Prediction Of Soil Organic Matter Research Articles

A new method to estimate soil organic matter using the combination model based on short memory fractional order derivative and machine learning model

In order to improve the quantitative inversion accuracy of soil organic matter in arid areas, the soils affected by different human disturbances in Xinjiang were used as research objects (Regions I, II and III). This study compared and analyzed the preprocessing hyperspectral data effects of GMFOD (global memory fractional order derivative) with 1759 memory length and SMFOD (short memory fractional order derivative) with different memory lengths (L = 1500, 1200, 1000, 800, 500, 300, 250, 200, 150, 100), machine learning models based on RBF (radial basis function) and ELM (extreme learning machine) were used to explore the best model for predicting soil organic matter content. Simulation results showed that (1) SMFOD had shorter computational runtime compared to GMFOD, and the 100 SMFOD had the shortest running time. When all orders of fractional order derivative were calculated, the time required for 100 SMFOD was 115.27 s, while the time required for GMFOD was 1025.32 s. These indicated that the running time of the former was 789.49% shorter than that of GMFOD. (2) Comparing the correlation coefficients of hyperspectral and soil organic matter in all fractional orders calculated based on GMFOD and SMFOD, it was found that the maximum correlation coefficient of 100 SMFOD and GMFOD differed by 0.0103, 0.0058 and 0.0287 in Regions I, II and III, which means that GMFOD only increased by 0.94%, 1.71% and 4.25%. These indicated that the maximum correlation coefficient of SMFOD and GMFOD was very similar at each fractional order. (3) The best model for inversion of soil organic matter in all three regions was SMFOD-CC1-RBF (CC1 denotes bands passed the 0.01 significance test), which was located at the fractional orders of 1.4, 1.8 and 1.1, respectively. Meanwhile, R2 (coefficient of determination), RMSE (root mean square error), RPD (ratio of the performance to deviation) were 0.8249, 0.7656, 1.9256 in Region I; were 0.7630, 1.4422, 1.9561 in Region II; and were 0.6799, 1.5343,1.8513 in Region III, which indicated that the higher-order SMFOD had better performance in inverting organic matter than the lower-order SMFOD, and the prediction ability of this model for soil organic matter in the three regions was good.

Impact of Spectral Resolution and Signal-to-Noise Ratio in Vis–NIR Spectrometry on Soil Organic Matter Estimation

Recently, considerable efforts have been devoted to the estimation of soil properties using optical payloads mounted on drones or satellites. Nevertheless, many studies focus on diverse pretreatments and modeling techniques, while there continues to be a conspicuous absence of research examining the impact of parameters related to optical remote sensing payloads on predictive performance. The main aim of this study is to evaluate how the spectral resolution and signal-to-noise ratio (SNR) of spectrometers affect the precision of predictions for soil organic matter (SOM) content. For this purpose, the initial soil spectral library was partitioned into to two simulated soil spectral libraries, each of which were individually adjusted with respect to the spectral resolutions and SNR levels. To verify the consistency and generality of our results, we employed four multiple regression models to develop multivariate calibration models. Subsequently, in order to determine the minimum spectral resolution and SNR level without significantly affecting the prediction accuracy, we conducted ANOVA tests on the RMSE and R2 obtained from the independent validation dataset. Our results revealed that (i) the factors significantly affecting SOM prediction performance, in descending order of magnitude, were the SNR levels > spectral resolutions > estimation models, (ii) no substantial difference existed in predictive performance when the spectral resolution fell within 100 nm, and (iii) when the SNR levels exceeded 15%, altering them did not notably affect the SOM predictive performance. This study is expected to provide valuable insights for the design of future optical remote sensing payloads aimed at monitoring large-scale SOM dynamics.

Mapping soil organic matter in cultivated land based on multi-year composite images on monthly time scales

Rapid and accurate acquisition of soil organic matter (SOM) information in cultivated land is important for sustainable agricultural development and carbon balance management. This study proposed a novel approach to predict SOM with high accuracy using multiyear synthetic remote sensing variables on a monthly scale. We obtained 12 monthly synthetic Sentinel-2 images covering the study area from 2016 to 2021 through the Google Earth Engine (GEE) platform, and reflectance bands and vegetation indices were extracted from these composite images. Then the random forest (RF), support vector machine (SVM) and gradient boosting regression tree (GBRT) models were tested to investigate the difference in SOM prediction accuracy under different combinations of monthly synthetic variables. Results showed that firstly, all monthly synthetic spectral bands of Sentinel-2 showed a significant correlation with SOM (P<0.05) for the months of January, March, April, October, and November. Secondly, in terms of single-monthly composite variables, the prediction accuracy was relatively poor, with the highest R2 value of 0.36 being observed in January. When monthly synthetic environmental variables were grouped in accordance with the four quarters of the year, the first quarter and the fourth quarter showed good performance, and any combination of three quarters was similar in estimation accuracy. The overall best performance was observed when all monthly synthetic variables were incorporated into the models. Thirdly, among the three models compared, the RF model was consistently more accurate than the SVM and GBRT models, achieving an R2 value of 0.56. Except for band 12 in December, the importance of the remaining bands did not exhibit significant differences. This research offers a new attempt to map SOM with high accuracy and fine spatial resolution based on monthly synthetic Sentinel-2 images.

Application of fractional-order differential and ensemble learning to predict soil organic matter from hyperspectra

Application of fractional-order differential and ensemble learning to predict soil organic matter from hyperspectra

Estimating Soil Organic Matter (SOM) Using Proximal Remote Sensing: Performance Evaluation of Prediction Models Adjusted at Local Scale in the Brazilian Cerrado

The quantification of soil organic matter (SOM) has increased over the years, especially in the Brazilian Cerrado region, one of the most important areas for grain production in the country. In this area, SOM content tends to be low, which directly impacts the physical, chemical and biological quality of soils. Thus, the use of spectroradiometry has been widely evaluated to investigate whether it can be used as a faster, more reliable and cheaper solution to meet the SOM estimation. In this context, the objective of the present paper was to evaluate the performance of a local spectral model for SOM prediction generated through the spiking strategy. The research was developed in the municipality of Passos, Minas Gerais State, located in the Brazilian Cerrado. Soil samples (0–0.2 m and 0.2–0.4 m depths) were collected in a zigzag pattern and split in calibration of the local models from a test area (90 soil samples) and recalibration and validation from a target area (46 soil samples). After this stage, the SOM contents were determined in a laboratory, and the spectral responses (350–2500 nm) of each soil sample were collected. From the target area, 10, 25 and 50% of soil spectra were selected for recalibration of the local models generated for the test area. Although median results were observed in the post-recalibration, due to the type of sample selected and the relative similarity among the spectral curves of both areas, improvement was observed for all statistical indices, especially when using 50% (23) of samples for recalibration of the local models, reaching r2 = 0.43, RMSEP = 2.34 gdm−3 and RPIQ = 4.58. These results are important for the SOM estimation in the Brazilian Cerrado considering its importance to the food security and socioeconomic activities. However, considering the lack of similar research in the study area, it is necessary to further investigate the development of spectral models on a local scale and their contribution to improve the identification of SOM spatial variability.

Improving digital mapping of soil organic matter in cropland by incorporating crop rotation

Despite human activities are key influencing factors for cropland soil organic matter (SOM), detailed characterization of human activities has always been limited in the digital mapping of SOM due to the lack of proper representations of human’s cropland use activities. Crop rotation is an essential human agricultural practice significantly affecting the spatial–temporal variations of SOM due to the periodically dynamic changes of crops. Thus, incorporating crop rotation in the digital soil mapping holds high potential for improving SOM prediction. Here, we applied time-series radar Sentinel-1 and optical Sentinel-2 to map crop rotation systems by a hierarchical rule-based method. Then we explored the effectiveness of incorporating such information in predicting SOM by implementing various combinations of predictive variables. We chose a typical multiple cropping region with various crop rotations in southern China. The model performance was evaluated by 10-fold cross-validation. Results showed significant differences in SOM among the crop rotation systems, and the single rice rotated with vegetables has the highest SOM followed by the high-diversity vegetables and long-term orchard systems. Adding crop rotation enhanced the predictability of SOM with a decrease in RMSE by 7% and an increase in R2 by 24%. Furthermore, the crop rotation systems appeared more important in the predictive models than the soil, topographic, and climatic variables. Our results demonstrated the effectiveness of including crop rotation in predicting SOM over complex agricultural landscapes. Our study indicated that human activities should be characterized more detailedly in cropland soil mapping, and that crop rotation containing information on the seasonal dynamics of cropland may be an option for such characterization.

Strategies and methods for predicting soil organic matter at the field scale based on the provincial near infrared spectral database

Strategies and methods for predicting soil organic matter at the field scale based on the provincial near infrared spectral database

Comparison of Depth-Specific Prediction of Soil Properties: MIR vs. Vis-NIR Spectroscopy.

The prediction of soil properties at different depths is an important research topic for promoting the conservation of black soils and the development of precision agriculture. Mid-infrared spectroscopy (MIR, 2500-25000 nm) has shown great potential in predicting soil properties. This study aimed to explore the ability of MIR to predict soil organic matter (OM) and total nitrogen (TN) at five different depths with the calibration from the whole depth (0-100 cm) or the shallow layers (0-40 cm) and compare its performance with visible and near-infrared spectroscopy (vis-NIR, 350-2500 nm). A total of 90 soil samples containing 450 subsamples (0-10 cm, 10-20 cm, 20-40 cm, 40-70 cm, and 70-100 cm depths) and their corresponding MIR and vis-NIR spectra were collected from a field of black soil in Northeast China. Multivariate adaptive regression splines (MARS) were used to build prediction models. The results showed that prediction models based on MIR (OM: RMSEp = 1.07-3.82 g/kg, RPD = 1.10-5.80; TN: RMSEp = 0.11-0.15 g/kg, RPD = 1.70-4.39) outperformed those based on vis-NIR (OM: RMSEp = 1.75-8.95 g/kg, RPD = 0.50-3.61; TN: RMSEp = 0.12-0.27 g/kg; RPD = 1.00-3.11) because of the higher number of characteristic bands. Prediction models based on the whole depth calibration (OM: RMSEp = 1.09-2.97 g/kg, RPD = 2.13-5.80; TN: RMSEp = 0.08-0.19 g/kg, RPD = 1.86-4.39) outperformed those based on the shallow layers (OM: RMSEp = 1.07-8.95 g/kg, RPD = 0.50-3.93; TN: RMSEp = 0.11-0.27 g/kg, RPD = 1.00-2.24) because the soil sample data of the whole depth had a larger and more representative sample size and a wider distribution. However, prediction models based on the whole depth calibration might provide lower accuracy in some shallow layers. Accordingly, it is suggested that the methods pertaining to soil property prediction based on the spectral library should be considered in future studies for an optimal approach to predicting soil properties at specific depths. This study verified the superiority of MIR for soil property prediction at specific depths and confirmed the advantage of modeling with the whole depth calibration, pointing out a possible optimal approach and providing a reference for predicting soil properties at specific depths.

A Framework for Retrieving Soil Organic Matter by Coupling Multi-Temporal Remote Sensing Images and Variable Selection in the Sanjiang Plain, China

Soil organic matter (SOM) is an important soil property for agricultural production. Rising grain demand has increased the intensity of cultivated land development in the Sanjiang Plain of China, and there is a strong demand for SOM monitoring in this region. Therefore, Baoqing County of the Sanjiang Plain, an important grain production area, was considered the study area. In the study, we proposed a framework for high-accuracy SOM retrieval by coupling multi-temporal remote sensing (RS) images and variable selection algorithms. A total of 73 surface soil samples (0–20 cm) were collected in 2010, and Landsat 5 images acquired during the bare soil period (April, May, and June) were selected from 2008 to 2011. Three variable selection algorithms, namely, Genetic Algorithm, Random Frog and Competitive Adaptive Reweighted Sampling (CARS), were combined with Partial Least Squares Regression (PLSR) to build SOM retrieval models on the spectral bands and indices of the images. The results using a single-date image showed that the combination of variable selection algorithms and PLSR outperformed using PLSR alone, and CARS showed the best performance (R2 = 0.34, RMSE = 15.66 g/kg) among all the algorithms. Therefore, only CARS was applied to SOM retrieval in the different year interval groups. To investigate the effect of the image acquisition time, all images were divided into various year interval groups, and the resulting images were then stacked. The results using multi-temporal images showed that the SOM retrieval accuracy improved as the year interval lengthened. The optimal result (R2 = 0.59, RMSE = 11.81 g/kg) was obtained from the 2008–2011 group, wherein the difference indices derived from the images of 2009, 2010, and 2011 dominated the selected spectral variables. Moreover, the spatial prediction of SOM based on the optimal model was consistent with the distribution of SOM. Our study suggested that the proposed framework that couples stacked multi-temporal RS images with variable selection algorithms has potential for SOM retrieval.

Improved multivariate modeling for soil organic matter content estimation using hyperspectral indexes and characteristic bands.

Soil organic matter (SOM) is a key index of soil fertility. Calculating spectral index and screening characteristic band reduce redundancy information of hyperspectral data, and improve the accuracy of SOM prediction. This study aimed to compare the improvement of model accuracy by spectral index and characteristic band. This study collected 178 samples of topsoil (0-20 cm) in the central plain of Jiangsu, East China. Firstly, visible and near-infrared (VNIR, 350-2500 nm) reflectance spectra were measured using ASD FieldSpec 4 Std-Res spectral radiometer in the laboratory, and inverse-log reflectance (LR), continuum removal (CR), first-order derivative reflectance (FDR) were applied to transform the original reflectance (R). Secondly, optimal spectral indexes (including deviation of arch, difference index, ratio index, and normalized difference index) were calculated from each type of VNIR spectra. Characteristic bands were selected from each type of spectra by the competitive adaptive reweighted sampling (CARS) algorithm, respectively. Thirdly, SOM prediction models were established based on random forest (RF), support vector regression (SVR), deep neural networks (DNN) and partial least squares regression (PLSR) methods using optimal spectral indexes, denoted here as SI-based models. Meanwhile, SOM prediction models were established using characteristic wavelengths, denoted here as CARS-based models. Finally, this research compared and assessed accuracy of SI-based models and CARS-based models, and selected optimal model. Results showed: (1) The correlation between optimal spectral indexes and SOM was enhanced, with absolute value of correlation coefficient between 0.66 and 0.83. The SI-based models predicted SOM content accurately, with the coefficient of determination (R2) and root mean square error (RMSE) values ranging from 0.80 to 0.87, 2.40 g/kg to 2.88 g/kg in validation sets, and relative percent deviation (RPD) value between 2.14 and 2.52. (2) The accuracy of CARS-based models differed with models and spectral transformations. For all spectral transformations, PLSR and SVR combined with CARS displayed the best prediction (R2 and RMSE values ranged from 0.87 to 0.92, 1.91 g/kg to 2.56 g/kg in validation sets, and RPD value ranged from 2.41 to 3.23). For FDR and CR spectra, DNN and RF models achieved more accuracy (R2 and RMSE values ranged from 0.69 to 0.91, 1.90 g/kg to 3.57 g/kg in validation sets, and RPD value ranged from 1.73 to 3.25) than LR and R spectra (R2 and RMSE values from 0.20 to 0.35, 5.08 g/kg to 6.44 g/kg in validation sets, and RPD value ranged from 0.96 to 1.21). (3) Overall, the accuracy of SI-based models was slightly lower than that of CARS-based models. But spectral index had a good adaptability to the models, and each SI-based model displayed the similar accuracy. For different spectra, the accuracy of CARS-based model differed from modeling methods. (4) The optimal CARS-based model was model CARS-CR-SVR (R2 and RMSE: 0.92 and 1.91 g/kg in validation set, RPD: 3.23). The optimal SI-based model was model SI3-SVR (R2 and RMSE: 0.87 and 2.40 g/kg in validation set, RPD: 2.57) and model SI-SVR (R2 and RMSE: 0.84 and 2.63 g/kg in validation set, RPD: 2.35).

Strategies for predicting soil organic matter in the field using the Chinese Vis-NIR soil spectral library

The large-scale soil spectral library (SSL) can provide abundant information for predicting soil properties at a local scale, especially in places lacking data. However, since all the existing large-scale SSLs only contain dry spectra recorded under laboratory conditions, the challenge remains in using SSL for predicting soil information using in situ soil spectra. Previous studies have focused on the methods of transforming in situ spectra to dry spectra when using SSLs, while few studies have compared which strategies are optimal in predicting soil properties. To determine the optimal strategies for predicting soil organic matter (SOM) from an area not located in the archived Chinese Soil Spectral Library (CSSL), we investigated the prediction accuracy of memory-based learning (MBL) using spectra transformed by external parameter orthogonalization (EPO) on the CSSL (MBL_EPO) and on the CSSL spiked with subset samples selected by the conditioned Latin hypercube (cLH) algorithm (MBL_EPO_spiking) and using the data from CSSL spiked directedly by the subset in situ samples (MBL_wet_spiking). We also evaluated the prediction accuracy of the in situ and dry spectra using the selected subset and the partial least squares regression (PLSR) model directly. The results showed that the mean squared Euclidean distance (msd) calculated from spectra was an optimal indicator for selecting the representative samples for both the laboratory and in situ conditions. When only 20 samples with both in situ and dry spectra are available to predict SOM, MBL_EPO_spiking is suggested; otherwise, MBL_wet_spiking with the spiking of in situ spectra determined by the smallest msd is optimal. Our findings pave the way for efficient SOM prediction in situ by integrating a large-scale SSL.

Mapping Forest Cover and Estimating Soil Organic Matter by GIS-Data and an Empirical Model at the Subnational Level in Mexico

Forests play an essential role in climate change as they are the terrestrial ecosystems that store the highest C content in their soils and biomass. Despite this, the lack of information at the subnational level hinders their proper management and conservation. This study aimed to identify the extension and distribution of forests and to develop an empirical model for the spatial prediction of soil organic matter (SOM) in Ixtacamaxtitlan, Puebla, Mexico, based on environmental variables generated through Geographical Information Systems. A supervised classification in Landsat 8 images was used to define the forest cover, and environmental variables related to topography, climate and vegetation were generated. Finally, a Multiple Linear Regression model validated with the leave-one-out cross-validation method was used to examine the relationships between the covariates and the SOM and estimate its content in forest. The results show that the forest cover extension is 41%, with an overall accuracy of 97.7%. The model shows a good fit (R2cv = 0.69, RMSEcv = 1.53). The mean of SOM was 5.2%, and upper values were consistent with higher altitude, precipitation and cooler temperature. Estimating SOM content in forest areas is essential in developing planning strategies at the subnational level to mitigate the harmful effects of climate change.

High-Resolution Mapping of Soil Organic Matter at the Field Scale Using UAV Hyperspectral Images with a Small Calibration Dataset

The rapid acquisition of high-resolution spatial distribution of soil organic matter (SOM) at the field scale is essential for precision agriculture. The UAV imaging hyperspectral technology, with its high spatial resolution and timeliness, can fill the research gap between ground-based monitoring and remote sensing. This study aimed to test the feasibility of using UAV hyperspectral data (400–1000 nm) with a small-sized calibration sample set for mapping SOM at a 1 m resolution in typical low-relief black soil areas of Northeast China. The experiment was conducted in an approximately 20 ha field. For calibration, 20 samples were collected using a 100 × 100 m grid sampling strategy, while 20 samples were randomly collected for independent validation. UAV captured hyperspectral images with a spatial resolution of 0.05 × 0.05 m. The extracted spectra within every 1 × 1 m were then averaged to represent the spectra of that grid; this procedure was also performed across the whole field. Upon applying various spectral pretreatments, including absorbance conversion, multiple scattering correction, Savitzky–Golay smoothing filtering, and first-order differentiation, the absolute maximum values of the correlation coefficients of the spectra for SOM increased from 0.41 to 0.58. Importance analysis from the optimal random forest (RF) model showed that the characterized bands of SOM were located in the 450–600 and 750–900 nm regions. When the RF model was used, the UAV hyperspectra data (UAV-RF) were able to successfully predict SOM, with an R2 of 0.53 and RMSE of 1.48 g kg−1. The prediction accuracy was then compared with that obtained using ordinary kriging (OK) and the RF model based on proximal sensing (PS-RF) with the same number of calibration samples. However, the OK method failed to predict the SOM accuracy (RMSE = 2.17 g kg−1; R2 = 0.02) due to a low sampling density. The semi-covariance function was unable to describe the spatial variability of SOM effectively. When the sampling density was increased to 50 × 50 m, OK successfully predicted SOM, with RMSE = 1.37 g kg−1 and R2 = 0.59, and its results were comparable to those of UAV-RF. The prediction accuracy of PS-RF was generally consistent with that of UAV-RF, with RMSE values of 1.41 g kg−1 and 1.48 g kg−1 and R2 values of 0.57 and 0.53, respectively, which indicated that SOM prediction based on UAV-RF is feasible. Additionally, compared with the PS platforms, the UAV hyperspectral technology could simultaneously provide spectral information of tens or even hundreds of continuous bands and spatial information at the same time. This study provides a reference for further research and development of UAV hyperspectral techniques for fine-scale SOM mapping using a small number of samples.

A simple but effective evaluation criterion for parameters optimization of EPO and its application to moisture insensitive prediction of soil organic matter

Field measurement using portable NIR spectrometer is becoming a popular way to provide in situ, rapid estimation of soil organic matter content. However, in situ NIR spectra is altered by external environmental factors, such as soil moisture. The external parameter orthogonalization (EPO) has been recognized as one of the most effective method to minimize effect of soil moisture on spectra. In general, the root mean squared error of cross validation (RMSECV) was mainly used as an evaluation criterion to optimize the parameters involved in EPO algorithm, which ensures that the wet soil spectra corrected by EPO could have a good quantitative performance, but ignores the spectral similarity between the dry soil spectra and the wet soil spectra corrected by EPO. In this study, we present a simple but effective evaluation criterion for optimization of EPO, which we refer to as the mean of spectral correlation coefficient (MSCC). Compared with the commonly used RMSECV criterion, the spectra corrected by EPO (MSCC) are more similar to the corresponding dry soil spectra, and MSCC is more robust to different spectral pretreatment methods. In addition, the quantitative results of MSCC are equivalent to those of RMSECV, and sometimes slightly better than those of RMSECV.

Mineralogy, elemental composition, and organic matter prediction of wetland soils in the Cerrado biome, Brazil

Context The Cerrado, an area of savannah and wetlands in Brazil, is an important wet ecosystem characterised by organic-rich soils that has important environmental functions. In this work, we hypothesised that the total elemental composition obtained via portable X-ray fluorescence spectrometry (pXRF) could be useful a tool to predict soil organic matter. Aims We aimed to predict the total organic carbon (TOC) content and its fractions (carbon fulvic acid, C-FA; carbon humic acid, C-HA; carbon humin, C-HU) of soils occurring in representative Veredas biome from the Cerrado, Brazil. Methods We collected 144 soil samples from surface (0–20 cm) and subsurface (40–70 cm) layers from six representative Veredas environment located in the Cerrado bome, Triângulo Mineiro region, Brazil. These samples were analysed using a pXRF to obtain the total elemental composition. TOC, C-HA, C-FA, and C-HU fractions were determined by wet-oxidation and titration methods. The pXRF data was correlated with TOC and humic substances and later subjected to simple and multiple linear regression (MLR). Key results The major elements corroborated the mineralogy of wetland soils. TOC and humic substances, mainly C-HU, were reasonably predicted using pXRF data. All humic substances and TOC were strongly correlated to Pb. Thus, by parsimony, TOC could be reasonable predicted exclusively based on Pb concentration. Conclusions We demonstrated that pXRF can be an efficient tool to assess the elemental composition of the Cerrado wetlands soils contributing to still scarce inventory. Further, the elements germane to soil organic matter compounds can be used to predict TOC and its fractions. Implications A detailed characterisation of the Cerrado wetland soils in situ can be successfully performed using pXRF in further studies.

Remote Sensing Prediction Model of Cultivated Land Soil Organic Matter Considering the Best Time Window

Soil organic matter (SOM) is very important to the quality evaluation of cultivated land, especially in fertile black soil areas. Many studies use remote sensing images combined with different machine learning algorithms to predict the regional SOM content. However, the information provided by remote sensing images in different time windows is very different. Taking Youyi Farm, a typical black soil area in Northeast China, as the research area, this study obtains all available Sentinel-2 images covering the research area from 2019 to 2021, calculates the spectral index of single-phase and multi-temporal synthesis images, takes the spectral index and band of each image as the input, and employs the random forest regression algorithm to evaluate the performance of SOM prediction using remote sensing images with different time windows. The results show that: (1) the accuracy of SOM prediction using image band and spectral index is generally improved compared to using only the band; (2) when using single-phase images, the R2 range of SOM prediction using image band and spectral index is from 0.16 to 0.59 and the RMSE ranges from 0.82% to 1.23%; When using multi-temporal synthesis images, the R2 range of SOM prediction using image band and spectral index is from 0.18 to 0.56 and the RMSE ranges from 0.85% to 1.19%; (3) the highest accuracy of SOM prediction using synthetic images is lower than that of single-phase images; (4) the best time window of the bare soil period in the study area is May. This study emphasizes the importance of the time window to SOM prediction. In subsequent SOM prediction research, remote sensing images with appropriate time windows should be selected first, and then the model should be optimized.

Mapping Soil Organic Matter Content during the Bare Soil Period by Using Satellite Data and an Improved Deep Learning Network

Soil function degradation has impaired global work in the implementation of sustainable development goals (SDGs), and soil organic matter (SOM) is a basic and the most important indicator. The deep learning neural network (i.e., DL network) has become a popular tool for mapping SOM content at a regional scale. However, outlier sample data caused by environmental factors (e.g., moisture and vegetation) and uncertain noise (e.g., random noise and instability effects) have interfered with the determination of the function mapping relationship between target soil properties and spectral features, leading to DL networks with low generalization capability. Therefore, we introduced a spatial association module into a deep neural network to remove outlier sample data in order to construct an optimal sample set for calibrating an improved deep learning (IDL) network for SOM mapping. A total of 707 soil samples and a Sentinel-2B multispectral image were acquired during the bare soil period in Weibei, China. The variable importance in the projection approach was used to select the SOM-responsive spectral features for model inputs. Measured SOM contents were taken as the dependent variable, and the IDL network was constructed and applied to map SOM at the regional scale. The results showed that Band 11 was the most important band for SOM prediction. The band difference transformation method was able to integrate multiple-band information and enhance the absorption signal of SOM. The optimal SOM-responsive spectral features included B11, B1–B11, B2−B11, B3−B11, B4−B11, and B1−B12. The IDL network exhibited better performance (R2 = 0.92; RPIQ = 4.57) regarding SOM estimation compared with the DL model performance (R2 = 0.84; RPIQ = 2.84), being improved by 9.52% (for R2) and 60.92% (for RPIQ). After introducing the spatial association module, the DL network generalization capability was enhanced. SOM distribution showed a high-value (&gt;20 g kg−1) area in the south, and a low-value (&lt;6 g kg−1) area in the north of the study area (the area affected by seawater intrusion). These results provide a strategy based on an IDL network and satellite data for effectively and accurately mapping SOM at the regional scale.

Soil organic matter prediction with benchtop and implement‐mounted optical reflectance sensing approaches

AbstractProximal sensing technologies can densely quantify soil organic matter (OM) variability using visible and near infrared (VNIR) reflectance spectroscopy. However, issues with global calibrations and varying water content can reduce accuracy. Therefore, research was conducted to determine OM prediction accuracy across soil volumetric water content (VWC) with (a) a planter‐mounted optical sensor and (b) multiple combinations of reflectance bands within the VNIR spectrum. Ninety soils collected across Missouri and Illinois were used in the study. Data were collected at three VWC with the Precision Planting SmartFirmer and a benchtop spectrometer. SmartFirmer OM was derived through an internal algorithm, while spectral pre‐processing and machine learning were used for OM prediction in all approaches that used reflectance from the benchtop spectrometer. Results found that SmartFirmer OM predictions were affected by soil VWC, with predicted OM decreasing with increasing VWC. Findings from three modeling approaches using the benchtop spectrometer showed that a continuous spectrum (i.e., 400–1,500 nm) improved performance (RMSE = 5.25 g kg–1) over the discrete waveband approach (RMSE = 9.23 g kg–1). Furthermore, including the entire VNIR region (400–2,500 nm) resulted in the best predictive capability (RMSE = 1.42 g kg–1). However, because a full‐spectrum approach increases cost, using reflectance from 400 to 1,500 nm along with spectral pre‐processing and machine learning may be an acceptable method for estimating OM. These findings contribute to the improvement of commercially available proximal sensors that may be used to monitor soil carbon (C) stocks, assess soil health, or for other precision agriculture applications.

Precise prediction of soil organic matter in soils planted with a variety of crops through hybrid methods

Determining the spatial distribution of soil organic matter (SOM) of cultivated lands efficiently and quickly using available data is crucial for the development of smart agriculture with low consumption and high efficiency. In this study, a total of 205 black soil samples from Hulunbuir, China, planted with a variety of crops were investigated for multiple soil nutrients. The SOM content in the studied black soil was rich, ranging from 32.1 to 62.8 g/kg. In addition, machine learning models based on spectral data and environmental variables were established to predict the overall spatial distribution of SOM. The prediction effect of the Random Forest (RF) model was the best (R2 = 0.84, CCC = 0.872, RMSE = 3.31, RPD = 2.19), and the environmental variables (mainly clay index) containing more information in recursive feature elimination had the greatest influence on the prediction accuracy of SOM. Further, to reduce the dependence on predictive indicators, this study quantitatively analyzed the disturbance factors (crop types) of cultivated land using Bayesian Maximum Entropy (BME), and the BME-RF model was established to combine crop types with prediction indicators to further optimize SOM prediction accuracy. The prediction accuracy of the model was improved to varying degrees in multiple groups of experiments by adding several groups of soft data to compare and analyze the effect of the BME-RF model, and it was the highest (R2 = 0.89, RI = 15.5%) when 1200 soft data were used. Crop types have certain influence on the pattern of SOM content. The SOM content was relatively low in rapeseed and barley planting areas, but was relatively high in potato and beet growing areas. In this paper, BME-RF method is innovated to combine historical data with as little experimental data as possible, which greatly improved the accuracy of digital map of soil nutrients. This study can provide a scheme for the development of smart agriculture and promote the sustainability of cultivated land environment.

Fine-Scale Mapping of Soil Organic Matter in Agricultural Soils Using UAVs and Machine Learning

The fine-scale mapping of soil organic matter (SOM) in croplands is vital for the sustainable management of soil. Traditionally, SOM mapping relies on laboratory methods that are labor-intensive and costly. Recent advances in unmanned aerial vehicles (UAVs) afford new opportunities for rapid and low-cost SOM mapping at the field scale. However, the conversion from UAV measurements to SOM maps requires specific transfer models that still rely on local sampling. This study aimed to develop a method for predicting topsoil SOM at a high resolution on the field scale based on soil color information gained from low-altitude UAV imagery and machine learning. For this, we performed a UAV survey in cropland within the German loess belt. We used two fields, one for training and one for validation of the model, to test the model transferability. We analyzed 91 soil samples for SOM in the laboratory for the model calibration and 8 additional samples for external model validation. A random forest model (RF) showed good performance for the prediction of SOM based on UAV-derived color information with an RMSE of 0.13% and with an RPIQ of 2.42. The RF model was used to predict SOM at a point-support of 1 × 1 m. The SOM map revealed spatial patterns within the fields with a uniform spread of the prediction uncertainty. The validation of the model performed similarly to the calibration with an RMSE of 0.12% and an RPIQ of 2.05, albeit with a slight bias of 0.05%. This validation using external data showed that prediction models are transferable to neighboring fields, thus permitting the prediction on larger scale farms or enabling carbon monitoring over time.