Search

• The selection of satellite sensor and radar system parameters greatly affected the model output. • The model was improved when more polarizations, orbital directions and frequencies were involved. • Models built using multiband radar datasets achieved a comparable accuracy to models based on optical data. • The model built by fusing SAR-optical data achieved better results than the model without SAR data. • The development of this continental-scale DSM work largely benefits from GEE. The advent of cloud computing platforms (e.g., Google Earth Engine (GEE)) and the massive amounts of optical and radar Earth Observation (EO) data hosted by these platforms present new opportunities for mapping soil pH at large scales. However, existing studies generally lack consensus on the effects of satellite and radar sensor parameters on GEE-based soil pH prediction models. In this study, we assessed the suitability of long-term radar (C-band Sentinel-1 and L-band PALSAR-1/2) and optical (Sentinel-2) EO data on GEE for the digital mapping of soil pH on a continental (Europe) scale and determined the most appropriate radar sensor parameters. Thirteen scenarios with different data configurations were simulated and combined with the 2018 LUCAS soil database and two machine learners (boosted regression trees and extreme gradient boosting) to develop soil prediction models. Results showed that the selection of modeling techniques, satellite sensors and radar system parameters largely affected the model output. Models involving a single polarization mode of PALSAR-1/2 data performed the worst (RPD = 1.24). Models based on Sentinel-1 data performed better than those built using PALSAR-1/2 data. The model performance was improved when a model involved more polarization bands, orbital directions, and band frequencies. The multiband model built using the two radar datasets achieved a comparable accuracy to the model based on optical data. Moreover, the model that fused radar-optical data achieved better results, with RPD values of 1.56 and 1.46 for the models with and without radar data, respectively; its performance was comparable to that of models built with commonly used variables (topography and climate). The analysis of importance indicated that long-term optical and radar EO data on GEE were important in our model. The modelling of soil pH at the continental scale largely benefits from GEE. The predicted maps exhibited strong spatial heterogeneity among different biogeographic regions, with similar spatial patterns under different modelling scenarios. [ABSTRACT FROM AUTHOR]

• SOC inversion models were established for various land-use in an arid mining area. • SOC spatiotemporal variability and its effect factors were analyzed. • Changes in vegetation and soil erosion dominate the spatial distribution of SOC. • Coal mining affects plant, terrain and soil erosion, then weakens SOC accumulation. Soil organic carbon (SOC) undergoes rapid changes due to human production activities, which have an impact on the land carbon cycle and ultimately global change. As one of the main human production activities, coal mining significantly impacts the soil carbon cycle. However, due to the lack of remote sensing modeling of soil carbon in mining areas, the spatio-temporal changes and driving mechanisms of SOC in mining areas remain unclear. Therefore, this study investigated and determined SOC data from 300 sampling points (depth of 0–20 cm) located in an arid mining area of China. Remote sensing images were then used to established a soil organic carbon density (SOCD) prediction model within the Random Forest (RF) model to achieve digital mapping of soil organic carbon stocks (SOCS). The spatiotemporal changes of SOCS were analyzed using SOCS digital mapping, and the influencing mechanism of SOCS was revealed using path analysis. The results showed that the constructed SOCD predictive model meets the demand for SOCD prediction (R2 ≥ 0.74, p < 0.01, RMSE ≤ 1.72 kg/m2). Under the combined influence of coal mining and land reclamation, the total amount of surface SOCS in the mining area exhibited a fluctuating upward trend from 1990 (6.34 Tg) to 2021 (7.73 Tg), with an annual growth rate of 0.038 Tg/a. The spatial distribution of SOCS generally increased from southeast to northwest. Precipitation, Normalized Difference Vegetation Index (NDVI), and land use were positively correlated to SOCS spatial distribution, while temperature, elevation, soil erosion, and mining intensity were negatively correlated to SOCS. The impact degree of factors on SOCS was as follows: NDVI > soil erosion > mining intensity > precipitation > elevation > land use > temperature. The negative impact of coal mining on SOCS was mainly indirect, through disturbance to elevation, vegetation, and soil erosion. The uneven ground subsidence and stretching caused by coal mining contribute to intensified soil erosion and vegetation degradation in the affected area, leading to a reduction in SOCS. However, SOCS did not decrease under high intensity mining, which was related to the increase in vegetation and the reduction in soil erosion in the mining area. In this study, a soil carbon prediction model was established based on remote sensing modeling to evaluate the temporal and spatial distribution of soil carbon in an arid mining area. The results can serve as valuable references for the scientific improvement of the ecological environment in mining areas, the rational planning of mining area construction, as well as low-carbon land reclamation and ecological compensation assessments. [ABSTRACT FROM AUTHOR]

• RFR with climate classification is an effective method to predict global soil δ15N. • Key ecological limits on soil δ15N in 5 climate zones were respectively identified. • The global map of soil δ15N at 0.1° resolution was reliably generated. Spatial information of the natural abundance of soil nitrogen stable isotope (δ15N) is beneficial for deeply understanding the terrestrial nitrogen (N) cycle. However, to date, the precise global map of soil δ15N still lacks. In this study, based on the measured soil δ15N data by Craine et al. (2015) (https://doi.org/10.1038/srep08280) and related environmental variables including soil, topography, vegetation, and climate, we constructed the optimal relationship model between soil δ15N and environmental variables, and mapped the global pattern of soil δ15N at 0.1° × 0.1° resolution (in natural terrestrial ecosystem). Results indicated that satisfied performance was achieved (R2 = 0.68 and RMSE = 1.26‰) by separately building the optimal relationship models for soil δ15N in each of five climate zones (Tropical, Arid, Temperate, Cold and Polar) using the random forest regression algorithm. In addition, critical controls of the soil δ15N in different climate zones were thus identified based on the variable importance calculated by each random forest regression model. In the Tropical zone, soil δ15N might be primarily regulated by microbial N loss, and soil pH and organic matter were identified as two most important factors. In the Arid zone, abiotically gaseous N loss regulated by solar radiation would be the critical controls of soil δ15N. In the Temperate zone, temperature-related variables were identified as the critical controlling factors, and in the Cold zone, soil water and heat conditions had the equally greater importance, and bulk density was the dominated factor in the Polar zone. Furthermore, the predicted global soil δ15N ranged from −0.44‰ to 12.59‰, with the mean value of 5.06‰, and the standard deviation of 1.74‰. Significantly higher soil δ15N (P < 0.05) were observed in the Tropical and Arid zones with mean values of 6.52‰ and 6.11‰, respectively. This indicated that the soil N cycles were more open than those in the Temperate, Cold and Polar zones (mean soil δ15N of 4.37‰, 3.67‰ and 2.76‰, respectively). This study provides clues for potential environmental regulations on terrestrial N cycle in different climates, and the global soil δ15N map can be a reliable data support for future research. [ABSTRACT FROM AUTHOR]

• Mapping the distribution of heavy metals in soil using multiple spatial data. • The spatial scene is a crucial covariate in estimating soil heavy metals. • Heavy metals accumulate most severely in industrial and residential soils. • The impact distance of various spatial scenes on heavy metals is different. Monitoring the spatial distribution and sources of heavy metals (HM) in soil is essential for avoiding health risks and achieving sustainable soil utilization. Multiple geospatial data, including remote sensing, climate, soil and topography data, were used to extract environmental covariates. Additionally, the spatial scene was employed as the alternative data of land use/land cover to describe the urban functions and human activity intensity in more detail. After converting to a uniform resolution of 30 m, these environmental covariates were adopted to characterize several common HM in soil, including copper (Cu), chromium (Cr), lead (Pb), nickel (Ni), and zinc (Zn). The RReliefF algorithm was used to identify several important variables. The quantification models of HM were established using back-propagation neural network (BPNN) and deep neural network (DNN). Besides, the impact of distance from the spatial scenes on HM were analyzed. The result demonstrated that the spatial scene is a key environmental covariate in estimating HM in soil. Compared with BPNN, the DNN model provided better accuracy (R2 = 0.67–0.75) for estimation of five HM elements. Therefore, the DNN model was used to map HM concentrations at a grid scale of 30 m. The spatial scenes with the highest risk of HM pollution are industrial areas, residential areas, road, and commercial areas, and the concentration of HM is negatively correlated with the distance from these spatial scenes. The effective impact distances of industrial and residential areas are about 2000 m, and the effective impact distances of road and commercial areas are 500 m. [ABSTRACT FROM AUTHOR]

Highlights • A MGWRK technique was introduced to map soil electrical conductivity concentration. • MGWRK considers both spatial dependence and local and global relationships. • MGWRK outperformed MGWR, GWRK, GWR, RK, and MLR based on cross validation. • MGWRK is a promising method for digital mapping of soil EC in complex environments. Abstract Spatial prediction is an important approach to obtain location-specific values of soil electrical conductivity (EC), which is a proxy of soil salinity and important for agricultural management in arid and semi-arid areas. Linear regression models assume that the relation between soil EC and environmental covariates is constant over the area to be predicted. This is problematic at the regional scale, at which some of the regression parameters may indeed be globally constant, whereas others may vary locally. Moreover, model residuals often exhibit spatial dependence, which invalidates the ordinary least squares linear regression. This study examined the combination of a mixed geographically weighted regression model with simple kriging of the residuals (MGWGK) for mapping soil EC in the Heihe River Basin, an inland river basin in arid northwest China. We compared the performance of MGWRK with those of multiple linear regression (MLR), regression kriging (RK), geographically weighted regression (GWR), geographically weighted regression kriging (GWRK) and mixed geographically weighted regression (MGWR). Environmental covariates were developed from spatial information on topography, climate, vegetation, and geographic position. A ten-fold cross-validation was applied to evaluate predictive accuracy of the various methods. Soil EC ranged from 0.031 to 182.100 dS m−1, exhibiting a contrasting distribution of soil EC in the upper, middle, and lower river reaches. The MGWRK method outperformed other methods. The effects of different environmental covariates on soil EC were revealed by the fixed and geographically varying parameters of MGWRK. The nugget-to-sill ratios of fitted variogram models all fell between 25 and 32%, exhibiting moderate spatial autocorrelation of models residuals. Predictive accuracy was improved by MGWRK, as the spatial dependence of model residuals were included in the prediction. When selecting an optimal linear regression model, covariates should be tested to see if they are constant (as in MLR) or spatially varying (as in GWR) or semi-varying (as in MGWR), and model residuals should be tested for spatial dependence (as in MGWRK). [ABSTRACT FROM AUTHOR]

Highlights • Kriging methods are the most accurate and RF provides reliable confidence intervals. • BRT yielded less accurate but more precise confidence intervals. • The local uncertainty must be mapped robustly or it will have little practical value. • ML algorithms can be modified to compute the local uncertainty. • Algorithms differ in their approaches to computing uncertainty, which should be considered. Abstract In recent years, the environmental modeling community has moved away from kriging as the main mapping algorithm and embraced machine learning (ML) as the go-to method for spatial prediction. The drawback of this shift has been a gradual decline in the number of papers in which uncertainty is presented and mapped alongside estimates of the target variables because in some ML algorithms, computing the local uncertainty can be challenging. This drawback has been recently identified in the literature as one of the key areas in DSM where progress is most needed. The main objective of this work is to compare geostatistical techniques, ML methods and hybrid methods, e.g., regression kriging, in terms of not only their overall accuracy but also their precision in providing useful confidence intervals at unsampled locations. We aim to provide clear application guidelines for future mapping exercises. For this experiment, we used a legacy soil dataset (n = 414) of topsoil observations from the semi-arid Mediterranean region of Sicily. This dataset was collected in a 2008 survey with a pedo-landscape sampling design; hence, it is ideal for comparing geostatistics and ML. In the comparison, we included algorithms that have been widely adopted in the literature: ordinary and universal kriging, linear regression, random forest (RF), quantile regression forest, boosted regression trees (BRT) and hybrid forms of kriging (e.g., regression kriging with RF and BRT used as regressors). To evaluate the accuracy of each algorithm, a validation test that was based on the random exclusion of 25% of the samples was repeated 100 times. In addition, we performed a test of the transferability, in which the locations with the largest nearest-neighbor distances were excluded from training and re-predicted. The validation results demonstrate that ordinary and universal kriging are the best performers, followed closely by random forest (RF) and quantile regression forest (QRF). In terms of local uncertainty, RF and QRF provide confidence intervals that most often include the observed values of SOC. However, they both provide very wide confidence intervals, which may be problematic in some studies. Other algorithms, such as boosted regression trees and boosted regression kriging, performed slightly worse (on this dataset), but produced narrower ranges of uncertainty. Hence, they may be more attractive since their estimates are very robust against changes and noise in the predictors. [ABSTRACT FROM AUTHOR]

[Display omitted] • Soil salinity can be well reflected by vegetation and salinity indicators. • The conversion of spectral indicators can effectively reduce the upscaling error. • Precise land regionalization will improve the accuracy of scale conversion. • The risk of soil salinization gradually decreases from coastal to inland. • Soil salinization experienced a change of increasing first and then decreasing. As a global problem of soil degradation, salinization has become a major obstacle to the sustainable development of the ecological environment and agriculture in coastal plains. However, the traditional process of salinity survey is too cumbersome, expensive and time-consuming to meet the mapping needs in a large scale. Remote sensing technology has become an important tool for digital soil mapping because of its rich sources, real-time and low cost. In order to meet the objective demand for rapid, accurate, and efficient acquisition and monitoring of soil salinization. This paper collected 61 soil samples from the Kenli District (experimental area) and extracted vegetation and salinity indicators from the Landsat image to construct the salinity inversion model by random forest algorithm. Then, taking the Yellow River Delta as the study area, the conversion coefficient of spectral indicators between Landsat and MODIS images was constructed in the form of the ratio of the mean value. Through optimization, the upscaling conversion method based on land use regionalization was proposed to realize the upscaling inversion and dynamic monitoring of soil salinization. The results showed that: (1) The random forest model based on NDVI , RVI , EVI , SI3 , and SI5 can better predict the soil salinity in the experimental area, with R2 = 0.821 and RMSE = 2.811 (validation accuracy). (2) The upscaling conversion method based on land use regionalization can effectively reduce the statistical error and collinearity of spectral indicators constructed by MODIS images and improve their correlation with OLI data and soil salinity. (3) From coastal to inland, soil salinization gradually decreases in the Yellow River Delta. From 2000 to 2020, soil salinization increased first and then decreased, and the salinized soil accounted for 20.35%∼35.10%. This study used multi-source remote sensing data to realize the collaborative inversion at different scales, which was significant for the quantitative estimation of soil salinity, salinization control, and sustainable agricultural development in coastal plains. [ABSTRACT FROM AUTHOR]

Soil organic carbon (SOC) and total nitrogen (TN) influence physical, chemical and biological properties and process in soils that determine soil fertility and enhance or maintain agricultural productivity and food security. Knowledge on the spatial distribution of SOC and TN is necessary for sustainable soil resource management. In this study, we applied boosted regression trees (BRT) model to predict and map the spatial distribution of SOC and TN in the northeastern coastal areas of China, and C:N ratio map was generated from the predicted maps of SOC and TN. A total of 149 topsoil (0–20 cm) samples and 12 environmental variables that included topography, landuse and remote sensing indices, and climatic variables were selected. The performance of the models was evaluated based on cross-validation using mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R 2 ). The BRT model was run for 100 iterations where the average of 100 SOC and TN maps was considered as final predicted maps and its standard deviation as prediction uncertainty. The BRT model showed a good predictive performance due to its higher R 2 and lower MAE, and RMSE indices. The model explained approximately 64% and 56% of the total SOC and TN variability. Topographic variables showed a maximum influence in the prediction of SOC and TN, followed by vegetation and climate. The SOC and TN contents were reduced from the northwest towards the southeast of the study area; average predicted SOC and TN contents were 15.4 g kg −1 and 1.01 g kg −1 , respectively. The spatial distribution of C:N ratio was closely related to landuse types in a decreasing order: woodland > orchard > cultivated land > grassland. We found topographic variables as main environmental indicators of SOC and TN distribution, and recommend to include them in future SOC and TN mapping studies in the coastal agroecosystems in China. [ABSTRACT FROM AUTHOR]

[Display omitted] • Predicting SOC and SIC from two sampling periods (1990 vs 2015); • Quantifying the impact of environmental variables on SOC and SIC changes; • Mapping soil carbon stocks change during 25 years (1990–2015). Soil organic carbon (SOC) and soil inorganic carbon (SIC) has important effects on soil physical, chemical and biological properties. They play an important role in coordinating the relationship between soil water and air, increasing soil water holding capacity and improving plant productivity. In this study, a boosted regression trees (BRT) model was developed to map the spatial distribution carbon stocks in the semi-arid region of Northeast China in 1990 and 2015. During the two periods, 10-fold cross-validation technology was used to test the performance and uncertainty of BRT model. In order to construct the model, 9 environmental variables (derived from climate, topography and biology) and 173 (1990) and 223 (2015) topsoil (0–30 cm) samples were used. The comparison between estimated and observed data shows that the RMSE of SOC and SIC stocks were 0.53 kgm−2 and 0.19 kgm−2 in 1990, and 0.65 kgm−2 and 0.20 kgm−2 in 2015, respectively. Elevation, normalized difference vegetation index, mean annual precipitation and Landsat band 3 were identifies as critical environmental factors for simulating the spatial distribution of SOC, accounting for 76.6 % and 70.3 % of the total relative importance in 1990 and 2015, respectively. Mean annual precipitation, mean annual temperature and topographic wetness index were the critical environmental factors for simulating the spatial variation of SIC during the two periods. Land use change also played an important role in the spatial variability of SOC and SIC stocks. In the past 25 years, soil carbon stocks decreased from 6.2 kg m−2 in 1990 to 5.9 kg m−2 in 2015. The spatial distribution pattern of SOC was high in northeastern area and low in southwestern area during the two periods, while the spatial distribution pattern of SIC was opposite to that of SOC stocks. The mapped soil carbon stock distribution is fundamental to future study of soil carbon cycle and regional carbon balance in semi-arid regions. [ABSTRACT FROM AUTHOR]

• We conducted spatial prediction and uncertainty analysis of SOC and STC. • RFRK gave best predictions for SOC and STC. • RFRK produced higher local accuracy than RF. • Land use types, NDVI and elevation were key factors for SOC and STC predictions. Soil carbon is the largest terrestrial carbon pool. Reliable mapping of soil organic carbon (SOC) and soil total carbon content (STC) is essential for agricultural ecosystem management and carbon accounting under global warming conditions. This study was conducted at a fine scale and aimed to perform spatial distribution prediction and uncertainty mapping of SOC and STC and quantify the contribution of environmental variables affecting the variability of SOC and STC. Three machine learning models, namely, Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Random Forest plus residuals Kriging (RFRK), were developed using 4345 agricultural topsoil samples and 16 environmental covariates. Mean absolute error (MAE), root mean square error (RMSE), Nash–Sutcliffe model efficiency coefficient (NSE), and Lin's concordance correlation coefficient (LCCC) were used to evaluate the prediction global accuracy. Accuracy plot was used to quantify the uncertainty (i.e. local accuracy) of SOC and STC predictions. RFRK performed best with MAE, RMSE, NSE, and LCCC of 20.76%, 27.61%, 0.39, and 0.56 for SOC and 27.20%, 38.59%, 0.35, and 0.53 for STC, respectively. RF outperformed XGBoost in terms of NSE (0.33 vs 0.29 for SOC and 0.36 vs 0.32 for STC). Accuracy plots showed that RFRK produced higher local accuracy than RF both in quantifying the prediction uncertainty of SOC and STC. XGBoost performed excellently in the uncertainty estimation of SOC. Land use types, mean annual Normalized Difference Vegetation Index, and elevation were the top three important indicators in determining the spatial variability of SOC and STC. These results could provide inspiration and support for monitoring soil carbon in complex terrain areas. [ABSTRACT FROM AUTHOR]

Soil organic carbon (SOC) plays an important role in soil fertility and carbon sequestration, and a better understanding of the spatial patterns of SOC is essential for soil resource management. In this study, we used boosted regression tree (BRT) and random forest (RF) models to map the distribution of topsoil organic carbon content at the northeastern edge of the Tibetan Plateau in China. A set of 105 soil samples and 12 environmental variables (including topography, climate and vegetation) were analyzed. The performance of the models was evaluated using a 10-fold cross-validation procedure. Maps of the mean values and standard deviations of SOC were generated to illustrate model variability and uncertainty. The results indicate that the BRT and RF models exhibited very similar performance and yielded similar predicted distributions of SOC. The two models explained approximately 70% of the total SOC variability. The BRT and RF models robustly predicted the SOC at low observed SOC values, whereas they underestimated high observed SOC values. This underestimation may have been caused by biased distributions of soil samples in the SOC space. Vegetation-related variables were assigned the highest importance in both models, followed by climate and topography. Both models produced spatial distribution maps of SOC that were closely related to vegetation cover. The SOC content predicted by the BRT model was clearly higher than that of the RF model in areas with greater vegetation cover because the contributions of vegetation-related variables in the two models (65% and 43%, respectively) differed significantly. The predicted SOC content increased from the northwestern to the southeastern part of the study area, average values produced by the BRT and RF models were 27.3 g kg −1 and 26.6 g kg −1 , respectively. We conclude that the BRT and RF methods should be calibrated and compared to obtain the best prediction of SOC spatial distribution in similar regions. In addition, vegetation variables, including those obtained from remote sensing imagery, should be taken as the main environmental indicators and explicitly included when generating SOC maps in Alpine environments. [ABSTRACT FROM AUTHOR]

• The ranking of environmental factors for soil type mapping was determined. • The optimal environmental covariates for DSM were clarified. • The optimal variable combination was determined. • The best overall accuracy produced by the optimal variable combination was 65.74%. The scientific selection of environmental covariates is an important factor that affects the accuracy of digital soil mapping (DSM).An investigation of the influences of different environmental covariates on soil type identification is vital to understand soil development and predictive soil mapping. In this study, the soil groups in the typical plain and hilly areas of the Jiaodong Peninsula in China were taken as the objects of investigation. Based on five environmental characteristics and remotely sensed indexes (texture, terrain, biology, major spectral components and moisture), 21 covariates sourced from Landsat 8 remote sensing images and a digital elevation map were utilized for the predictive mapping of soils. The identification accuracies of the soil groups were acquired through maximum likelihood classification, and the relevant environmental factors that are important for soil mapping and the influence of different environmental covariates on soil mapping were explored. The results show that texture features had the greatest influence on DSM, followed by terrain, major spectral components, moisture and biological factors. The single environmental covariate with the greatest impact on DSM was the texture entropy parameter. For areas containing plain and hilly landforms, the entropy, elevation, slope, normalized difference vegetation index (NDVI), third principal component (PC3) and normalized difference moisture index (NDMI) indicators compose the optimal covariate combination for soil identification. Clarifying the impacts of different environmental covariates on DSM can improve the soil classification applicability and accuracy. The environmental covariates discussed in this study would be useful in similar environmental areas and would provide a reference for DSM. [ABSTRACT FROM AUTHOR]

Soil organic carbon (SOC) is a key indicator of ecosystem health, with a great potential to affect climate change. This study aimed to develop, evaluate, and compare the performance of support vector regression (SVR), artificial neural network (ANN), and random forest (RF) models in predicting and mapping SOC stocks in the Eastern Mau Forest Reserve, Kenya. Auxiliary data, including soil sampling, climatic, topographic, and remotely-sensed data were used for model calibration. The calibrated models were applied to create prediction maps of SOC stocks that were validated using independent testing data. The results showed that the models overestimated SOC stocks. Random forest model with a mean error (ME) of −6.5 Mg C ha −1 had the highest tendency for overestimation, while SVR model with an ME of −4.4 Mg C ha −1 had the lowest tendency. Support vector regression model also had the lowest root mean squared error (RMSE) and the highest R 2 values (14.9 Mg C ha −1 and 0.6, respectively); hence, it was the best method to predict SOC stocks. Artificial neural network predictions followed closely with RMSE, ME, and R 2 values of 15.5, −4.7, and 0.6, respectively. The three prediction maps broadly depicted similar spatial patterns of SOC stocks, with an increasing gradient of SOC stocks from east to west. The highest stocks were on the forest-dominated western and north-western parts, while the lowest stocks were on the cropland-dominated eastern part. The most important variable for explaining the observed spatial patterns of SOC stocks was total nitrogen concentration. Based on the close performance of SVR and ANN models, we proposed that both models should be calibrated, and then the best result applied for spatial prediction of target soil properties in other contexts. [ABSTRACT FROM AUTHOR]

Soil organic matter (SOM) content is considered as an important indicator of soil quality. An accurate spatial prediction of SOM content is so important for estimating soil organic carbon pool and monitoring change in it over time at a regional scale. Due to the unfavourable natural conditions in Tibetan Plateau, soil sampling with high density is time consuming and expensive. As a result, little research has focused on the spatial prediction of SOM content in Tibet because of shortage of data. We used a two-stage process that integrated an artificial neural network (ANN) and the estimation of its residuals by ordinary kriging to produce accurate SOM content maps based on sparsely distributed observations and available auxiliary information. SOM content data were obtained from a soil survey in Tibet and were used to train and validate the ANN-kriging methodology. Available environmental information including elevation, temperature, precipitation, and normalized difference vegetation index were used as auxiliary variables in the ANN training. The prediction accuracy of SOM content was compared with those of ANN, universal kriging, and inverse distance weighting (IDW). A more accurate prediction of SOM content was obtained by ANN-kriging, with lower global prediction errors (root mean square error = 6.02 g kg-1) and higher Lin's concordance correlation coefficient (0.75) for validation sampling sites compared with other methods. Relative improvements of 26.94-37.10% over other methods were observed in the prediction of SOM content. In conclusion, the proposed ANN-kriging methodology is particularly capable of improving the accuracy of SOM content mapping at large scale. [ABSTRACT FROM AUTHOR]

• Filter, wrapper, embedded and ensemble feature selection approaches (FS) were evaluated. • The proposed ensemble FS incorporates ten individual selectors to identify top significant variables. • Ensemble FS produced the highest predictive variable subset for mapping SOM. • XGBoost outperformed RF in mapping SOM in subtropical restored forests. Mapping Soil organic matter (SOM) over a complex forest landscape is challenging due to the difficulty in selecting the most insightful variables from high-dimensional datasets in the recent explosion of geospatial-data. Feature selection (FS) is necessary to reduce data redundancy and noise as well as to achieve more reliable SOM spatial predictions. However, it is still unclear that which is most effective among various FS methods in mapping SOM. Therefore, four types of FS approaches (i.e., filter, wrapper, embedded and ensemble) were adopted to generate optimum variable subsets from an original variable dataset of 60 candidates, respectively, for mapping SOM of restored forest land in a typical subtropical region of southern China. The most used methods for each type of FS approaches were selected in this study, including three filters (Chi-square, InfoGain and pearson correlation analysis), three wrappers (genetic algorithm, simulated annealing algorithm and support vector machine-recursive feature elimination) and three embedded methods (Boruta, random forest (RF) and extreme gradient boosting (XGBoost)), as well as an ensemble method (robust rank aggreg algorithm (RRA)). Meanwhile, the RF and XGBoost models were applied with a 10-fold cross-validation method to compare the relative advantages of the different FS methods in SOM mapping, by utilizing the correlation coefficients R2 between observed and predicted values and predicting errors of root mean square error (RMSE). The results show that the SOM prediction accuracies with optimized variable subsets generated by the different FS methods are better than those with full variables, yet the improvements of prediction performance are different among the four types of FS approaches. The ensemble method (RRA) is superior to the other three types of approaches with an average RMSE reduction of 9.16% comparing to that without using FS methods, followed by wrapper and embedded methods which obtained the average RMSE reduction by 7.81%, 7.32%, respectively, and the filter methods are the weakest in the RMSE reduction with slight decreases of 4.32%. The XGBoost model achieved a better performance in predicting SOM than the RF model regardless of input variables, and the XGBoost model combined with RRA FS method shows the greatest potential to map SOM in the restored forest land. This study provides a reference for obtaining more parsimonious and robust variable sets from the available big geo-data freely for soil mapping in other areas. [ABSTRACT FROM AUTHOR]

Highlights: [•] Land surface diurnal temperature differences (DTD) were negatively correlated with soil organic matter (SOM) in low-relief plains. [•] The soil-adjusted vegetation index (SAVI) was positively correlated with SOM in these areas. [•] Regression kriging with a mixed model using DTD, SAVI and soil type predicted SOM. [•] The generic mixed model RK method must be calibrated for each area separately. [Copyright &y& Elsevier]

[Display omitted] • The determinants of MHT were identified both statistically and pedologically. • Distribution patterns of MHT in alpine environment were first revealed. • Partial dependence plots revealed the non-linear predictor-response relationships. • The proposed approach can be adopted into the Digital Soil Mapping framework. Mattic horizon, grassroots intertwined surface horizon for soils in alpine meadow, is of great importance for alpine ecosystems with its thickness serving as a critical indicator for the extent of alpine meadow degradation. However, the distribution patterns and key environmental factors controlling mattic horizon thickness (MHT) remain unknown. We evaluated the potential of a boosted regression trees (BRT) based variable selection approach to identify the determinants of MHT, to reveal the predictor-response relationship, and to map the spatial distributions of MHT. Results showed that the selected six key environmental variables can reliably reflect the pedogenic processes of mattic horizon in such alpine environment. Among which, the most important environmental variables controlling MHT were modified soil-adjusted vegetation index 2 (relative importance of 31.9%), elevation (22.7%), followed by slope (13.2%), topographic wetness index (11.4%), relative slope position (10.4%), and mean annual precipitation (10.4%). Additionally, the BRT model using those six key variables for mapping MHT performed an acceptable prediction with R2 value of 0.447 (standard deviation of 0.019). The variable selection procedure dramatically reduced the number of environmental variables by 77% (from 26 to 6), which could not only improve computational efficiency but also significantly improve model performance with a 30.3% increase in R2. Our results demonstrate that the proposed BRT algorithm-based variable selection approach can reliably identify the determinants of MHT both statistically and pedologically. It can be used as a promising approach for revealing the predictor-response relationship, mapping the distribution patterns of MHT, and may provide a solution to map soil depth over large areas. [ABSTRACT FROM AUTHOR]

• Simulation accuracy and cost are two core issues in soil organic carbon mapping. • A zonal simulation approach was proposed for realizing high accuracy and low cost. • A typical subzone with an area proportion <20% of the study area was identified. • The zonal simulation achieves high accuracy with a cost decrease by more than 50%. • Geographical integrity, typicalness, and scale are key points of zonal simulation. Digital soil mapping (DSM) has been developed for decades and aims to accurately simulate the soil factors with low cost. However, it still cost considerably in coastal wetlands because of the high difficulty in field survey. A zonal simulation approach was proposed for decreasing the cost of field survey in the premise of high accuracy for soil organic carbon stock mapping in coastal wetlands, and Chongming Island, an important estuarine wetland in China, was used to demonstrate the study. A subzone with an area proportion lower than 20% was identified based on the typicalness of land surface characteristics and soil influencing factors for the whole study area. Entire and zonal simulations were conducted using abundant predictors sourced from remote sensing, partial least square regression, and 10-fold cross validation. The results revealed that root mean squared errors of entire and zonal simulations were 1.70 g/kg and 1.95 g/kg, respectively, which were in a low level compared with those in previous studies and indicated the high simulation accuracy. The zonal simulation exhibited a slightly lower accuracy than the entire simulation, while considerably decreased the cost by more than 50%. The decrease in the cost showed a much more practical significance than the increase in the accuracy in areas with high difficulty in field survey, and the geographical integrity and continuity of the whole study area, the typicalness of the selected subzone, and the spatial extent were the three key points when promoting the zonal simulation in DSM. The soil organic carbon stock in the study area was generally high in areas with good vegetation condition, low soil salinity, complex landscape configuration, and long distances from the sea. Human activities and the resulting landscape fragmentation have been the dominant factor that drives the spatial pattern of soil organic carbon stock. [ABSTRACT FROM AUTHOR]

Digital soil mapping approaches related to soil organic matter (SOM) are crucial to quantify the process of the carbon cycle in terrestrial ecosystems and thus, can better manage soil fertility. Recently, many studies have compared machine learning (ML) models with traditional statistical models in digital soil mapping. However, few studies focused on the application of hybrid models that combine ML with statistical models to map SOM content, especially in loess areas, which have a complicated geomorphologic landscape. In this study, the trend prediction used two ML models, i.e., gradient boosting modeling and random forest (RF), and a traditional stepwise multiple linear regression plus interpolated residuals generated from two classic geostatistical models, i.e., ordinary kriging and inverse distance weighting, and a high accuracy surface modeling (HASM) were implemented to map SOM content in the Dongzhi Loess Tableland area of China. A total of 145 topsoil samples and heterogeneous environmental variables were collected to develop the hybrid models. Results showed that 18 variables related to soil properties, climate variables, terrain attributes, vegetation indices, and location attributes played an important role in SOM mapping. The models that incorporate ML algorithms and interpolated residuals to predict SOM variation were found to have a better ability to handle complex environment relationships. The HASM model outperformed traditional geostatistical models in interpolating the residuals. In contrast, RF combined with HASM residuals (RF_HASM) gave the best performance, with the lowest mean absolute error (1.69 g/kg), root mean square error (2.30 g/kg), and the highest coefficient of determination (0.57) and concordance correlation coefficient (0.69) values. Moreover, the spatial distribution pattern obtained with RF_HASM yielded a spatial distribution of SOM that better fit the actual distribution pattern of the study area. In conclusion, these results suggest that RF_HASM is particularly capable of improving the mapping accuracy of SOM content at the regional scale. [ABSTRACT FROM AUTHOR]

Abstract: Terrain attributes derived from digital elevation models have been used widely for mapping soil organic matter (SOM). Among these attributes, the topographic wetness index (TWI), an index for quantitatively indicating the balance between water accumulation and drainage conditions at the local scale, has been shown to correlate with SOM. However, TWIs used in most studies are calculated using a single-flow-direction (SFD) algorithm, which assumes that all water from a grid cell flows into only one neighboring cell. This assumption is not always valid, especially in areas with low relief where movement of water may be divergent. To overcome this SFD limitation, a multiple-flow-direction (MFD) algorithm has been developed, which distributes flow from a grid cell to several downslope neighbors. In this study we compared the effect of TWI calculations based on SFD and MFD in predictive mapping of SOM by incorporating them into different kriging methods over a 51.76km2 area in Nenjiang County of northeastern China. We found that the MFD-based TWI was better correlated with SOM than was the SFD-based index. We then compared the accuracies of SOM maps which were derived from MFD-based TWI and SFD-based TWI incorporated by ordinary kriging (OK), simple kriging with varying local means (SKlm), kriging with external drift (KED) and collocated cokriging (CC). The MFD-based TWI, used as a secondary variable in SKlm and CC, outperforms the SFD-based TWI. For the different kriging methods, CC (incorporating either MFD-based TWI or SFD-based TWI) showed the best performance, and OK generated a better result than SKlm and KED. Both the MFD-based TWI and SFD-based TWI proved to be incompatible with KED and SKlm due to their numerical instability caused by the rough TWI surfaces. Among all predictive methods, CC incorporating the MFD-based TWI produced the best results. This is because: (1) the MFD-based TWI is best able to indicate quantitatively soil moisture and therefore has the strongest correlation with SOM; (2) CC is capable of utilizing effectively the spatial auto-correlation of SOM and the cross-correlation between SOM and the MFD-based TWI. [Copyright &y& Elsevier]

• For small sample sizes, modelling methods have a greater impact on accuracy of DSM. • Large sample sizes have a greater impact than modelling methods on accuracy of DSM. • An optimal combination of modelling method and sample size exists for SOM mapping. Digital soil mapping (DSM) is the most widely used method for producing spatial information of soil organic matter (SOM). Accuracy of the information is generally determined by modelling methods and sample sizes used for DSM. However, different studies present different importance of modelling method and sample size on accuracy of DSM, while they do not explore various combinations of modelling method and sample size. Based on the studies, it is supposed that there exists an optimal combination of modelling method and sample size for producing information of SOM accurately and economically. With SOM data of 1861 soil samples collected in Guangdong, China, the present study first assessed importance of modelling method and sample size and then examined if an optimal combination of modelling method and sample size existed for the area. Six modelling methods were explored, while 12 sample sizes were used, ranging from 100 to 1200 with an interval of 100. For each size, 10 repeated samples were randomly taken from a data of 1311 samples which were randomly selected from all the 1861 soil samples based on the probability distribution of the SOM data. The results showed that, for small sample sizes, the modelling methods have a greater impact on accuracy of DSM. However, for large sample sizes, e.g., more than 1000, the sample sizes have a much greater impact. Due to the varying importance of modelling method and sample size, there exists an optimal combination of modelling method and sample size for spatial prediction of SOM in the area, i.e., the combination of regression kriging and a sample size of 800. Thus, for economically producing detailed and accurate information on spatial distribution of SOM, it is recommended that a series of modelling methods and sample sizes are tried to identify an optimal combination of modelling method and sample size. [ABSTRACT FROM AUTHOR]

Many jurisdictions have put forward guidelines to identify sites at risks of soil degradation with the extraction of logging residues. Most guidelines are based on expert opinion and use the precautionary principle because of the lack of strong understanding of what makes a site sensitive to intensive biomass removal. Two main approaches are used: 1-identifying thresholds for specific site properties 2- developing nutrient budget indicators to rate the potential for nutrient deficit. Thanks to the development of digital soil mapping, it is becoming easier to develop maps for such indicators. A crucial question is the reliability of the different indicators. One way of evaluating their reliability is to test the coherence between the geographic locations of sites rated as sensitive by different indicators. In this study, we developed maps of key soil properties and of biogeochemical fluxes for the managed forest land of Canada at a resolution of 250 m. We used three site properties (slope, pH and sand content), as well as three nutrient budget indicators (N Budget, Base cation (BC) Budget and N Stability ratio) that were mapped and compared. The results indicated very little overlap between the areas identified as sensitive by the three site property indicators and very little concordance between sites rated as sensitive by site property indicators and those identified as sensitive by nutrient budget indicators while nutrient budget indicators showed coherence among themselves. Because nutrient budget indicators were found to be more dependent on the amount of nutrient extracted in the harvested biomass, than on the rates of nutrient inputs from the soil or the atmosphere, they tended to rate productive sites as sensitive, to the opposite of site property indicators. These results suggest that different indicators are assessing different processes and different aspects of site sensitivity. Because of the lack of coherence amongst the indicators, it is advisable 1) to use indicators based on the results of long-term monitoring plots, 2) to maintain such long-term observations and 3) to leave on the ground a substantial proportion of harvest residues, especially on sites evaluated as sensitive. [ABSTRACT FROM AUTHOR]

• Spatial-temporal variation of soil quality index was digitally mapped in central Iran. • Temporal analysis of the predicted maps explored a decreasing trend in soil quality index values from 1986 to 2030. • Groundwater quality indicators and human activities had the highest correlation with soil quality. Soil degradation reduces soil quality by a loss of organic matter, reduced soil fertility, and structural breakdown, erosion, undesirable changes in salinity, acidity or alkalinity and the effects of toxic chemicals, contaminants or excessive flooding. In this study, we investigated spatial and temporal changes of soil quality between 1986 and 2030 in the Central Iranian desert. 201 topsoil samples (0–20 cm) in Yazd-Ardakan plain of Iran were chosen with conditioned Latin hypercube sampling strategy and analysed for soil organic carbon, electrical conductivity, dry bulk density, aggregate stability, and soil heavy metals. A soil quality index was calculated using weighted index method. To predict the spatial distribution of the soil quality index, we implemented a random forest model based on a set of covariates with a coefficient of determination of 0.69 between soil quality index and covariates. The results of soil quality index changes from 1986 to 2030 show that during this period, the areal extent of the Very Low SQI Class (<0.4) increased by 78,242 ha (increase of 16.2%), while the extent of the Good SQI Class (>0.6) decreased by 366,018 ha (decrease of 75.8%). This indicates that the soil quality in the study area, over time, is progressively becoming poorer. The results of the study of soil quality changes can be used in land evaluation, environmental studies and integrated planning and management in order to properly and reasonably utilize natural resources and reduce future soil degradation. [ABSTRACT FROM AUTHOR]

• The feasibility of adding multi-temporal radar Sentinel-1A data to SOC prediction was explored. • Climate variables contributed the most to explaining SOC content variation. • In this ecosystem, RF and BRT models showed similar performance and produced similar SOC spatial distribution patterns. Soil organic carbon (SOC) has a large impact on soil quality and global climate change. It is therefore important to be able to predict SOC accurately to promote sustainable soil management. Although the synthetic aperture radar (SAR) has many advantages and has been widely used in soil science research, it has rarely been used in previous SOC mapping studies based on remote sensing images. The purpose of this study was to investigate the ability of multi-temporal Sentinel-1A data in SOC prediction, by comparing the predictive performance of random forest (RF) and boosted regression tree (BRT) models in the Heihe River Basin in northwestern China. A set of 162 topsoil (0–20 cm) samples were taken and 15 environmental variables were obtained including land use, topography, climate, and remote sensing images (optical and SAR data). Using a cross-validation procedure to evaluate the performance of the models, three statistical indices were calculated. Overall, both RF and BRT models effectively predicted SOC content, exhibiting similar performance and producing similar spatial distribution patterns of SOC. The results showed that the addition of multi-temporal Sentinel-1A images improved prediction accuracy, with the root mean squared error (RMSE), the mean absolute error (MAE) and the coefficient of determination (R2) improving by 9.0%, 8.3% and 13.5%, respectively. Furthermore, the combination of all environmental variables had the best prediction performance explaining 75% of SOC variation. The most important environmental variables explaining SOC variation were precipitation, elevation, and temperature. The multi-temporal Sentinel-1A data in RF and BRT models explained 9% and 7%, respectively. The results from our case study highlight the usefulness of multi-temporal Sentinel-1 data in SOC mapping. [ABSTRACT FROM AUTHOR]

• Soil TN is controlled by soil moisture and geographic attributes. • Soil AK is controlled by soil salinity, vegetation and geographic attributes. • Spatial distribution of TN and AK were characterized in YRD. • Digital soil mapping can be widely applied to map soil properties in wetland. Coastal wetland soils play a vital role in providing habitats for wildlife and vegetation, and in affecting water quality and primary production in coastal wetland ecosystem. The spatial prediction of soil properties provides important data support to help sustainable wetland management. This research explored the relationships between soil total nitrogen (TN) and soil available potassium (AK) and multiple environmental variables such as remote sensing spectral indices, geographic attributes and land use types of coastal wetlands in the Yellow River Delta (YRD). Soil moisture and geographic attributes were the major factors controlling the spatial distribution of soil TN. Soil salinity, vegetation and geographic attributes were the key factors affecting the spatial distribution of soil AK. The research established spatial prediction models of soil TN and AK by random forest method, and showed heterogeneous spatial patterns of TN and AK in the study area. Soil TN model showed relatively higher prediction performance compared with soil AK model. The research framework utilized in this study can be promoted to predict and map other soil properties, and has significant implication for the site-specific soil management of coastal wetlands in the YRD. [ABSTRACT FROM AUTHOR]