them. A total of five independent variables were analyzed area, established rural areas, roads and other con- using the original indices. For the variables selection, it structions and infrastructures); agricultural areas was used the Forward method, and the indices that (cultivated areas with undefined crop types); under contributed the most to the identification of land use fallow areas (newly harvested areas, from farming or and occupation classes were used, as it allows to examine forestry, and areas prepared for the next planting); the contribution of each independent index to the regres- and native fields (areas where there is no forest pre- sion model. The models were evaluated based on the sence and characteristic vegetation is a natural pas- adjusted coefficient of determination (R aj), standard ture, planted pastures were also included). error of the estimate (Syx) and coefficient of variation After obtaining the map of land use and cover, in (CV%), where the fitting level of the selected models for order to estimate the accuracy of the classification each class of land use and occupation was determined by data, we performed statistical analyzes that relate the distribution of the residuals (Alba et al., 2017)and by the sum of the statistical scores proposed by Thiersch (1997). It was assigned the lowest weight (one) for the best statistical results of each evaluated index, the best Table 1. Vegetation Indices analyzed in the present study. model was designated by the sum of the scores, values Vegetation index Formula Autors from one to N, where the lowest sum of the scores ρnir ρred NDVI Rouse, Haas, Schell, and NDVI ¼ indicates the selection of the best equation. ρnirþ ρred Deering (1974) ρnir ρred SAVI Huete (1988) SAVI ¼ðÞ 1 þ L ρnirþ ρredþL 0;69SAVI Ln LAI ðÞ Allen, Tasumi, and 0;59 LAI ¼ Results 0;91 Trezza (2002) ρnir ρred EVI Justice et al. (1998) EVI ¼ 2; 5 ðÞ ρnirþ6ρred7;5ρblueþ1 Through the land use and cover mapping, it was ρnir ρswir NDWI Gao (1996) NDWI ¼ ρnirþ ρswir possible to identify seven different thematic classes 162 V. S. DA SILVA ET AL. in the study area (Figure 2). The visual analysis Table 2. Area occupied by thematic classes expressed in hectares and percentage. allowed to identify the predominance of native forests Classes Area (ha) Area (%) in the southern region, associated to watercourses. Native Forests 39,059.37 38.01 The mapping quantitative analysis (Table 2) Native Fields 34,950.87 34.01 expresses the area occupied by the thematic classes, Monocultures 11,945.34 11.63 Agricultural areas 7,510.41 7.31 which corresponds to the original classification. It Under fallow areas 6,181.2 6.01 should be noted that the native forests, native fields, Built areas 1,786.5 1.74 Water 1,314.09 1.28 and established monocultures are the classes that Total area 102,747.78 100 comprise the highest percentages, occupying the lar- gest areas in the municipality. On the other hand, Table 3. Percentage and producer accuracy. built areas and watercourses are less expressive in Classes Producer Accuracy User Accuracy Campo Belo do Sul. Native Forests 89.46 97,48 The Kappa index, calculated by the contingency Native Fields 94.73 92,24 matrix of the thematic map and the reference points, Monocultures 93.52 77,76 Agricultural areas 98.95 97,59 presented a value of 0.88. Thus, the classification was Under fallow areas 89.87 83,14 considered as very good according to the evaluation Built areas 74.60 81,31 criteria of Galparsoro and Fernández (1999), as well Water 99.89 98,43 Global Accuracy 91.59 as its overall accuracy, which expressed a value of approximately 92% (Table 3). The vegetation indices resulting from the maps with each other, except for the native forests class, algebra are shown in Figure 3, demonstrating the which showed little correlation in most of the vari- variations occurring in the classification after applica- ables, showing a high correlation only in relation to tion of the indices when compared to the initial the LAI to the SAVI. classification made by the MaxVer classifier algo- It was also observed that for the native forest, the rithm (Figure 2). relationship between NDVI and NDWI expressed It is inferred that there are variations in the classi- a correlation coefficient of 0.68, representing one of fication of the land cover according to each index, the highest correlations for this class. It should be which are corroborated by observing the different noted that EVI and SAVI were strongly correlated class intervals obtained for each vegetation index when the native field, planted forest, agriculture, and tested in the study (Table 4). urbanization were evaluated, showing the highest The Pearson correlation analysis performed in the correlation coefficients. However, presenting vegetation indices obtained in each use class is a higher correlation does not mean that both indices demonstrated in Figure 4, in which 100% correlated are adequate for the classes, as can be seen in Table 5. variables expressed the value one. It is possible to From the linear regression models, based on the analyze that for the different land uses and cover, scores of each class of land use (Table 5), it can be the vegetation indices showed a high correlation inferred that for native forests the most adequate Figure 2. Map of land use and classification of the municipality of Campo Belo do Sul-SC, Brazil. GEOLOGY, ECOLOGY, AND LANDSCAPES 163 Figure 3. Land use and cover mapping based on vegetation indices NDVI (a), EVI (b), NDWI (c), LAI (d) e SAVI (e). Discussion Table 4. Class intervals for each vegetation index analyzed for the municipality of Campo Belo do Sul, Santa Catarina, Brazil. According to the initial analysis of land use and cover Class Interval of the municipality of Campo Belo do Sul, it is note- Vegetation Index Minimum Value Maximum Value worthy that native compositions of the region, repre- EVI −0.235 −0.054 NDWI −0.246 0.143 sented by forests and native fields, occupy LAI 0.745 4.09 approximately 70% of the municipality. This result SAVI 0.357 0.964 NDVI −0.603 0.936 corroborates with the percentage of the classes related to the anthropic action being less expressive, allowing to infer that the municipality environmental degrada- tion is not an expressive practice. The native fields are indices were NDVI and EVI; for planted forests, present in 34% of the area of the municipality, SAVI, followed by EVI and NDVI; for the class of a percentage below only the native vegetation, agricultural areas were SAVI and EVI, the same which can be explained due to the extensive livestock indices that were more adequate for the native fields; practice in this territory, in which the native fields for the fallow areas class the best adjustments were (natural pastoral ecosystem) offer the support for the with the NDWI and NDVI indices; NDWI was also development of the activity (Kaibara, 2014). the best performing index to determine constructed It is also observed that approximately 50% of the areas; and, for the waterbodies, NDVI and SAVI were municipal territory is covered by shrub-tree vegeta- more indicated. Making it possible to infer that there tion, adding native forests (38%) and established was not an index that excelled in all classes. 164 V. S. DA SILVA ET AL. Figure 4. Correlation of vegetation indices for different land uses and cover: Native forests (a), Native fields (b), Monocultures (c), Agricultural areas (d), Under fallow areas (e) and Built areas (g). monocultures (12%) that are composed of Eucalyptus for the development of agriculture and livestock pro- and Pinus tree individuals. However, in 8% of the duction in the region. Agricultural areas (7%) and municipality, there is no vegetation cover, analyzing watercourses (1%) were also present in the identified areas under fallow and also built areas; the areas classes, being less expressive. under fallow have its soil exposed, which are The variation of accuracy among the classes eval- a result of the intense soil exploitation carried out uated, can be associated with common remote GEOLOGY, ECOLOGY, AND LANDSCAPES 165 Table 5. Linear regression models resulting from the Forward selection method for estimating land use and land cover. Regression Model F R aj Syx CV(%) Scores Native Forests Y= −0,3532 + 0,9790 (EVI) 56.88*** 0.3608 0.0008 1.55 8 Y = 0,5543–0,4937 (NDVI) 528.3*** 0.8419 0.0023 2.78 6 Y = 0,2046–0,1567 (NDWI) 311.3*** 0.7581 0.0070 16.58 10 n.s 2.32 0.01316 0.0048 2.65 - n.s 2.342 0.01338 0.0005 0.88 - Monocultures Y = 0,00937 + 0,23398(EVI) 417.1*** 0.8078 0.0055 10.91 10 Y= −0,1086 + 0,2782(NDVI) 17.37*** 0.1419 0.0020 2.36 10 Y = 0,09230 + 0,06745(NDWI) 5.715* 0.04546 0.0050 9.44 16 Y = 0,07622 + 0,02715(IAF) 279.3*** 0.7376 0.0456 23.88 16 Y= −0,04891 + 0,30,626(SAVI) 328.2*** 0.7701 0.0041 7.14 8 Agricultural Areas Y = 0,02528 + 0,29251(EVI) 911.1*** 0.9019 0.0029 3.14 10 Y= −0,4443 + 0,8072(NDVI) 129.3*** 0.5644 0.0008 0.91 12 Y = 0,008275 + 0,695017(NDWI) 213.6*** 0.6823 0.0011 2.57 13 Y = 0,35177–0,03338(IAF) 229.3*** 0.6975 0.0225 13.79 16 Y= −0,1577 + 0,5512(SAVI) 624.9*** 0.863 0.0015 1.83 9 Under Fallow Areas n.s 0.1003 0.009171 0.0082 25.16 - Y = 0,22976–0,08759(NDVI) 10.66** 0.08893 0.0114 20.43 6 Y = 0,1926–0,1492(NDWI) 54.31*** 0.35 0.0128 162.27 6 n.s 0.2263 0.007876 0.0306 38.69 - n.s 0.4321 0.005769 0.0085 21.65 - Built Areas n.s 0.001082 0.01019 0.0096 38.66 - n.s 2.035 0.01035 0.0135 35.03 - Y = 0,20189–0,05121(NDWI) 5.595* 0.04435 0.0090 233.07 4 n.s 0.2402 0.007734 0.0335 72.95 - n.s 0.4388 0.005701 0.0103 36.07 - Native Fields Y = 0,07465 + 0,22079(EVI) 65.92*** 0.396 0.0064 12.21 6 n.s 0.06224 0.009563 0.0045 5.89 - n.s 1.325 0.003276 0.0063 24.94 - Y = 0,1537 + 0,0191(IAF) 38.5*** 0.2747 0.0617 32.57 12 Y = 0,05102 + 0,24281(SAVI) 39.95*** 0.2823 0.0049 8.60 6 Water n.s 0.4378 0.005711 0.0002 2.38 - Y = 0,05341 + 0,06213(NDVI) 524.8*** 0.841 0.0023 5.32 9 Y = 0,03615–0,08531(NDWI) 162.6*** 0.6202 0.0015 12.44 11 Y = 0,07703 0,15110(IAF) 15.22*** 0.1256 0.0004 1.14 11 Y = 0,04929 + 0,20745(SAVI) 15.25*** 0.1258 0.0003 2.50 9 n.s. Signif, codes: 0 ‘***‘; 0,001 ‘**‘; 0,01 ‘*‘; not significant ‘ ‘ sensing limitations, among them the distinction of When comparing the classification of land use by targets with partially similar spectral responses, such MaxVer (initial classification) with the vegetation as: native and planted forests, native fields, agricul- indices, it was noticed variations in the classification tural areas and areas under fallow. Another possible of the municipality. Among vegetation indices, this confounding factor in the automatic classification is difference is also expected, since they are based on the characteristic relief, because the sun-earth-sensor different algorithms to classify the images, each index alignment prevents the electromagnetic radiation having its specificities and different levels of detail. from being reflected by the targets located on steep The correlogram within each thematic class slopes in full shape, detecting only shadows. This showed that some indices have a high correlation factor interferes in the spectral response of targets between them for most classes. Macedo, Sousa, located on slopes and consequently in the quality of Gonçalves, Silva, and Rodrigues (2017), seeking to the automatic classification (Caixeta, Edmundo, develop allometric functions to estimate total and Rodrigues, Moreira, & Medeiros, 2012). commercial volume, using as a dependent variable The understanding of the land use and occupation a vegetation index, infers that the high correlations forms, with modeling purposes and representing obtained for vegetation indices are justified mainly large areas with good levels of detail, can be consid- due to their compositions, which essentially use the erably improved using geoprocessing tools. In addi- spectral bands in the region of the electromagnetic tion to reducing fieldwork, the extraction of spectrum related to red and infrared. Thus, such information in medium-resolution digital images, behaviors observed in the present study are expected, such as the Landsat 8 satellite, allows larger areas to since in order to determine some indices there is be studied in a shorter time. dependence on others. 166 V. S. DA SILVA ET AL. From the regression analysis, it was possible to noting that both NDWI and NDVI are prominent in notice that in each thematic class the indices are identifying vegetation and water coverages consider- presented in a different way, some being more ade- ing their individual restrictions. quate than others in the determination of the differ- With another approach, Choung and Jo (2016) ent coverages. used NDWI to monitor changes in water resources In determining the native forests class, NDVI was in South Korea using Landsat multitemporal images; the best index, also suitable for determining waterbo- their results showed significant differences in water- dies (best fit), areas under fallow (second best fit) and body sizes throughout the study years. Sarp and planted forests (third best fit). NDVI is a relevant index Ozcelik (2018) also used NDWI to assess changes in for areas of medium to high vegetation density, since it water resources in southwestern Turkey, and the is less susceptible to the soil and to the effects of the results effectively showed the detection of change in atmosphere. According to Ponzoni et al. (2012), this the water surface between specified time intervals. On index has been used to detect the effects of seasonality, the other hand, aiming to determine indicators for phenological stage of vegetation, duration of the growth assessing the vulnerability of forests and fires in period, green peak, physiological changes of leaves, Indonesia, Nurdiana and Risdiyanto (2015) used senescence periods and other important vegetation NDVI and NDWI as key parameters for the identifi- related situations. However, it is not suitable for areas cation of fire outbreaks, where both indices were able with low vegetation coverage (Karimi et al., 2018). The to provide the highest contribution to the vulnerabil- index values range from −1to +1and isbased on the ity level. high reflection of the healthy plants in the wavelengths Efforts around the world have been observed to in the infrared band and its low reflection in the pre- justify the use of this index (associated or not to sence of the red band of the electromagnetic spectrum others) to identify dry areas. Gu, Brown, Verdin, (Hesketh, Sánchez-Azofeifa, & Azofeifa, 2014), there- and Wardlow (2007) evaluated the results of the use fore, healthy forest cover usually has higher NDVI of satellite-derived indices (NDVI and NDWI) to values. monitor vegetation drought through soil moisture This is the most commonly used vegetation index, observations in the United States. They found that which minimizes topographic effects by producing there is a strong relationship between the vegetation a linear scale of measurement, in which the closer to indices with the heterogeneity of the land cover, soil 1 the greater the vegetation cover density and, in its type and humidity, and for areas of homogeneous turn, the 0 would be the approximate value for the forest cover, both were sensitive to changes in soil absence of vegetation, that is, represents non-vegetated moisture and are strongly related to vegetation surfaces (Rosendo, 2005). In this regard, it is justified drought conditions; suggesting that both indices are that this index had good results to determine water- appropriate to monitor water stress in vegetation. bodies and areas under fallow, being an index that Szabó, Gácsi, and Balázs (2016) investigated ranges presents comprehensive values of class intervals. of three spectral indices for land cover types by com- In the determination of areas under fallow and paring the spectral ratios by land cover types and areas with buildings, as well as waterbodies (after assessed their efficiency in discriminating land cover NDVI), the best fit index was NDWI. This is classes. As well as the multiple functions that the amodification of the NDVI, that allows to highlight authors suggest for NDWI, the results of this study water features and minimize the rest of the targets. foster the idea that their use can be indicated to The NDWI is highly correlated with the water content evaluate various land coverages. in the vegetation cover, making it possible to measure It was from the use of the abovementioned indices biomass changes and to evaluate vegetation water that the need for an index that considered soil stress, through mathematical operations using near- response, which could be dominant on the vegetation infrared and medium infrared bands (Jensen, 2009). response, was found, depending on the coverage per- However, some authors have used NDWI for centage. Thus, in order to mitigate this soil effect, numerous purposes. When investigating phenological SAVI (Huete, 1988) was created, which is based on metrics in dry and dormancy periods in semi-arid the principle that the vegetation curve tends to pastures, Ding, Liu, Huan, Li, and Zou (2017) have approach the soil curve for low vegetation densities, found results that suggest that phenological studies passing through a mixture of spectral responses to using NDWI can expand the understanding of the average densities and almost no influence of the soil terrestrial surface phenology; furthermore, they sug- to high vegetation densities (Sousa & Ponzoni, 1998). gest that the combination of this index with climatic In this study, SAVI was the best fit index to deter- variability could contribute to the study of ecosystem mine planted forests, agricultural areas, native fields, processes in semi-arid pastures. Ahmed and Akter and waterbodies. According to Huete (1988), the (2017) studied changes in land use and cover follow- SAVI adjusted to different soil conditions that may ing regular flooding in coastal areas of Bangladesh, exert considerable influence on the canopy, thus in GEOLOGY, ECOLOGY, AND LANDSCAPES 167 areas where there are considerable variations in the Disclosure statement soil brightness resulting from differences in moisture, No potential conflict of interest was reported by the variations in roughness, shade or differences in authors. organic matter, there are soil-induced influences on vegetation index values. For this reason, this index reduces the soil effects and it is probable that in this Funding study, its good adjustment in the different coverages This work was supported by the Conselho Nacional de was provoked precisely because these areas have soils Desenvolvimento Científico e Tecnológico [1]; where the reflectance is favored. Coordenação de Aperfeiçoamento de Pessoal de Nível Corroborating with the results obtained in the Superior [3];Fundação de Amparo à Ciência e Tecnologia present study, Liaqata et al. (2017) found that SAVI do Estado de Pernambuco [1]. was the index that best adjusted for estimated agri- cultural production in irrigated areas in Pakistan; González-Dugo and Mateos (2008) observed that References SAVI also excels when used in irrigated cotton and Ahmed, K. R., & Akter, S. (2017). Analysis of landcover beet crops in southern Spain. For planted forests change in southwest Bengal delta due to floods by areas, Alba et al. (2017) demonstrated that SAVI NDVI, NDWI and K-means cluster with landsat was the index that had the best correlation with the multi-spectral surface reflectance satellite data. Remote volume estimation for a Pinus elliottii forest. In Sensing Applications Society and Environment, 8(1), 168–181. a study developed by Cassol (2013), Maciel (2002) Alba, E., Mello, E., Marchesan, J., Silva, E. A., and Bernardes (1998), the SAVI index also stood out Tramontina, J., & Pereira, R. S. (2017). Spectral charac- as one of the spectral variables with the greatest terization of forest plantations with Landsat 8/OLI relation to the forest biomass in a Mixed images for forest planning and management. Pesquisa Ombrophilous Forest, which has a high population agropecuária brasileira, 52(11), 1072–1079. density. These authors reinforce that soil brightness Allen, R., Tasumi, M., & Trezza, R. (2002). SEBAL (Surface Energy Balance Algorithms for Land) – advanced training influences the spectral response, even in closed cano- and users manual – Idaho implementation, version 1.0. pies and with high individual density.b Antunes, J. F. G., Mercante, E., Esquerdo, J. C. D. M., The EVI was the second-best index for the four land Lamparelli, R. A. C., & Rocha, J. V. (2012). Estimativa uses, being the most sensitive to the canopy structure de área de soja por classificação de imagens normalizada (Huete et al., 2002), although it is more influenced by pela matriz de erros. Pesquisa Agropecuária brasileira, 47 (9), 1288–1294. the solar zenith angle (Galvão et al., 2011) under cer- Bernardes, S. (1998). Índices de vegetação e valores de tain conditions. This index has higher sensitivity in proporção na caracterização de floresta tropical regions with high biomass production and reduces primária e estádios sucessionais na área de influência da the atmospheric and soil influences (Jiang, Huete, Floresta Nacional do Tapajós-Estado do Pará. Didan, & Miura, 2008). Risso et al. (2012)evaluated Dissertação (Mestrado em Sensoriamento Remoto): the performance of EVI and NDVI to discriminate Instituto Nacional de Pesquisas Espaciais. 94f. Caixeta, P. R. C., Edmundo, I. S. B., Rodrigues, A. F. S., soybean areas from sugarcane, pasture, forested and Moreira, G. F., & Medeiros, N. G. (2012). Processamento Cerrado areas, they found that EVI excelled in the de Imagens de Alta Resolução RapidEye para image classification, quite possibly due to this charac- Mapeamento do Meio Físico. IV Simpósio Brasileiro de teristic of a greater sensitivity to higher biomass values. Ciências Geodésicas e Tecnologias da Geoinformação (pp.001–008). Recife: PE-BR. Canetti, A., Garrastazu, M. C., Mattos, P. P., Braz, E. M., & Pellico Netto, S. (2018). Understanding multi-temporal Conclusion urban forest cover using high resolution images. Urban There is no specific vegetation index that best repre- Forestry and Urban Greening, 29(1), 106–112. Cassol, H. L. G. (2013). Estimativa de biomassa e estoque de sents all the classes evaluated in the study; however, carbono em um fragmento de floresta ombrófila mista NDVI, EVI, and SAVI had good adjustments in most com uso de dados ópticos de sensores remotos of the thematic classes. The choice of the best vegeta- (Dissertação (Mestrado em Sensoriamento Remoto)). tion index for land use and occupation classification Universidade Federal do Rio Grande do Sul, 143. depends, particularly on the predominant soil use. Choung, Y. J., & Jo, M. H. (2016). Monitoring water resource changes using multi-temporal landsat imagery In general, it is possible to suggest that the infor- taken in changnyeong, South Korea. Procedia mation spectral mixture and the determination of Engineering, 154(1), 348–352. several thematic classes favored the non-definition Churches, C. E., Wampler, P. J., Sun, W., & Smith, A. J. of the best vegetation index, and thus it can be (2014). Evaluation of forest cover estimates for Haiti inferred that depending on the objective and the using supervised classification of Landsat data. type of predominant use, more adequate indexes International Journal of Applied Earth Observation and Geoinformation, 30(1), 203–216. can be indicated. 168 V. S. DA SILVA ET AL. Cohen, J. (1960). A coefficient of agreement for nominal Land remote sensing for global change research. IEEE scales. Educational and Psychological Measurement, 20 Transactions on Geosciences and Remote Sensing, 36(4), (1), 37–46. 1228–1249. CPRM. (2017) Serviço Geológico Brasileiro. Geobank. Retrieved Kaibara, M. A. (2014). Rendimento animal em uma pasta- from http://geobank.sa.cprm.gov.br/ gem nativa melhorada no planalto sul catarinense (pp. Ding, H., Liu, X., Huan, F., Li, Y., & Zou, X. (2017). Onset 60). Monografia, Florianópolis (SC): UFSC – of drying and dormancy in relation to water dynamics of Universidade Federal de Santa Catarina. semi-arid grasslands from MODIS NDWI. Agricultural Karimi, A., Abdollahi, S., Balajadeh, H. R. K., Ostad-Ali- and Forest Meteorology, 234–235,22–30. Askari, K., Eslamian, S., & Singh, V. P. (2018). The use EMBRAPA - Brazilian Agricultural Research Corporation. of remote sensing techniques in detecting and predicting (2004). Solos do Estado de Santa Catarina. Rio de Janeiro: forest vegetation change using MODIS satellite data, Embrapa Solos. Boletim de Pesquisa e Desenvolvimento, Golestan, Iran. American Journal of Engineering and 46,745. Applied Sciences, 11(1), 387–396. Explorer, U. E. (2016). Retrieved from http://earthexplorer. Köppen, W. (1948). Climatologia: Um studio do los climas usgs. gov/. de la tierra (pp. 478). Mexico (DF): Fondo del Cultura Furtado, L. F. A., Francisco, C. N., & Almeida, C. M. Economica. (2013). Análise de imagem baseada em objeto para Liaqata, M. U., Cheemaa, M. J. M., Huangc, W., classificação das fisionomias da vegetação em imagens Mahmoodd, T., Zamana, M., & Khan, M. M. (2017). de alta resolução espacial. Geociências, 32(3), 441–451. Evaluation of MODIS and Landsat multiband vegetation Galparsoro, L. U., & Fernández, S. P. (1999). Medidas de indices used for wheat yield estimation in irrigated Indus concordancia: El índice de Kappa. Cad Aten Primaria, 6, Basin. Computers and Electronics in Agriculture, 138, 169–171. 39–47. Galvão, L. S., Santos, J. R., Roberts, D. A., Breunig, F. M., Liesenberg, V., & Gloaguen, R. (2013). Evaluating SAR Toomey, M., & Moura, Y. M. (2011). On intra-annual polarization modes at L-band for forest classification pur- EVI variability in the dry season of tropical forest: A case poses in Eastern Amazon, Brazil. International Journal of study with MODIS and hyperspectral data. Remote Applied Earth Observation and Geoinformation, 21, Sensing of Environment, 115(9), 2350–2359. 122–135. Gao, B. C. (1996). NDWI – A Normalized Difference Macedo, F. L., Sousa, A. M. O., Gonçalves, A. C., Water Index for remote sensing of vegetation liquid Silva, H. R., & Rodrigues, R. A. F. (2017). Estimativa water from space. Remote Sensing of Environment, 58 do volume de madeira para Eucalyptus sp. com imagens (3), 257–266. de satélite de alta resolução espacial. Scientias Forestalis, Ghebrezgabher, M. G., Yang, T., Yang, X., Wang, X., & 45(114), 237–247. Khan, M. (2016). Extracting and analyzing forest and Maciel, M. N. M. (2002). Estimativa de parâmetros estru- woodland cover change in Eritrea based on landsat data turais de uma floresta primária na Amazônia Oriental using supervised classification. The Egyptian Journal of através de dados orbitais. Tese (Doutorado em Ciências Remote Sensing and Space Science, 19(1), 37–47. Florestais): Universidade Federal do Paraná. 144f. González-Dugo, M. P., & Mateos, L. (2008). Spectral vege- Mallmann, C. L., Prado, D. A., & Pereira Filho, W. (2015). tation indices for benchmarking water productivity of Índice de vegetação por diferença normalizada para irrigated cotton and sugarbeet crops. Agricultural Water caracterização da dinâmica florestal no parque estadual Management, 95(1), 48–58. Quarta Colônia, estado do Rio Grande do Sul - Brasil. Gu, Y., Brown, J. F., Verdin, J. P., & Wardlow, B. (2007). A Revista Brasileira de Geografia Física, 8(5), 1454–1469. five-year analysis of MODIS NDVI and NDWI for grass- Mancino, G., Nolè, A., Ripullone, F., & Ferrara, A. (2014). land drought assessment over the central great plains of the Landsat TM imagery and NDVI differencing to detect United States. Geophysical Research Letters, 34(6), 1–6. vegetation change: Assessing natural forest expansion in Hesketh, M., Sánchez-Azofeifa, A., & Azofeifa, A. (2014). Basilicata, southern Italy. iForest, 7,75–84. A review of remote sensing of tropical dry forests. In Maranhão, D. D. C., Pereira, M. G., Costa, E. M., & Tropical dry forests in the Americas (pp. 95–112). Boca Anjos, L. H. C. (2017). Correção de imagens Raton: CRC Press. e caracterização do uso da terra no município de Huete, A. R. (1988). A soil-adjusted vegetation index. Pinheiral, estado do Rio de Janeiro, Brasil. Caminhos Remote Sensing of Environment, 25(3), 295–309. de Geografia, 18(62), 174–188. Huete, A. R., Didan, K., Miura, T., Rodriguez, E. P., Meneses, P. R., & Sano, E. E. (2012). Classificação pixel Gao, X., & Ferreira, L. G. (2002). Overview of the radio- a pixel de imagens. In MENESES, P. R.; ALMEIDA, metric and biophysical performance of the MODIS vege- T. Introdução ao Processamento de imagens de tation indices. Remote Sensing of Environment, 83(1–2), Sensoriamento Remoto. Brasília: UNB/CNPQ. 195–213. Moreira, M. A., Rudorff, B. F., Barros, M. A., Faria, V. G., IBGE - Brazilian Institute of Geography and Statistics. & Adami, M. (2010). Geotecnologias para mapear (2011) Demographic census 2010. Rio de Janeiro. lavouras de café nos estados de Minas Gerais e São Jensen, J. R. (2009). Remote sensing of the environment: An Paulo. Engenharia Agrícola, 30(6), 1123–1135. earth resource perspective 2/e (pp. 671). New Delhi: Nurdiana, A., & Risdiyanto, I. (2015). Indicator determina- Pearson Education India. tion of forest and land fires vulnerability using landsat-5 Jiang, Z., Huete, A. R., Didan, K., & Miura, T. (2008). TM data (Case Study: Jambi Province). Procedia Development of a two-band enhanced vegetation index Environmental Sciences, 24, 141–151. without a blue band. Remote Sensing of Environment, Oliveira, F. P., Fernandes Filho, E. I., Soares, V. P., & 112(10), 3833–3845. Souza, A. L. (2013). Mapeamento de fragmentos florestais Justice, C. O., Vermote, E., Townshend, J. R., Defries, R., com monodominância de aroeira a partir da classificação Roy, D. P., Hall, D. K., . . . Lucht, W. (1998). The supervisionada de imagens Rapideye. Revista Árvore, 37 Moderate Resolution Spectroradiometer (MODIS): (1), 151–161. GEOLOGY, ECOLOGY, AND LANDSCAPES 169 Oliveira, L. C., Bertol, I., Barbosa, F. T., Campos, M. L., & Plains with ERTS. Third Symposium of ERTS. (pp. SP- Mecabô Junior, J. (2015). Perdas de solo, água 351. 1,301–317). Greenbelt, Maryland, USA: NASA. e nutrientes por erosão hídrica em uma estrada florestal Sarp, G., & Ozcelik, M. (2018). Water body extraction and na Serra Catarinense. Ciência Florestal, 25(3), 655–665. change detection using time series: A case study of Lake Ponzoni, F. J., Shimabukuro, Y. E., & Kuplich, T. M. Burdur, Turkey. Journal of Taibah University for Science, (2012). Sensoriamento remoto aplicado ao estudo da 11(3), 381–391. vegetação (2 ed., pp. 160). São José Dos Campos: Saraiva. Silva, D. E. M., & Costa, D. F. S. (2015). A caracterização Ramirez, G. M., Zullo Junior, J., Assad, E. D., & Pinto, H. S. do uso e ocupação do solo de um sistema flúvio-lagunar (2006). Comparação de dados dos satélites Ikonos-II no litoral semiárido potiguar: Uso da classificação super- e Landsat/ETM+ no estudo de áreas cafeeiras. Pesquisa visionada maxver. Revista CERES, 1(2), 196–200. Agropecuária Brasileira, 41(4), 661–666. Silva, E. A. (2011). Mapeamento do uso e cobertura vegetal Risso, J., Rizzi, R., Rudorff, B. F. T., Adami, M., do município de Floresta - PE ((Dissertação em Ciências Shimabukuro, Y. E., Formaggio, A. R., & Florestais)). Universidade Federal Rural de Pernambuco, Epiphanio, R. D. V. (2012). Índices de vegetação Modis Recife-PE. 56. aplicados na discriminação de áreas de soja. Pesquisa Silva, W. S. (2015). Histórico - Município de Campo Belo Agropecuária Brasileira, 47(9), 1317–1326. do Sul. Retrieved from http://www.campobelodosul.sc. Robinson, N. P., Allred, B. W., Jones, M. O., Moreno, A., gov.br/cms/pagina/ver/codMapaItem/15697. Kimball, J. S., Naugle, D. E., . . . Richardson, A. D. Szabó, S., Gácsi, Z., & Balázs, B. (2016). Specificfeatures of (2017). A dynamic landsat derived Normalized NDVI, NDWI AND MNDWI as reflected in land cover Difference Vegetation Index (NDVI) product for the categories. Landscape and Environment, 10(3–4), conterminous United States. Remote Sensing, 9(8), 863. 194–202. Rosendo, J. S. R. (2005). índices de vegetação Thiersch, A. (1997). Aeficiência das distribuições diamétricas e monitoramento do uso do solo e cobertura vegetal na para prognose da produção de Eucalyptus camaldulensis bacia do Rio Araguari – MG – utilizando dados do sensor (Dissertação (Mestrado em Engenharia Florestal)). MODIS. (Dissertação (Mestrado em Geografia)). Universidade Federal de Lavras, Lavras-MG. 155. Universidade Federal de Uberlândia, Uberlândia. 130. Xue, J., & Su, B. (2017). Significant remote sensing vegeta- Rouse, J., Jr, Haas, R. H., Schell, J. A., & Deering, D. W. tion indices: A review of developments and applications. (1974). Monitoring vegetation systems in the Great Journal of Sensors, 2017(1353691), 1–17.

Journal

Geology Ecology and Landscapes – Taylor & Francis

Published: Apr 2, 2020

Keywords: Remote sensing; image processing; Landscape

There are no references for this article.