Prediction of soil organic carbon and the C:N ratio on a national scale using machine learning and satellite data: A comparison between Sentinel-2, Sentinel-3 and Landsat-8 images

Inversion of soil organic carbon content based on the two-point machine learning method

Soil organic carbon (SOC) is vital for the global carbon cycle and environmentally sustainable development. Meanwhile, the fast, convenient remote sensing technology has become one of the notable means to monitor SOC content. Nowadays, limitations are found in the inversion of SOC content with high-precision and complex spatial relationships based on scarce ground sample points. It is restrained by the spatial difference in the relationship between SOC content and remote sensing spectra due to the problem of different spectra for the same substance and the influence of topographic and environment (e.g. vegetation and climate). In this regard, the two-point machine learning (TPML) method, which can overcome above problems and deal with complex spatial heterogeneity of relationships between SOC and remote sensing spectra, is used to invert the SOC content in Hailun County, Heilongjiang Province, combined with derived variables from Sentinel-1, Sentinel-2, topography and environment. Based on 10-fold cross-validation and t-test, results indicate that the TPML method boasts the highest inversion accuracy, followed by random forest, gradient boosting regression tree, partial least squares regression and support vector machine. The average r, MAE, RMSE, and RPD of TPML are 0.854, 0.384 %, 0.558 %, and 1.918. Further, the TPML method has been proven to be equal to evaluating the uncertainty of inversion results, by comparing the actual and theoretical error of the inversion result in one subset. The spatial inversion result of SOC content with 10 m resolution by TPML is smoother and has more real details than other models, which are consistent with the distribution of SOC content in different land use types. This study provides both theoretical and technical guidance for using TPML method combined with spectral information of remote sensing to predict soil attributes and offer accurate uncertainty estimation, thereby opening up the opportunity for low-cost, high-precision, and large-scale SOC inversion.

Geospatial prediction of total soil carbon in European agricultural land based on deep learning

Accurate geospatial prediction of soil parameters provides a basis for large-scale digital soil mapping, making efficient use of the expensive and time-consuming process of field soil sampling. To date, few studies have used deep learning for geospatial prediction of soil parameters, but there is evidence that it may provide higher accuracy compared to machine learning methods. To address this research gap, this study proposed a deep neural network (DNN) for geospatial prediction of total soil carbon (TC) in European agricultural land and compared it with the eight most commonly used machine learning methods based on studies indexed in the Web of Science Core Collection. A total of 6209 preprocessed soil samples from the Geochemical mapping of agricultural and grazing land soil (GEMAS) dataset in heterogeneous agricultural areas covering 4,899,602 km2 in Europe were used. Prediction was performed based on 96 environmental covariates from climate and remote sensing sources, with extensive comprehensive hyperparameter tuning for all evaluated methods. DNN outperformed all evaluated machine learning methods (R2 = 0.663, RMSE = 9.595, MAE = 5.565), followed by Quantile Random Forest (QRF) (R2 = 0.635, RMSE = 25.993, MAE = 22.081). The ability of DNN to accurately predict small TC values and thus produce relatively low absolute residuals was a major reason for the higher prediction accuracy compared to machine learning methods. Climate parameters were the main factors in the achieved prediction accuracy, with 23 of the 25 environmental covariates with the highest variable importance being climate or land surface temperature parameters. These results demonstrate the superiority of DNN over machine learning methods for TC prediction, while highlighting the need for more recent soil sampling to assess the impact of climate change on TC content in European agricultural land.

Estimation of Soil Characteristics from Multispectral Sentinel-3 Imagery and DEM Derivatives Using Machine Learning

In this paper, different machine learning methodologies have been evaluated for the estimation of the multiple soil characteristics of a continental-wide area corresponding to the European region, using multispectral Sentinel-3 satellite imagery and digital elevation model (DEM) derivatives. The results confirm the importance of multispectral imagery in the estimation of soil properties and specifically show that the use of DEM derivatives improves the quality of the estimates, in terms of R2, by about 19% on average. In particular, the estimation of soil texture increases by about 43%, and that of cation exchange capacity (CEC) by about 65%. The importance of each input source (multispectral and DEM) in predicting the soil properties using machine learning has been traced back. It has been found that, overall, the use of multispectral features is more important than the use of DEM derivatives with a ration, on average, of 60% versus 40%.

Predicting Rice Lodging Risk from the Distribution of Available Nitrogen in Soil Using UAS Images in a Paddy Field

Rice lodging causes a loss of yield and leads to lower-quality rice. In Japan, Koshihikari is the most popular rice variety, and it has been widely cultivated for many years despite its susceptibility to lodging. Reducing basal fertilizer is recommended when the available nitrogen in soil (SAN) exceeds the optimum level (80-200 mg N kg^-1). However, many commercial farmers prefer to simultaneously apply one-shot basal fertilizer at transplant time. This study investigated the relationship between the rice lodging and SAN content by assessing their spatial distributions from unmanned aircraft system (UAS) images in a Koshihikari paddy field where one-shot basal fertilizer was applied. We analyzed the severity of lodging using the canopy height model and spatially clarified a heavily lodged area and a non-lodged area. For the SAN assessment, we selected green and red band pixel digital numbers from multispectral images and developed a SAN estimating equation by regression analysis. The estimated SAN values were rasterized and compiled into a 1 m mesh to create a soil fertility map. The heavily lodged area roughly coincided with the higher SAN area. A negative correlation was observed between the rice inclination angle and the estimated SAN, and rice lodging occurred even within the optimum SAN level. These results show that the amount of one-shot basal fertilizer applied to Koshihikari should be reduced when absorbable nitrogen (SAN + fertilizer nitrogen) exceeds 200 mg N kg^-1.

Effects of optical and radar satellite observations within Google Earth Engine on soil organic carbon prediction models in Spain

The modeling and mapping of soil organic carbon (SOC) has advanced through the rapid growth of Earth observation data (e.g., Sentinel) collection and the advent of appropriate tools such as the Google Earth Engine (GEE). However, the effects of differing optical and radar sensors on SOC prediction models remain uncertain. This research aims to investigate the effects of different optical and radar sensors (Sentinel-1/2/3 and ALOS-2) on SOC prediction models based on long-term satellite observations on the GEE platform. We also evaluate the relative impact of four synthetic aperture radar (SAR) acquisition configurations (polarization mode, band frequency, orbital direction and time window) on SOC mapping with multiband SAR data from Spain. Twelve experiments involving different satellite data configurations, combined with 4027 soil samples, were used for building SOC random forest regression models. The results show that the synthesis mode and choice of satellite images, as well as the SAR acquisition configurations, influenced the model accuracy to varying degrees. Models based on SAR data involving cross-polarization, multiple time periods and "ASCENDING" orbits outperformed those involving copolarization, a single time period and "DESCENDING" orbits. Moreover, combining information from different orbital directions and polarization modes improved the soil prediction models. Among the SOC models based on long-term satellite observations, the Sentinel-3-based models (R² = 0.40) performed the best, while the ALOS-2-based model performed the worst. In addition, the predictive performance of MSI/Sentinel-2 (R² = 0.35) was comparable with that of SAR/Sentinel-1 (R² = 0.35); however, the combination (R² = 0.39) of the two improved the model performance. All the predicted maps involving Sentinel satellites had similar spatial patterns that were higher in northwest Spain and lower in the south. Overall, this study provides insights into the effects of different optical and radar sensors and radar system parameters on soil prediction models and improves our understanding of the potential of Sentinels in developing soil carbon mapping.

Prediction of the concentration of cadmium in agricultural soil in the Czech Republic using legacy data, preferential sampling, Sentinel-2, Landsat-8, and ensemble models

The current study assesses and predicts cadmium (Cd) concentration in agricultural soil using two Cd datasets, namely legacy data (LD) and preferential sampling-legacy data (PS-LD), along with four streams of auxiliary datasets extracted from Sentinel-2 (S2) and Landsat-8 (L8) bands. The study was divided into two contexts: Cd prediction in agricultural soil using LD, ensemble models, 10 and 20 m spatial resolution of S2 and L8 (context 1), and Cd prediction in agricultural soil using PS-LD, ensemble models and 10 and 20 m spatial resolution of S2 and L8 (context 2). In context 1, ensemble 1, L8 with PS-LD was the cumulative optimal approach that predicted Cd in agricultural soil with a higher R² value of 0.76, root mean square error (RMSE) of 0.66, mean absolute error (MAE) of 0.35, and median absolute error (MdAE) of 0.13. However, with R² = 0.78, RMSE = 0.63, MAE = 0.34, and MdAE = 0.15, ensemble 1, S2 of PS-LD was the best prediction approach in predicting Cd concentration in agricultural soil in context 2. Overall, the predictions from both contexts indicated that ensemble 1 of S2 combined with PS-LD was the most appropriate and best model for Cd prediction in agricultural soil. The modeling approaches' uncertainty in both contexts was assessed using ensemble-sequential gaussian simulation (EnSGS), which revealed that the degree of uncertainty propagated in the study area was within 5% in both contexts. The combination of the PS dataset and the LD along with ensemble models and the remote sensing dataset, produced promising results. Nonetheless, the results demonstrated that the 20 m spatial resolution band dataset used in the prediction of Cd in agricultural soil outperformed the 10 m spatial resolution. When PS is combined with LD, an appropriate modeling approach, and a well-correlated remote sensing dataset are used, good results are obtained.

Digital mapping of soil pH and carbonates at the European scale using environmental variables and machine learning

Soil pH and carbonates (CaCO₃) are important indicators of soil chemistry and fertility, and the prediction of their spatial distribution is critical for the agronomic and environmental management. Digital soil mapping (DSM) techniques are widely accepted for the geospatial analysis of the soil properties. They are rapid and cost-efficient approaches that can provide quantitative prediction. However, the digital mapping of soil pH and CaCO₃ are not well studied, especially at a continental scale. In this research, we mapped the soil pH and CaCO₃ at the European scale using multisource environmental variables and machine learning approaches. Moderate Resolution Imaging Spectroradiometer (MODIS) products, terrain attributes, and climatic variables were considered. Meanwhile, nine machine learning algorithms, namely, three linear and six nonlinear models, were used for the spatial prediction of soil pH and CaCO₃. The land use and cover area frame statistical survey (LUCAS) 2015 topsoil dataset provided by the European Soil Data Centre was utilised. The performances of different models were compared and analysed in terms of coefficient of determination (R²), root mean square error (RMSE), and ratio of performance to deviation (RPD). Specifically, nonlinear machine learning models outperformed the linear ones, and extremely randomized trees (ERT) gave the most satisfactory result for soil pH (R² = 0.70, RMSE = 0.75, and RPD = 1.84) and CaCO₃ (R² = 0.53, RMSE = 93.49 g/kg, and RPD = 1.46). The results revealed that MODIS products and climatic variables were important in predicting soil pH and CaCO₃. Moreover, spatial distribution of soil pH and CaCO₃ in Europe were mapped at 250 m resolution, and the areas with high CaCO₃ content always showed high soil pH value.

Oil spills: Detection and concentration estimation in satellite imagery, a machine learning approach

The method's development to detect oil-spills, and concentration monitoring of marine environments, are essential in emergency response. To develop a classification model, this work was based on the spectral response of surfaces using reflectance data, and machine learning (ML) techniques, with the objective of detecting oil in Landsat imagery. Additionally, different concentration oil data were used to obtain a concentration-estimation model. In the classification, K-Nearest Neighbor (KNN) obtained the best approximations in oil detection using Blue (0.453-0.520 μm), NIR (0.790-0.891 μm), SWIR1 (1.557-1.717 μm), and SWIR2 (1.960-2.162 μm) bands for 2010 spill images. In the concentration model, the mean absolute error (MAE) was 1.41 and 3.34, for training and validation data. When testing the concentration-estimation model in images where oil was detected, the concentration-estimation obtained was between 40 and 60 %. This demonstrates the potential use of ML techniques and spectral response data to detect and estimate the concentration of oil-spills.

ALOS-2 L-band SAR backscatter data improves the estimation and temporal transferability of wildfire effects on soil properties under different post-fire vegetation responses

Remote sensing techniques are of particular interest for monitoring wildfire effects on soil properties, which may be highly context-dependent in large and heterogeneous burned landscapes. Despite the physical sense of synthetic aperture radar (SAR) backscatter data for characterizing soil spatial variability in burned areas, this approach remains completely unexplored. This study aimed to evaluate the performance of SAR backscatter data in C-band (Sentinel-1) and L-band (ALOS-2) for monitoring fire effects on soil organic carbon and nutrients (total nitrogen and available phosphorous) at short term in a heterogeneous Mediterranean landscape mosaic made of shrublands and forests that was affected by a large wildfire. The ability of SAR backscatter coefficients and several band transformations of both sensors for retrieving soil properties measured in the field in immediate post-fire situation (one month after fire) was tested through a model averaging approach. The temporal transferability of SAR-based models from one month to one year after wildfire was also evaluated, which allowed to assess short-term changes in soil properties at large scale as a function of pre-fire plant community type. The retrieval of soil properties in immediate post-fire conditions featured a higher overall fit and predictive capacity from ALOS-2 L-band SAR backscatter data than from Sentinel-1 C-band SAR data, with the absence of noticeable under and overestimation effects. The transferability of the ALOS-2 based model to one year after wildfire exhibited similar performance to that of the model calibration scenario (immediate post-fire conditions). Soil organic carbon and available phosphorous content was significantly higher one year after wildfire than immediately after the fire disturbance. Conversely, the short-term change in soil total nitrogen was ecosystem-dependent. Our results support the applicability of L-band SAR backscatter data for monitoring short-term variability of fire effects on soil properties, reducing data gathering costs within large and heterogeneous burned landscapes.

Estimation of Soil Organic Carbon Content in the Ebinur Lake Wetland, Xinjiang, China, Based on Multisource Remote Sensing Data and Ensemble Learning Algorithms

Soil organic carbon (SOC), as the largest carbon pool on the land surface, plays an important role in soil quality, ecological security and the global carbon cycle. Multisource remote sensing data-driven modeling strategies are not well understood for accurately mapping soil organic carbon. Here, we hypothesized that the Sentinel-2 Multispectral Sensor Instrument (MSI) data-driven modeling strategy produced superior outcomes compared to modeling based on Landsat 8 Operational Land Imager (OLI) data due to the finer spatial and spectral resolutions of the Sentinel-2A MSI data. To test this hypothesis, the Ebinur Lake wetland in Xinjiang was selected as the study area. In this study, SOC estimation was carried out using Sentinel-2A and Landsat 8 data, combining climatic variables, topographic factors, index variables and Sentinel-1A data to construct a common variable model for Sentinel-2A data and Landsat 8 data, and a full variable model for Sentinel-2A data, respectively. We utilized ensemble learning algorithms to assess the prediction performance of modeling strategies, including random forest (RF), gradient boosted decision tree (GBDT) and extreme gradient boosting (XGBoost) algorithms. The results show that: (1) The Sentinel-2A model outperformed the Landsat 8 model in the prediction of SOC contents, and the Sentinel-2A full variable model under the XGBoost algorithm achieved the best results R² = 0.804, RMSE = 1.771, RPIQ = 2.687). (2) The full variable model of Sentinel-2A with the addition of the red-edge band and red-edge index improved R² by 6% and 3.2% over the common variable Landsat 8 and Sentinel-2A models, respectively. (3) In the SOC mapping of the Ebinur Lake wetland, the areas with higher SOC content were mainly concentrated in the oasis, while the mountainous and lakeside areas had lower SOC contents. Our results provide a program to monitor the sustainability of terrestrial ecosystems through a satellite perspective.

Simulation of Soil Organic Carbon Content Based on Laboratory Spectrum in the Three-Rivers Source Region of China

Soil organic carbon (SOC) changes affect the land carbon cycle and are also closely related to climate change. Visible-near infrared spectroscopy (Vis-NIRS) has proven to be an effective tool in predicting soil properties. Spectral transformations are necessary to reduce noise and ensemble learning methods can improve the estimation accuracy of SOC. Yet, it is still unclear which is the optimal ensemble learning method exploiting the results of spectral transformations to accurately simulate SOC content changes in the Three-Rivers Source Region of China. In this study, 272 soil samples were collected and used to build the Vis-NIRS simulation models for SOC content. The ensemble learning was conducted by the building of stack models. Sixteen combinations were produced by eight spectral transformations (S-G, LR, MSC, CR, FD, LRFD, MSCFD and CRFD) and two machine learning models of RF and XGBoost. Then, the prediction results of these 16 combinations were used to build the first-step stack models (Stack1, Stack2, Stack3). The next-step stack models (Stack4, Stack5, Stack6) were then made after the input variables were optimized based on the threshold of the feature importance of the first-step stack models (importance > 0.05). The results in this study showed that the stack models method obtained higher accuracy than the single model and transformations method. Among the six stack models, Stack 6 (5 selected combinations + XGBoost) showed the best simulation performance (RMSE = 7.3511, R2 = 0.8963, and RPD = 3.0139, RPIQ = 3.339), and obtained higher accuracy than Stack3 (16 combinations + XGBoost). Overall, our results suggested that the ensemble learning of spectral transformations and simulation models can improve the estimation accuracy of the SOC content. This study can provide useful suggestions for the high-precision estimation of SOC in the alpine ecosystem.

Estimating Carbon, Nitrogen, and Phosphorus Contents of West–East Grassland Transect in Inner Mongolia Based on Sentinel-2 and Meteorological Data

Estimating the carbon (C), nitrogen (N), and phosphorus (P) contents of a large-span grassland transect is essential for evaluating ecosystem functioning and monitoring biogeochemical cycles. However, the field measurements are scattered, such that they cannot indicate the continuous gradient change in the grassland transect. Although remote sensing methods have been applied for the estimation of nutrient elements at the local scale in recent years, few studies have considered the effective estimation of C, N, and P contents over large-span grassland transects with complex environment including a variety of grassland types (i.e., meadow, typical grassland, and desert grassland). In this paper, an information enhancement algorithm (involving spectral enhancement, regional enhancement, and feature enhancement) is used to extract the weak information related to C, N, and P. First, the spectral simulation algorithm is used to enhance the spectral information of Sentinel-2 imagery. Then, the enhanced spectra and meteorological data are fused to express regional characteristics and the fractional differential (FD) algorithm is used to extract sensitive spectral features related to C, N, and P, in order to construct a partial least-squares regression (PLSR) model. Finally, the C, N, and P contents are estimated over a West–East grassland transect in Inner Mongolia, China. The results demonstrate that: (i) the contents of C, N, and P in large-span transects can be effectively estimated through use of the information enhancement method involving spectral enhancement, regional feature enhancement, and information enhancement, for which the estimation accuracies (R2) were 0.88, 0.78, and 0.85, respectively. Compared with the estimation results of raw Sentinel-2 imagery, the RMSE was reduced by 3.42 g/m2, 0.14 g/m2, and 13.73 mg/m2, respectively; and (ii) the continuous change trend and spatial distribution characteristics of C, N, and P contents in the west–east transect of the Inner Mongolia Plateau were obtained, which showed decreasing trends in C, N, and P contents from east to west and the characteristics of meadow > typical grassland > desert grassland. Thus, the information enhancement algorithm can help to improve estimates of C, N, and P contents when considering large-span grassland transects.

Using Sentinel-2 Images for Soil Organic Carbon Content Mapping in Croplands of Southwestern France. The Usefulness of Sentinel-1/2 Derived Moisture Maps and Mismatches between Sentinel Images and Sampling Dates

In agronomy, soil organic carbon (SOC) content is important for the development and growth of crops. From an environmental monitoring viewpoint, SOC sequestration is essential for mitigating the emission of greenhouse gases into the atmosphere. SOC dynamics in cropland soils should be further studied through various approaches including remote sensing. In order to predict SOC content over croplands in southwestern France (area of 22,177 km²), this study addresses (i) the influence of the dates on which Sentinel-2 (S2) images were acquired in the springs of 2017–2018 as well as the influence of the soil sampling period of a set of samples collected between 2005 and 2018, (ii) the use of soil moisture products (SMPs) derived from Sentinel-1/2 satellites to analyze the influence of surface soil moisture on model performance when included as a covariate, and (iii) whether the spatial distribution of SOC as mapped using S2 is related to terrain-derived attributes. The influences of S2 image dates and soil sampling periods were analyzed for bare topsoil. The dates of the S2 images with the best performance (RPD ≥ 1.7) were 6 April and 26 May 2017, using soil samples collected between 2016 and 2018. The soil sampling dates were also analyzed using SMP values. Soil moisture values were extracted for each sample and integrated into partial least squares regression (PLSR) models. The use of soil moisture as a covariate had no effect on the prediction performance of the models; however, SMP values were used to select the driest dates, effectively mapping topsoil organic carbon. S2 was able to predict high SOC contents in the specific soil types located on the old terraces (mesas) shaped by rivers flowing from the southwestern Pyrénées.

Use of Multi-Seasonal Satellite Images to Predict SOC from Cultivated Lands in a Montane Ecosystem

Although algorithms are well developed to predict soil organic carbon (SOC), selecting appropriate covariates to improve prediction accuracy is an ongoing challenge. Terrain attributes and remote sensing data are the most common covariates for SOC prediction. This study tested the predictive performance of nine different combinations of topographic variables and multi-season remotely sensed data to improve the prediction of SOC in the cultivated lands of a middle mountain catchment of Nepal. The random forest method was used to predict SOC contents, and the quantile regression forest for quantifying the prediction uncertainty. Prediction of SOC contents was improved when remote sensing data of multiple seasons were used together with the terrain variables. Remote sensing data of multiple seasons capture the dynamic conditions of surface soils more effectively than using an image of a single season. It is concluded that the use of remote sensing images of multiple seasons instead of a snapshot of a single period may be more effective for improving the prediction of SOC in a digital soil mapping framework. However, an image with the right timing of cropping season can provide comparable results if a parsimonious model is preferred.

Spatial variations in soil organic carbon, nitrogen, phosphorus contents and controlling factors across the "Three Rivers" regions of southwest China

Based on the data of China Geochemical Baselines project, geostatistical analysis was used to investigate the spatial variations in soil organic carbon (SOC), total nitrogen (TN) and total phosphorus (TP) contents across the "Three Rivers" regions of southwest China, and the factors affecting them were analyzed by the redundancy analysis (RDA) and Person's correlation. Results showed that, the average content of SOC, TN and TP in the study area were 7.20 g/kg, 0.84 g/kg and 0.55 g/kg, respectively. The SOC and TN contents showed an obvious enrichment characteristic with great spatial variability, while TP content was stable on regional scale. The SOC, TN and TP contents decreased with elevation increase in the northern frigid highland, but showed an opposite character in the southern tropical & subtropical, which actually reflected the control of temperature on them. Combined with that there were higher SOC, TN and TP contents in subalpine meadow soil and red earth-yellow earth of the middle latitude zone, these suggested that the suitable temperature was conducive to the accumulation of soil nutrients. The weak positive correlation between population density and soil nutrients, together with high level of soil nutrients in the vicinity of large cities, demonstrated that human activities had significantly increased the soil nutrients contents in the study area. The RDA results showed that soil nutrients in the northern frigid highland were mainly controlled by the environmental factors dominated by temperature and soil structural factors dominated by parent materials with the total explanatory power as high as 75.87%, while in the southern tropical & subtropical mainly by the environmental factors dominated by chemical and biological weathering and the biological factors dominated by vegetation with the total explanatory power as high as 88.53%. The above factors superimposing a certain degree of human activities converged to cause that the SOC and TP contents in the south were higher than that in the north while the TN content was lower than that in the north.

Spatial heterogeneity modeling of water quality based on random forest regression and model interpretation

A systematic understanding of the spatial distribution of water quality is critical for successful watershed management; however, the limited number of physical monitoring stations has restricted the evaluation of spatial water quality distribution and the identification of features impacting the water quality. To fill this gap, we developed a modeling process that employed the random forest regression (RFR) to model the water quality distribution for the Taihu Lake basin in Zhejiang Province, China, and adopted the Shapley Additive exPlanations (SHAP) method to interpret the underlying driving forces. We first used RFR to model three water quality parameters: permanganate index (COD_Mn), total phosphorus (TP), and total nitrogen (TN), based on 16 watershed features. We then applied the built models to generate water quality distribution maps for the basin, with the COD_Mn ranging from 1.39 to 6.40 mg/L, TP from 0.02 to 0.23 mg/L, and TN from 1.43 to 4.27 mg/L. These maps showed generally consistent patterns among the COD_Mn, TN, and TP with minor differences in the spatial distribution. The SHAP analysis showed that the TN was mainly affected by agricultural non-point sources, while the COD_Mn and TP were affected by agricultural and domestic sources. Due to differences in sewage collection and treatment between urban and rural areas, the water quality in highly populated urban areas was better than that in rural areas, which led to an unexpected positive relationship between water quality and population density. Overall, with the RFR models and SHAP interpretation, we obtained a continuous distribution pattern of the water quality and identified its driving forces in the basin. These findings provided important information to assist water quality restoration projects.

Prediction of Soil Organic Carbon under Different Land Use Types Using Sentinel-1/-2 Data in a Small Watershed

Soil organic carbon (SOC) is a key property for evaluating soil quality. SOC is thus an important parameter of agricultural soils and needs to be regularly monitored. The aim of this study is to explore the potential of synthetic aperture radar (SAR) satellite imagery (Sentinel-1), optical satellite imagery (Sentinel-2), and digital elevation model (DEM) data to estimate the SOC content under different land use types. The extreme gradient boosting (XGboost) algorithm was used to predict the SOC content and evaluate the importance of feature variables under different land use types. For this purpose, 290 topsoil samples were collected and 49 features were derived from remote sensing images and DEM. Feature selection was carried out to prevent data redundancy. Coefficient of determination (R2), mean absolute error (MAE), mean squared error (MSE), percent root mean squared error (%RMSE), ratio of performance to interquartile range (RPIQ), and corrected akaike information criterion (AICc) were employed for evaluating model performance. The results showed that Sentinel-1 and Sentinel-2 data were both important for the prediction of SOC and the prediction accuracy of the model differed with land use types. Among them, the prediction accuracy of this model is the best for orchard (R2 = 0.86 and MSE = 0.004%), good for dry land (R2 = 0.74 and MSE = 0.008%) and paddy field (R2 = 0.66 and MSE = 0.009%). The prediction model of SOC content is effective and can provide support for the application of remote sensing data to soil property monitoring.

Digital Mapping of Soil Organic Carbon Using Sentinel Series Data: A Case Study of the Ebinur Lake Watershed in Xinjiang

As an important evaluation index of soil quality, soil organic carbon (SOC) plays an important role in soil health, ecological security, soil material cycle and global climate cycle. The use of multi-source remote sensing on soil organic carbon distribution has a certain auxiliary effect on the study of soil organic carbon storage and the regional ecological cycle. However, the study on SOC distribution in Ebinur Lake Basin in arid and semi-arid regions is limited to the mapping of measured data, and the soil mapping of SOC using remote sensing data needs to be studied. Whether different machine learning methods can improve prediction accuracy in mapping process is less studied in arid areas. Based on that, combined with the proposed problems, this study selected the typical area of the Ebinur Lake Basin in the arid region as the study area, took the sentinel data as the main data source, and used the Sentinel-1A (radar data), the Sentinel-2A and the Sentinel-3A (multispectral data), combined with 16 kinds of DEM derivatives and climate data (annual average temperature MAT, annual average precipitation MAP) as analysis. The five different types of data are reconstructed by spatial data and divided into four spatial resolutions (10, 100, 300, and 500 m). Seven models are constructed and predicted by machine learning methods RF and Cubist. The results show that the prediction accuracy of RF model is better than that of Cubist model, indicating that RF model is more suitable for small areas in arid areas. Among the three data sources, Sentinel-1A has the highest SOC prediction accuracy of 0.391 at 10 m resolution under the RF model. The results of the importance of environmental variables show that the importance of Flow Accumulation is higher in the RF model and the importance of SLOP in the DEM derivative is higher in the Cubist model. In the prediction results, SOC is mainly distributed in oasis and regions with more human activities, while SOC is less distributed in other regions. This study provides a certain reference value for the prediction of small-scale soil organic carbon spatial distribution by means of remote sensing and environmental factors.