The Effect of Land Use and Land Cover Changes on Soil Erosion in Semi-arid Areas Using Cloud-based Google Earth Engine Platform and GIS-based RUSLE Model

Soil erosion has recently attracted the attention of researchers and managers as an environmental crisis. One of the effective factors in soil erosion is land use/land cover change (LU/LCC). Use of satellite imagery is a method for generating LU/LCC maps. Recently, Google has launched the cloud-based Google Earth Engine (GEE) platform, which enabled the processing of satellite images online. Accordingly, the purpose of the present study is to investigate the effect of LU/LCC on soil erosion in a semi-arid region in the south-west of Iran. LU/LCC map was prepared over a period of 30 years (1989–2019) using a new approach and classi�cation of the Normalized Difference Vegetation Index (NDVI) index time series on the GEE. For classifying the NDVI time series, a non-parametric Support Vector Machine (SVM) classi�cation method was employed. The LU/LC maps were also used as an input factor in the soil erosion estimation model. The amount of soil erosion in the region was estimated using the Revised Universal Soil Loss Equation (RUSLE) empirical model in the Geographical Information System (GIS) environment. Validation of LU/LC maps generated in GEE indicated overall accuracy higher than 86% and the kappa coe�cient higher than 0.82. The study of LU/LCC trends showed that the area of forests, pastures, and rock outcrop in the region has diminished, but the area of agricultural and man-made LUs has been expanded. Also, the highest rate of LU/LC conversion was related to the conversion of forests to agricultural lands. Estimating the amount of soil erosion in the region using the RUSLE model revealed that the average annual erosion in 1989 and 2019 was 15.48 and 20.41 tons per hectare, respectively, which indicates an increase of 4.93 tons in hectares, while the hot spots of erosion in the area have increased at the con�dence levels of 90, 95, and 99%. Matching the LU/LCC map with the soil erosion map indicated that the degradation of forests and their conversion to agricultural lands had the greatest impact on increasing soil erosion. Based on the �ndings, we can conclude that GEE, as an online platform, has a high capability in preparing LU/LC maps and other effective factors in soil erosion estimation models.


Introduction
Soil is one of the non-renewable natural resources, which as an environment for the growth of plants and agricultural products, provides about 95% of human nutritional demands both directly and indirectly (FAO, 2015; Saha et al., 2022).For this reason, any factor that causes destruction of this vital layer of the earth's surface is considered a threat to the lives of humans and other organisms.On a global scale, there are various processes for land degradation, of which soil erosion is one of the most important issues (Prăvălie, 2021;Thomaz et al., 2022).According to the report of the Food and Agriculture Organization of the United Nations (FAO), the process of soil erosion is worsening in the continents of Asia, Africa, and Latin America (Pennock, 2019;Admas et al., 2022).Considering the increase in soil erosion and its negative impact on soil fertility, production of agricultural products, quality of water resources, the capacity of dams, quality of the environment, animal habitat, food security of living organisms, carbon sequestration cycle, and other ecosystem services, this phenomenon is considered as a complex and dynamic global environmental issue and attracted the attention of researchers, managers, and governments (Alkharabsheh et al., 2013;Aiello et al., 2015;Singh and Panda, 2017; Barman et al., 2020;Admas et al., 2022;Bag et al., 2022;Gong et al., 2022;Senanayake et al., 2022).Today, it has been proven that various factors such as climate change, LU/LCC, population increase, soil characteristics, physiographic and geomorphological characteristics of the watershed, land management systems, rainfall, human factors, and deforestation in uence the amount of soil erosion in different areas (Aiello et al., 2015;El Jazouli et al., 2019;Becker et al., 2021;Gong et al., 2022;Senanayake et al., 2022).
Managing and controlling soil erosion requires access to up-to-date data as well as estimating and predicting its amount in different areas.Nowadays, due to the costly and time-consuming soil erosion measurement using traditional and eld methods, the use of soil erosion estimation models has become popular (Alkharabsheh et al., 2013; Barman et al., 2020).In general, the models for estimating soil erosion can be classi ed into three categories: empirical, physical, and conceptual models (Ganasri and Ramesh, 2016; Singh and Panda, 2017).One of the widely used empirical models is the Universal Soil Loss Equation (USLE) model and its revised version called RUSLE, which are used globally (Ganasri and Ramesh, 2016; Barman et al., 2020).The advantages of using the RUSLE model include no need for complex input data, simple and comprehensible structure, suitability for a regional scale, and implementability in forests, pasture, agricultural, and man-made areas (Zhang et al., 2006;Uddin et al., 2018; Barman et al., 2020).In addition, the RUSLE model, with the potential to be implemented in the GIS environment and the use of remote sensing data, allows for spatial and temporal estimation of the soil erosion rate in different areas (Barman et al., 2020;Gong et al., 2022).As mentioned, LU/LCC is an effective and important factor in soil erosion rate in different regions.For this reason, access to the map of LU/LCC is a key point in identifying the critical point of soil erosion and its sustainable management.
Nowadays, the use of satellite imagery has become a common method for generating LU/LC maps, with advantages such as wide coverage, access to images at different times, and access to information on inaccessible areas (Naghavi et al., 2013;Adam et al., 2016;Naseri et al., 2019;Prasai et al., 2021;Faruque et al., 2022;Kuma et al., 2022).Satellite images with different spatial resolutions can be used to prepare the LU/LC map.Use of high spatial resolution satellite imagery can boost the accuracy of LU/LC maps (Luo and Ji, 2022), but application of these images is usually costly, which is a challenge in projects with a limited budget in developing and underdeveloped countries.For this reason, the use of images from the Landsat satellites with medium spatial resolution has been developed in these areas, which allows free access to their image archives from previous decades (Cohen and Goward, 2004;Delfan et al., 2020;Ang et al., 2021;Ding et al., 2022).
One of the challenges that researchers face in processing satellite images is the time-consuming processing of time series images as well as images related to large areas.In 2010, Google launched an advanced cloud-based platform called GEE for the online processing of satellite images (Ang et al., 2021;Peng and Dai, 2022;Nghia et al., 2022).This cloud space, while providing access to a huge archive of satellite images, eliminates the limitations of traditional methods and offers users the possibility of online processing of time series of satellite images with a large volume using millions of servers around the world (Gemitzi and Koutsias, 2022).Given the development of this platform in recent years and the mentioned advantages, its use is growing, and many researchers including Huang et al., (2017) Considering the importance of soil erosion, the purpose of this research is to examine the effect of LU/LCC on soil erosion in a semi-arid watershed using the GEE platform and GIS-based RUSLE model.In this way, the LU/LCC map was generated in a 30-year period using a new approach, by combining the time series of the NDVI spectral index related to each year and classifying it via the non-parametric SVM classi cation method in the GEE.Then, its effect on the rate of soil erosion changes was investigated using the RUSLE model.

Study Area
Khorramabad watershed with an area of 1608.22 km 2 is located in the geographical coordinate range of to east longitude and to north latitude in Lorestan and south-west of Iran (Fig. 1).This region has a semi-arid climate with an average annual temperature of 15 and an average annual rainfall of 405 mm.The minimum, maximum, and average altitudes of the area are 1174, 3000, and 1695.3 m above sea level, respectively.Also, the average slope of the watershed is 24.36% (Mohammadlou and Zeinivand, 2019).The main LU/LC in the region includes oak forests, agricultural lands, pastures, and man-made areas.℃ Landsat 5 satellite related to 1989 were utilized (https://earthengine.google.com).For this purpose, each year was divided into three periods of four months where the images of each period were called in GEE and clip using the region border vector layer.Then, all images were converted to NDVI vegetation index using Eq. 1, and all NDVI values of each period were converted into one band through the Maximum Value Composite (MVC) function (Eq.2) (Huang et al., 2017).Finally, an image containing three bands for each year was generated, with the values of each band of this image showing the maximum value of the NDVI in each four-month period (Fig. 2) (Ahrari, 2020).Next, it was necessary to introduce training samples to GEE for supervised classi cation.In this regard, the samples were randomly collected in different classes, including forest, pasture, man-made, agricultural, and rock outcrop, using eld data, aerial photos, and GEE.Overall, 3000 samples were collected for each year, of which 1000, 1000, 600, 200, and 200 samples were related to forest, agriculture, pasture, man-made, and rock outcrop parts, respectively.Also, 70% of the samples were used for classi cation, and the remaining 30% were employed to evaluate the accuracy of the classi cation results. 1 Where ρ NIR is the re ectance of near-infrared band and, ρ Red is the re ectance of red band.

2
Where i is the earliest scene and j is the last image acquired in a given four-months period.
Today, different classi cation methods are used for classifying satellite images.The SVM method is a non-parametric regression and classi cation method used by researchers, which does not have the limitations of parametric statistical methods.This method was introduced by Vapnik (1999), and today it has many applications in remote sensing (Mountrakis et al., 2011;Pourghasemi et al., 2020).This method tries to nd an optimal separating hyperplane that can separate classes (Kalantar et al., 2018).
The SVMs are applicable based on linear, polynomial, radial basis function (RBF), as well as sigmoid kernels, where the selection of each of these kernels affects the accuracy of the obtained outputs (Bag et al., 2022).In this research, image classi cation was done using the SVM algorithm and RBF kernel on GEE.Finally, to evaluate the accuracy of classi cation results using test samples, the kappa coe cient, overall accuracy, user accuracy, and producer accuracy were calculated.Finally, the classi ed maps in this stage were used to investigate the spatiotemporal LU/LCC in the 30-year period, as well as to estimate the cover management factor (C-factor) in the RUSLE model.

Estimation of Soil Erosion Rate Using RUSLE Model
In this study, the RUSLE model was used to calculate annual soil erosion.This model uses the following equation (Renard, 1997): Where A is the average annual soil loss (ton.ha − 1.y − 1 ), R denotes the rainfall erosivity factor (MJ.mm.ha − 1 .h− 1 .y− 1 ), K shows the soil erodibility factor (ton.h.MJ − 1 .mm− 1 ), LS re ects the slope length factor (unitless), C represents the cover management factor (unitless), and P is the support practice factor (unitless).The ve factors of the RUSLE model were calculated as follows.

R-factor
In this study, the R factor was calculated based on Fournier's index using the monthly and annual rainfall data of 17 rain gauge stations.The rain data were obtained from the statistics recorded by the Ministry of Energy and the Iranian Meteorological Organization during a 30-year period (1989-2019).
In order to calculate the R-factor, Fournier's index was applied based on the following equations (Renard and Ferreira, 1993): Where R represents the erosivity of rain (MJ.mm.ha − 1.h − 1 .y − 1 ), p i denotes the average rainfall (mm) in month i, p is the average annual rainfall (mm), and F shows the Fournier index.
The R-factor map of the Khorramabad watershed was prepared by interpolation of the R-factor values in the stations using the inverse distance weighting (IDW) method.The IDW interpolation method is based on the assumption that the estimated value of a point is more in uenced by known nearby points than distant points (Weber and Englund, 1992).The IDW method was chosen, since in this method the effect of the R-factor measured at the station points is considered very important and during the interpolation process weights are determined for the station points.Thus, as the distance from the point increases, the value of the R-factor decreases (Weber and Englund, 1994; Belasri and Lakhouili, 2016).For calculating the K factor, the equation proposed by Renard (1997) was used for limited data.This equation suggested by Römkens et al., (1997) for calculation of K-factor is as follows: In this study, the information related to the characteristics of soil granularity was used to provide the K factor map, which was prepared in previous studies by the General Department of Natural Resources, the Research and Agriculture and Natural Resources Center of Lorestan Province, and the Faculty of Agriculture and Natural Resources of Lorestan University.For this purpose, rst the value of K factor was calculated in the sampled points using Eq. 7.Then, the characteristics related to the variogram of the K factor values in different points were calculated in the GS + 9 software and its information was imported to the ArcGIS 10.8 software, with the corresponding map prepared using the Kriging method (

LS-factor
In this study, in order to prepare the LS factor map in Khorramabad watershed, the digital elevation model (DEM) map of Aster sensor with 30-meter pixel was employed.The LS-factor was prepared based on the method provided by Desmet and Govers, (1996) in the SAGA GIS 6 software.

C-factor
In the RUSLE model, the C-factor is usually determined based on empirical equations (Ochoa-cueva et al., 2015).According to past research, the vegetation map, and the conditions of the study area, the values of the C-factor, according to   The implementation process of this research is displayed in Fig. 3.

Investigation of LU/LCC Using GEE
Figure 4 depicts the LU/LC maps of 1989 and 2019, which were prepared through the classi cation of NDVI time series using the SVM method in GEE.The validation of the maps using test samples revealed the overall accuracy and kappa coe cient of 87.83% and 0.83 for 1989 and 86.51% and 0.82% for 2019 (Table 3).The investigation of the LU/LCC trend from 1989 to 2019 indicated that the area of man-made and agricultural areas increased by 2602.66 and 13303.35hectares, while the area of forests, rocks, and pastures decreased by 12584.3593,47.06, and 3274.02hectares.In other words, man-made and agricultural areas have grown by 77.34% and 28.39%, respectively, while forest, rock outcrop, and pasture have decreased by 17.40%, 0.85%, and 10.01%, respectively.According to the obtained results, most changes are related to agricultural lands and forests while the least changes are associated with rock outcrops (Table 4).Also, the investigation of the conversion rate of LU/LC showed that the highest conversion rate was related to the change of forests and pastures to agricultural lands (Table 5 and Fig. 5).

Estimation of RUSLE Model Factors
The estimation of the rain erosivity factor in 17 meteorological stations based on the Fournier index and the IDW interpolation method indicated that the value of the R-factor for the study area was between 98.50 and 292.89MJ.mm.ha − 1.h − 1 .y − 1 .The examination of this index in the watershed revealed that the highest value of this index was estimated in the northern, north-eastern, and southern regions of the watershed due to high altitudes and high rainfall.Also, the R-factor values have been lower in the central and western plains and lowlands where the amount of precipitation is less (Fig. 6a).The examination of the K-factor map, which was prepared using soil samples, indicated that the erodibility of the soil in the study area has been between 0.07 and 0.24 ton.h.MJ − 1 .mm− 1 (Fig. 6b).The LS-factor map shown in Fig. 6 indicates that the value of this factor in the study area has been between 0.03 and 38.14.The lower value of P-factor indicates that this factor plays a greater role in reducing water erosion (Fig. 6c).Based on the literature review, the value of the C-factor varies between 0.002 and 0.35, and when these values are closer to zero, it shows the desirable condition of vegetation management.Also, the value of P-factor in the studied study area was between 0.5 and 1(Fig.6d, e).

Estimation of Annual Soil Erosion Using RUSLE Model
After providing the map of RUSLE factors, the average annual soil erosion map was estimated for the years 1989 and 2019.The results of the statistical analysis of the erosion in these years are presented in Table 6.The results revealed that the estimated average extent of annual soil erosion in the watershed during 1989 and 2019 has been 15.48 and 20.41 (ton.ha − 1 .y − 1 ), respectively (Table 6).The results of the study of soil erosion changes during this 30-year period indicated that the average annual erosion rate has increased by 4.93 (ton.ha − 1 .y − 1 ).The results of the area and percentage of erosion classes indicated that in 1989, the erosion class of very low, low, medium, high, and very high covered 64.78%, 14.61%, 6.32%, 4.38%, and 9.91% of the Khorramabad watershed, respectively.Also, in 2019, very low, low, medium, high, and very high classes covered 60.91%, 14.54%, 6.84%, 4.88%, and 12.83% of the entire watershed, respectively (Table 7 and Fig. 7).According to the results, most changes are related to very low and very high erosion classes, while the least changes are associated with low erosion class (Table 7).Also, the investigation of the conversion rate soil erosion classes showed that the highest conversion rate was related to the change of very low erosion class to very high erosion class (Table 8 and Fig. 8).

Hot spots and cold spots analysis
The results of the hot spots analysis of soil erosion are presented in Fig. 9.The spatial clustering pattern showed that in both maps of 1989 and 2019, hot spots (99% con dence level) were distributed in the north-west to the south-east of the watershed.Also, the hot spots areas in 1989 and 2019 covered 19.2% and 21.7% of the Khorramabad watershed, respectively.These areas have the most vulnerability to soil erosion.On the other hand, cold spots are more widespread in the southwestern areas.Compared to other areas, these areas are the least vulnerable to soil erosion.Also, not-signi cant areas were scattered across the entire watershed between hot spots and cold spot areas.In not-signi cant areas, there is a possibility that they will soon become a hot spot area.The changes in the spots of soil erosion during 1990-2019 period showed that in 2019, compared to 1989, the area of hot spots increased by 90, 95 and 99% in all three levels.Although an increase can be seen in cold spots with a con dence level of 99% (6.2%) in 2019 compared to 1989, instead the area of cold spots has diminished at 95 and 90% levels (Table 9).

Discussion
Soil erosion is a complex environmental issue in different parts of the world, which has caused irreversible economic and environmental losses.One of the important factors affecting soil erosion rate is LU/LCC.A procedure to prepare LU/LCC map is to use satellite images.Researchers have used Landsat satellite images widely, especially in underdeveloped and developing countries, due to the free access to their images.In recent years, by launching the GEE online platform, Google has made it possible to access and process satellite images online using different servers worldwide.In this regard, the current research was conducted to examine the effect of LU/LCC on soil erosion in a semi-arid region using the GEE platform and the RUSLE model.

Investigation of LU/LCC using GEE
The LU/LC map was prepared using the time series classi cation of the NDVI index for each year in GEE.
Validation of the prepared maps revealed an overall accuracy value of more than 86% and a kappa coe cient of more than 0.82, which indicates the high accuracy of this method.In this regard,  The results of hot spots analysis showed the increasing trend of hot spots in 2019 compared to 1989.
This nding shows the increase of vulnerable areas to soil erosion and land destruction during this 30year period.However, cold spots have increased slightly in some areas, which indicates suitable LU/LC in these areas (Bagwan and Gavali, 2020).The investigation of the spatial distribution of the spots in 1989 and 2019 indicated that the hot spots were mostly located in steep areas and pasture lands, or in areas where the change from forests to pastures and agricultures has occurred.In contrast to cold spots, plain areas with a low slope and no change in LU/LC are more widespread.A signi cant part of the studied area is covered with non-signi cant areas.In non-signi cant areas, there is a possibility that they will soon become a hot spot area (Ranagalage et al., 2018;Dissanayake et al., 2019).Thus, in order to prevent deterioration of the erosion situation in these areas, it is necessary to pay special attention to soil protection operations.
Matching of LU/LCC maps to the map of soil erosion changes revealed that the change of forests to other uses, especially agricultural lands, had the greatest impact on the increase in erosion; the average annual erosion in the area related to forests in 1989 was 3.15 ton.ha − 1 .y− 1 , but this value within the same area in 2019 was 18.77 ton.ha − 1 .y− 1 .In other words, during a 30-year period, with a 17.40% reduction in the forests, soil erosion has grown by 15.62 ton.ha − 1 .y− 1 in the same area.When the forest is converted into other LU/LCs such as agriculture, the canopy cover of the trees is damaged, which acts as a protective layer for the soil against rain.Further, with the degradation of forests, the amount of soil organic matter and soil organisms also decreases, so that the soil permeability declines and as a result, the soil erosion increases.Ganasri and Ramesh (2016), El Jazouli et al., (2019), and Gong et al., (2022) pointed out that the degradation of forests and pastures as well as their conversion to other LU/LCs causes increased soil erosion.

Conclusion
research examined the effect of LU/LCC on the extent of soil erosion in a semi-arid region.LU/LC maps were prepared in a thirty-year period using a new approach and time series classi cation of NDVI spectral index of every year in GEE.Validation of the produced maps revealed that the utilized method is suitable for providing LU/LC maps with high accuracy.The study of the LU/LCC indicated an increase in the area of man-made areas and agricultural lands, while the area of forest, pasture, and rock outcrop areas decreased, with the highest rate of LU/LC conversion being linked to the conversion of forest to agricultural land.Estimation of erosion rate using RUSLE model in GIS environment showed that the average annual erosion in the region has increased by 4.93 ton.ha − 1 .y− 1 .The ndings of hot spots analysis also indicated increased vulnerable areas in the watershed.Matching the LU/LCC map to the soil erosion map revealed that the degradation of forests and pastures as well as their conversion to agricultural lands has had the greatest impact on the increase in soil erosion.Finally, it can be stated that GEE, as a platform that allows users to access the archive of satellite images and online image processing, has a high capability in preparing LU/LCC maps as well as other effective factors in soil erosion estimation models.The LU/LC conversion map.
Page 27/30 , Zurqani et al., (2018), Ghorbanian et al., (2020), Prasai et al., (2021), Ang et al., (2021), and Becker et al., (2021) have employed this cloud-based platform to prepare LU/LC maps.In order to estimate the soil erosion using satellite images and the RUSLE model, various studies have been conducted by researchers worldwide.Wang and Zhao (2020), estimated the amount of soil erosion in the Tahoe River area of China in 2005, 2010, 2015, and 2018 as 1424, 1195, 1129, 1099, and 1124 ton.ha − 1 .year − 1 , respectively.Enashar et al., (2021) estimated the average soil erosion in the middle, upper, and lower parts of the Blue Nile basin in Ethiopia as 39.73, 57.98, and 6.40 ton.ha − 1 .year − 1 , respectively.Petito et al., (2022) investigated the impact of conservation agriculture on soil erosion in the south of Italy and concluded that the area of soil erosion in the conservation agriculture system has diminished compared to traditional management.In all three mentioned studies, the RUSLE model and the GEE were used to estimate soil erosion.Also, researchers such as Alkharabsheh et al., (2013), Uddin et al., (2018), El Jazouli et al., (2019), and Gong et al., (2022) investigated the impact of LU/LCC on soil erosion in different areas using RUSLE model and satellite imagery.

7 8
Where, K is the soil erodibility factor (ton.h.MJ − 1 .mm− 1 ), m i denotes the average diameter of clay, silt and sand (mm), f i indicates the percentage of each component of silt, clay, and sand in the soil sample, and Dg represents the geometric mean diameter of the soil particles.

4. 2 .
The Effect of LU/LCC on Soil Erosion The average extent of annual soil erosion in the watershed for the years 1989 and 2019 was estimated as 15.48 and 20.41 ton.ha − 1 .y− 1 , respectively, which indicates an increase of 4.93 ton.ha − 1 .y− 1 .Investigating the spatial distribution of different erosion classes related to the years 1989 and 2019 revealed that classes with very low and low erosion risk are scattered throughout the entire region.The average erosion class is also more scattered in the central and north-western regions.Further, the classes with high and very high erosion risk are observed in the north, south-west, and central parts, which are affected by the large changes of LS, K and P-factors, showing the highest amount of soil erosion in these areas.In this regard, researchers such as Gupta and Kumar (2017), Uddin et al., (2018), Tilahum et al., (2018), and El Jazouli et al., (2019) reported the rising trend of soil erosion in their studies; in contrast, the research results of Alkharabsheh et al., (2013) showed descending trend of soil erosion, which seems to be the result of proper management of the watershed, conservation of forests and pastures, as well as afforestation in these areas.

Figure 7 Maps
Figure 7

Table 1
(Teng et al., 2016)h (1978)have presented Table2in order to estimate the P-factor through different slopes.As such, rst, using the DEM map, the slope map of the study area was prepared based on which the P-factor map was provided(Teng et al., 2016).

Table 2
maps were also classi ed into ve classes of very low, low, medium, high, and very high erosion based on the Natural break method.Note that the quantitative values of the range of erosion classes in each of the erosion intensity classes in both maps related to the years 1989 and 2019 were considered the same.2.2.2.7.Hot and Cold Spots Analysis of Soil Erosion in the Study AreaAccording to the method presented by Dissanayake et al., (2019), hot spots analysis was used for the spatial clustering pattern of soil loss for the years 1989 and 2019.This analysis was performed based on a 100×100 m grid in ArcGIS 10.8 using the tool box, optimized hotspot analysis (Getis-Ord Gi*).The average extent of soil erosion computed by the RUSLE model was calculated for each grid cell.The optimized hotspot analysis toolbox calculates the Gi* statistic, which represents the Z-score.Higher positive Z values indicate hot spots and lower negative Z values reveal cold spots.The z value determines the signi cance of clustering for a certain range based on the con dence level (ESRI, 2016 a, according to Eq. 3 and using the Raster calculator tool in the Arc GIS 10.8 software, the layers were multiplied together and the soil erosion map (A) was calculated.Finally, assuming that other factors of the RUSLE model are constant and the C-factor changes due to the effect of LU/LCC, the annual average soil loss map was prepared for the years 1989 and 2019.The nal soil loss b).

Table 3
Validation of satellite image classi cation results of 1989 and 2019.

Table 6
Values of average, maximum, and minimum soil erosion statistics in the study area.

Table 7
Area of annual soil erosion classes in the Khorramabad watershed.

Table 9
The area of hot and cold spots of erosion.
Ang et al., (2021)21)lso reported the overall accuracy and kappa coe cient of the LU/LCC map prepared using NDVI and GEE as 86.61% and 0.82, respectively.Prasai et al., (2021)also prepared the LU/LC map of Florida using GEE.They reported the overall accuracy and Kappa coe cient of 86% and 0.79, respectively.Also, Zurgani et al., (2018) reported an overall accuracy rate of 76-79% and a Kappa coe cient of 0.72 to 0.77 for LU/LC classi cation in a 16-year period.Note that in studies such asAng et al., (2021)andBecker et al., (2021), the overall accuracy and Kappa coe cient have been higher than the results of this study.It seems that factors such as the spatial and spectral resolution of satellite images, the type and number of LU/LC, the number of training samples, physiographic conditions, and classi cation algorithm affect the accuracy of output maps.Analysis of the LU/LCC trend indicated that compared to the initial area, man-made areas and agricultural lands have increased, while forest, rock outcrop, and pasture areas have decreased.Also, the results revealed that most changes were related to agricultural lands and forests while the least changes were linked to rock outcrop areas.The highest rate of conversion was related to the change of forests and pastures to agricultural lands.Degradation of forests and pastures in the region are mainly because of climatic changes, unauthorized cutting of trees for charcoal, res, and pests and diseases.Yet, the results of this study demonstrated that, one of the most important reasons for reduced area of forests and pastures is the cutting of trees and shrubs as well as the conversion of forests into agricultural lands.It seems that life issues and high unemployment rate among the people living in forests of the region have caused people to resort to illegal harvesting of these forests.In this regard, Naseri et al., (2019), Delpasand et al., (2022), Parma et al., (2017), and Vafaei et al., (2013) noted the reduction of forests and pastures as well as the increase of agricultural lands.In other regions of the world, researchers such as Khoi and Murayama (2011), Zurqani et al., (2018), Jazouli et al., (2019), Kumar et al., (2020), and Banyongha et al., (2020) have pointed out the reduction of forests in their studies.On the other hand, the results of Wang et al., (2021) and Gong et al., (2022) reported an increase in the area of forests, which seems to be due to enhanced level of implementation of appropriate protection and management measures as well as afforestation.