Prediction Of Soil Properties Research Articles

Evaluating three calibration transfer methods for predictions of soil properties using mid‐infrared spectroscopy

AbstractMid‐infrared (MIR) spectroscopy models have been developed for rapid assessment of soils but are often soil and instrument specific because of differences in laboratory conditions and sensor setup. Calibration transfer is required to apply a spectral model such as partial least squares (PLS) regression developed from a primary instrument to a spectral dataset measured by a secondary instrument with statistically retained accuracy and precision. The study aimed to compare the performance of three transfer methods (i.e., direct standardization [DS], piecewise direct standardization [PDS], and spectral space transfer [SST]) and investigate the effects of transfer sample size and sample selection methods. The transfer methods were developed for predicting total C, clay, silt, and sand contents, cation exchange capacity (CEC), pH in water (pHW) and CaCl2, CaCO3 equivalent, and −1,500‐kPa water retention using spectral measurements of a secondary instrument. Calibration transfer methods of three PLS models for estimating soil properties with a high (total C), intermediate (clay content), and low (pHW) predictability were discussed. The effect of sample size required for the development of the calibration transfer and the selection method of the transferred samples were investigated. It was found that SST was most favorable for a relatively small sample size used in calibration transfer (≤12 samples). The performance of transfer methods was optimal when the transfer samples accounted for the variability of MIR spectra from the secondary instrument. We conclude that SST and PDS have the potential to be applied in spectroscopy for predicting soil properties using secondary instruments.

Regional ensemble modeling reduces uncertainty for digital soil mapping

Recent country and continental-scale digital soil mapping efforts have used a single model to predict soil properties across large regions. However, different ecophysiographic regions within large-extent areas are likely to have different soil-landscape relationships so models built specifically for these regions may more accurately capture these relationships relative to a ‘global’ model. We ask the question: Is a single ‘global’ model sufficient or are regionally-specific models useful for accurate digital soil mapping? We test this question by modeling soil depth classes across the 432,000 km2 upper Colorado River Basin in the Western USA using a single global model, multiple ecophysiographic models, and ensembles of the ecophysiographic models.Effective soil depth class observations (n = 12,194) were derived from multiple soil databases. Fifty-seven environmental covariates were derived from a 30 m digital elevation model, climate data, satellite imagery, and aeroradiometric data. Three independent land classifications were used to stratify the area. Two expert-derived land classifications, USDA Major Land Resource Areas (MLRA) and US-EPA Level III ecoregions, divided the study area into multiple ecophysiographic regions based on vegetation and broad-scale physiographic differences. The third land classification divided the study area into broad landforms.Soil depth observations were split into separate training (n = 10,470) and validation (n = 1,724) datasets. First, a ‘global’ random forest model was used to model soil depth classes using all training observations and covariates. ‘Global’ denotes a model built with all training data across the extent of the area, not a model at world extent. Second, the land classifications were used to subset the observations into ecophysiographic sub-datasets and random forest models were refit for each region. Models fit by ecophysiographic region are referred to as regional models. Thirdly, predictions from each regional model were fused into regional-ensemble models. Accuracy, Brier scores, and Shannon’s entropy were used to compare model accuracy and uncertainty. Regional ecophysiographic models were also compared to models built for geographic areas that were defined solely to be approximately equal in area. Training dataset density and the imbalance ratio were investigated to determine if data characteristics influenced regional accuracy/uncertainty metrics.Accuracy for the global model using the validation set was 62.8%. Regional model accuracies ranged between 56.1% and 75.0%. We found: 1) useful inter-regional differences in global model accuracy were revealed when the global model was validated by region, 2) no consistent relationship between training observation density and accuracy/uncertainty metrics, 3) no meaningful differences in accuracy and uncertainty metrics between physiographic and geographic regions, 4) ensembles of regionally-specific models were approximately as accurate as global models, and 5) both region-specific models and ensembles of regional models were less uncertain than the global model. Overall, we recommend the use of soil depth class predictions made from MLRA regional ensemble models because this prediction had higher accuracy than the ecoregion ensemble model prediction, but lower uncertainty than both the global model and the landform ensemble model predictions. We answer our question: Ensembles of regionally-specific models are approximately as accurate as global models, but result in less uncertainty.

A dual search-based EPR with self-adaptive offspring creation and compromise programming model selection

Evolutionary polynomial regression (EPR) is extensively used in engineering for soil properties modeling. This grey-box technique uses evolutionary computing to produce simple, transparent and well-structured models in the form of polynomial equations that best explain the observed data. A key task is then to determine mathematical structures for modeling physical phenomena and to select the optimal EPR model. This requires an algorithm to search through the model structure space and successfully produce feasible solutions that honor a set of statistical metrics. The complexity of EPR models increases greatly, however, with the number of polynomial terms used to tune these models. In this paper, we propose an alternative EPR for modeling complex soil properties. We implement a dual search-based EPR with self-adaptive offspring creation as model structure search strategy and couple a compromise programming tool to select a model that is preferred statistically relative to models with different polynomial terms. We illustrate our method using real-world data to improve predictions of optimal moisture content and creep index for soils. Our results demonstrate that the models derived using the proposed methodology can predict soil properties with adequate accuracy, physical meaning and lower number of parameters and input variables.

Retrieval and Mapping of Soil Organic Carbon Using Sentinel-2A Spectral Images from Bare Cropland in Autumn

Soil is the largest carbon reservoir on the terrestrial surface. Soil organic carbon (SOC) not only regulates global climate change, but also indicates soil fertility level in croplands. SOC prediction based on remote sensing images has generated great interest in the research field of digital soil mapping. The short revisiting time and wide spectral bands available from Sentinel-2A (S2A) remote sensing data can provide a useful data resource for soil property prediction. However, dense soil surface coverage reduces the direct relationship between soil properties and S2A spectral reflectance such that it is difficult to achieve a successful SOC prediction model. Observations of bare cropland in autumn provide the possibility to establish accurate SOC retrieval models using the S2A super-spectral reflectance. Therefore, in this study, we collected 225 topsoil samples from bare cropland in autumn and measured the SOC content. We also obtained S2A spectral images of the western Guanzhong Plain, China. We established four SOC prediction models, including random forest (RF), support vector machine (SVM), partial least-squares regression (PLSR), and artificial neural network (ANN) based on 15 variables retrieved from the S2A images, and compared the prediction accuracy using RMSE (root mean square error), R2 (coefficient of determination), and RPD (ratio of performance to deviation). Based on the optimal model, the spatial distribution of SOC was mapped and analyzed. The results indicated that the inversion model with the RF algorithm achieved the highest accuracy, with an R2 of 0.8581, RPD of 2.1313, and RMSE of 1.07. The variables retrieved from the shortwave infrared (SWIR) bands (B11 and B12) usually had higher variable importance, except for the ANN model. SOC content mapped with the RF model gradually decreased with increasing distance from the Wei river, and values were higher in the west than in the east. These results matched the SOC distribution based on measurements at the sample sites. This research provides evidence that soil properties such as SOC can be retrieved and spatially mapped based on S2A images that are obtained from bare cropland in autumn.

Using autoencoders to compress soil VNIR–SWIR spectra for more robust prediction of soil properties

In the past two decades, research efforts have focused on using near infrared diffuse reflectance spectroscopy (350–2500nm) as a rapid and cost-effective method for soil analysis. Given the multi-faceted importance of the soil ecosystem, and considering the increasing pressures exerted upon it due to climate change, degradation and urbanization, the advantages of soil spectroscopy are significant. Large soil spectral libraries have been developed to this end throughout the world. The soil matrix is however complex, posing a challenge in the determination of key soil properties from the spectra. To tackle this challenge, two methodologies are generally used: a) the use of spectral pre-processing techniques to transfer the spectra into a new space in which the association between spectrum and soil property is supposed to be more clear, and b) the use of more sophisticated machine learning models (e.g. deep learning). In this paper, we propose a novel methodology using stacked autoencoders to transform the initially recorded spectra in a new compressed (i.e. latent) space which can help the chemometric models enhance the accuracy of prediction. This is an unsupervised learning approach which only depends on the input data (i.e. the spectra). Following the significant results obtained in the literature using combinations of different spectra pre-processing techniques and the simultaneous prediction of multiple soil properties, the proposed methodology is extended to facilitate these approaches. We demonstrate this capacity by applying it in the mineral samples of the LUCAS 2009 topsoil database, and simultaneously predicting eight properties (the particle size distribution, pH, CEC, organic carbon, calcium carbonate, and total nitrogen) using an artificial neural network. Compared to standard pre-processing techniques and other transformations such as the principal components space, the proposed methodology using only one spectral source as input decreases the RMSE on average by 8.4% and by 3.5%, respectively. With respect to the current state-of-the-art and in particular a multi-input convolutional neural network which was recently proposed and outperformed the compared methodologies, the results of our multi-input methodology exhibit an average RMSE decrease of 9.9%. The interpretability aspect of the transformed feature space and the compressed spectra was also examined to identify how the compressed information encodes the input data and enables better associations between input and output.

Spatial variability and mapping of soil fertility status in a high-potential smallholder farming area under sub-humid conditions in Zimbabwe

A study was conducted to examine spatial variability of soil properties related to fertility in maize fields across varying soil types in ward 10 of Hurungwe district, Zimbabwe; a smallholder farming area with sub-humid conditions and high yield potential. Purposively collected and geo-referenced soil samples were analyzed for texture, pH, soil organic carbon (OC), mineral N, bicarbonate P, and exchangeable K. Linear mixed model was used to analyze spatial variation of the data. The model allowed prediction of soil properties at unsampled sites by the empirical best linear unbiased predictor (EBLUP). Evidence for spatial dependence in the random component of the model was evaluated by calculating Akaike’s information criterion. Soil pH ranged from 4.0 to 6.9 and showed a strong spatial trend increasing from north to south, strong evidence for a difference between the home and outfields with homefields significantly higher and between soil textural classes with the sand clay loam fraction generally higher. Soil OC ranged from 0.2 to 2.02% and showed no spatial trend, but there was strong evidence for a difference between home and outfields, with mean soil OC in homefields significantly larger, and between soil textural classes, with soil OC largest in the sandy clay loams. Both soil pH and OC showed evidence for spatial dependence in the random effect, providing a basis for spatial prediction by the EBLUP, which was presented as a map. There were significant spatial trends in mineral N, available P and exchangeable K, all increasing from north to south; significant differences between homefields and outfields (larger concentrations in homefields), and differences between the soil textural classes with larger concentrations in the sandy clay loams. However, there was no evidence for spatial dependence in the random component, so no attempt was made to map these variables. These results show how management (home fields vs outfields), basic soil properties (texture) and other factors emerging as spatial trends influence key soil properties that determine soil fertility in these conditions. This implies that the best management practices may vary spatially, and that site-specific management is a desirable goal in conditions such as those which apply in Ward 10 of Hurungwe district in Zimbabwe.

Soil Properties Prediction for Precision Agriculture Using Visible and Near-Infrared Spectroscopy: A Systematic Review and Meta-Analysis

Reflectance spectroscopy for soil property prediction is a non-invasive, fast, and cost-effective alternative to the standard laboratory analytical procedures. Soil spectroscopy has been under study for decades now with limited application outside research. The recent advancement in precision agriculture and the need for the spatial assessment of soil properties have raised interest in this technique. The performance of soil spectroscopy differs from one site to another depending on the soil’s physical composition and chemical properties but it also depends on the instrumentation, mode of use (in-situ/laboratory), spectral range, and data analysis methods used to correlate reflectance data to soil properties. This paper uses the systematic review procedure developed by the Centre for Evidence-Based Conservation (CEBC) for an evidence-based search of soil property prediction using Visible (V) and Near-InfraRed (NIR) reflectance spectroscopy. Constrained by inclusion criteria and defined methods for literature search and data extraction, a meta-analysis is conducted on 115 articles collated from 30 countries. In addition to the soil properties, findings are also categorized and reported by different aspects like date of publication, journals, countries, employed regression methods, laboratory or in-field conditions, spectra preprocessing methods, samples drying methods, spectroscopy devices, wavelengths, number of sites and samples, and data division into calibration and validation sets. The arithmetic means of the coefficient of determination (R2) over all the reports for different properties ranged from 0.68 to 0.87, with better predictions for carbon and nitrogen content and lower performance for silt and clay. After over 30 years of research on using V-NIR spectroscopy to predict soil properties, this systematic review reveals solid evidence from a literature search that this technology can be relied on as a low-cost and fast alternative for standard methods of soil properties prediction with acceptable accuracy.

Comparing CalReg performance with other multivariate methods for estimating selected soil properties from Moroccan agricultural regions using NIR spectroscopy

Nowadays, near-infrared spectroscopy (NIRS) is widely used as an alternative method for analysing soil attributes. The successful determination of soil attributes by this method depends on the selection of an appropriate multivariate calibration technique. In this study, four multivariate techniques, including partial least squares regression (PLSR), memory-based learning (MBL), backpropagation neural network (BPNN) and a new proposed method named ‘combined classification & regression method’ (CalReg) were compared in terms of accuracy in predicting soil properties, including extractable calcium (Caex), extractable iron (Feex), zinc (Zn), extractable magnesium (Mgex), extractable potassium (Kex), phosphorus extracted by Olson method (P-Olson), calcium carbonate (CaCO3), soil organic carbon (SOC), total nitrogen (TN), clay, silt, sand, cation exchange capacity (CEC), and pH. A total of 660 soil samples collected from six Moroccan agricultural regions were used as the dataset for the laboratory’s calibration and validation procedure. Prediction accuracy was quantified using four statistical metrics of accuracy including the coefficient of determination (R2), root mean squared error (RMSE), the ratio of performance to interquartile distance (RPIQ) and residual prediction deviation (RPD). Both calibration and data test sets showed that the CalReg model outperformed the other tested models for SOC, TN, Caex, CEC, P-Olson, Kex and pH predictions. At the same time, MBL outperformed the other models for Feex, Zn, silt, sand and clay prediction. Moreover, BPNN outperformed the other models for Mgex and CaCO3 predictions. The best predictions were obtained by the CalReg model for SOC (R2ts ​= ​0.92; RMSEts ​= ​0.29; RPD ​= ​4.73, RPIQ ​= ​5.71), TN (R2ts ​= ​0.90; RMSEts ​= ​364.80; RPD ​= ​3.39, RPIQ ​= ​3.16), Caex (R2ts ​= ​0.88; RMSEts ​= ​3.25; RPD ​= ​3.33, RPIQ ​= ​4.40), CaCO3 (R2ts ​= ​0.89; RMSEts ​= ​3.86; RPD ​= ​3.18, RPIQ ​= ​3.95), and clay (R2ts ​= ​0.88; RMSEts ​= ​3.95; RPD ​= ​3.25, RPIQ ​= ​4.81). Furthermore, the results showed that CalReg helps improve the accuracy of soil property prediction. NIRS combined with CalReg has great potential to determine accurately the selected soil properties in the laboratory.

A novel framework to estimate soil mineralogy using soil spectroscopy

Soil minerals are usually quantified by the conventional laboratory soil analyses. However, developments in interpretations and analyses of the visible, near-infrared, and short-infrared (Vis-NIR-SWIR) diffuse reflectance have allowed the quantification of some soil minerals. In this study, we aimed to implement a novel framework using Vis-NIR-SWIR spectroscopy to quantify the main soil minerals. We also assessed the application of this framework to create new environmental variables for digital soil mapping (DSM). The soil spectra database comprised 2701 samples from 1008 sites in the spectral range of 350–2500 nm at 0–20, 40–60, and 80–100 cm depths. The specific bands in the Vis-NIR-SWIR spectra that identify the presence of soil mineral were selected based on the literature with the United States Geological Survey Spectral Library Version 7 and in the strong maxima and minima of the second-derivative curves of the soil mineral standards using the Savitzky-Golay method. We proposed an estimation and conversion of the measurement unit of soil minerals in amplitude to g kg−1 using a small dataset of mineral content quantified via X-Ray Powder Diffraction. We selected randomly 85 samples out of 2701 available at 0–20 cm depth and sent to conventional laboratory analyses to calibrate the final estimation, using the kaolinite soil mineral as an example. Therefore, a constant factor was determined to estimate mineral content in soils displaying RMSE, R2adj, the Lin's concordance coefficient (CCC), Bias, and RPIQ values of 7612 g kg−1, 0.28, 0.50, 13.09 g kg−1, and 0.56, respectively. This evaluation was assessed by splitting 85 samples into 80% to determine and 20% to validate the constant factor. For the DSM procedure, we used 2701 samples split into 80% and 20% for calibration and validation, respectively, of the models for each of the nine minerals. This study showed that the proposed framework using Vis-NIR-SWIR spectroscopy to estimate soil minerals is promising due to higher CCC and lower RMSE values obtained. Furthermore, the spectral amplitude for each mineral provides important information to be used as environmental variables for the prediction of soil attributes, soil types, and soil properties.

Digital mapping of soil chemical properties with limited data in the Thung Kula Ronghai region, Thailand

Prediction of soil chemical properties has significant implications for land management in Thailand but it is especially challenging in vast areas with limited soil data. In this study we identified important spectral and terrain indices for predicting various soil properties, evaluated the suitability of the digital soil mapping (DSM) technique for creating digital soil maps, and assessed the soil nutrients levels in the Thung Kula Ronghai (TKR) region of Thailand. A total of 186 soil samples were collected at 0–30 cm depth and analyzed for nutrients. A digital elevation model with 5 m resolution was used to derive the terrain variables of the study area. Landsat-8 images collected at bare soil conditions with 30 m resolution were used to determine the soil and vegetation indices. Models developed to predict soil properties using multiple linear regression (MLR) were evaluated in terms of the coefficient of determination, root mean square error and normalized root mean square error. We found that indices such as brightness, saturation, coloration, normalized difference water, and moisture stress are the most important predictor variables, significantly correlated with various soil properties. The accuracy of the MLR models developed for predicting soil properties in this study suggests that the DSM technique could be useful to predict soil nutrient status in the TKR region.

Spatial prediction of soil properties for the Busia area, Kenya using legacy soil data

Legacy soil data often contains considerable agronomic information that can help revitalize agriculture in countries with poor spatial data infrastructures. The objective of this study was to determine whether existing legacy soil data could be used to quantitatively predict soil properties at a higher resolution than the original legacy soil map using digital soil mapping techniques without conducting additional field work. A dataset of 76 soil profiles mined from the Reconnaissance Soil Survey of the Busia Area in Western Kenya together with a total of 23 environmental covariates were used to capture the soil forming factors and test the effectiveness of the individual predictive soil mapping (iPSM) technique to predict soil organic carbon, clay, silt, and sand over large areas in cases where data is limited. iPSM was compared to the stepwise multiple linear regression (SMLR) under two scenarios and ordinary kriging. In scenario one, SMLR prediction was achieved using the 23 original environmental soil covariates as predictors whereas in scenario two, SMLR prediction was achieved using principal components, generated from the 23 original covariates, as predictors. Ordinary kriging used only the soil property data for prediction. Ordinary kriging outperformed all the prediction models (RMSE = 0.02) for SOC, (RMSE = 0.30) for clay, (RSME = 0.10) for silt, and (RMSE = 0.32) for sand. This was attributed to the proximity between the calibration and the evaluation datasets and that these two datasets were not entirely independent. Ordinary kriging maps, however, provided very little spatial detail. Considering only those prediction models that used soil covariates, iPSM did not perform well in predicting soil properties compared to SMLR under the two scenarios. Regardless of the poor performance from iPSM relative to the other prediction models, SMLR (under the two scenarios) and ordinary kriging, iPSM best captured soil properties' spatial variability. A combination of the four models, using model averaging methods, resulted in a better prediction using the equal weights method for clay (%) (RMSE = 14.31), silt (%) (RMSE = 9.00), and sand (%) (RMSE = 16.67). On the contrary, SOC content (%) prediction from SMLR under the two scenarios performed best (RMSE = 0.43). Predicted soil property maps were at a resolution of 30 m, better than the current soil property maps for the study area with a 250 m resolution. The ability of predicted soil property maps to capture the variability of soils in our study area, except for ordinary kriging, supports the overall objective of this study. The sample size within legacy soil data, however, limits the use of complex DSM techniques. Therefore, efforts should continue to ensure that information mined from legacy soil data can be used to bolster the efforts made by various soil agencies to create harmonized soil profile databases to increase their usefulness for complex DSM methods.

Does the limited use of orthogonal signal correction pre-treatment approach to improve the prediction accuracy of soil organic carbon need attention?

Visible-Near-infrared (Vis-NIR) spectroscopy is a relatively modern method that can be used to predict soil properties such as soil organic carbon (SOC). Predictions of soil properties with Vis-NIR requires pre-processing algorithms. Applying a wrong type or applying too extreme pre-processing algorithms may result in the removal of valuable information or may even introduce unwanted variations. This can negatively affect the prediction accuracy of the property being studied. Orthogonal signal correction (OSC) pre-processing method has been used in other fields for visible and near infrared spectra improvement. However, the application of OSC application in soil science remains limited. The main idea behind OSC is the removal of only unwanted variation from the spectrum unlike some other pre-treatment methods which are believed to remove valuable information in the process of removing undesirable variation. This study verifies the effectiveness of the OSC against nine commonly used pre-treatment methods across three different agricultural fields for both lab-dry and in-field spectra. For the prediction, partial least square regression (PLSR) and support vector machine regression (SVMR) algorithms were used. In this study, the OSC method overall improved prediction accuracy the most (e.g. with OSC the best result was R2CV = 0.79, without OSC the best result was R2CV = 0.62) and is therefore a promising tool that should be included in further studies on different soils and other soil properties.

Convolutional Neural Network approach for the prediction of Soil texture properties

Background/Objectives: The main objective is to achieve improved performance of soil properties prediction for hyperspectral data. In this work, convolutional neural network is trained to understand the pattern of hyperspectral data by spatial interpolation. Methods/Statistical analysis: The proposed methodology is used to predict six soil properties- Organic Carbon content (OC), Cation Exchange Capacity (CEC), Nitrogen Content (N), pH level in water, Clay particle and Sand Particle. Soil texture which defines the relative content of soil particles is determined by the percentage of clay, sand and silt in the soil. The input to the Convolutional Neural Network (CNN) is the Hyperspectral data in the form of multiple arrays. The statistical evaluation of model performance is evaluated using root-mean-square error and r square. Findings: In this research, deep learning approach is used to capture the pattern hidden in the soil. Deep learning is a kind of neural network which can model complex relationship for representing non-linearity for a scalable data. The main challenge is predicting a soil type, as it involves complex structural characteristics and soil features. Novelty/Improvements: The performance of soil texture prediction is improved by automatic feature learning capability in the proposed CNN model. The average rmse value obtained in proposed method for all the six soil texture properties is 5.68%. Keywords: Soil texture; convolutional neural network; hyperspectral data; deep learning

Improving the predictions of soil properties from VNIR–SWIR spectra in an unlabeled region using semi-supervised and active learning

Monitoring the status of the soil ecosystem to identify the spatio-temporal extent of the pressures exerted and mitigate the effects of climate change and land degradation necessitates the need for reliable and cost-effective solutions. To address this need, soil spectroscopy in the visible, near- and shortwave-infrared (VNIR–SWIR) has emerged as a viable alternative to traditional analytical approaches. To this end, large-scale soil spectral libraries coupled with advanced machine learning tools have been developed to infer the soil properties from the hyperspectral signatures. However, models developed from one region may exhibit diminished performance when applied to a new, unseen by the model, region due to the large and inherent soil variability (e.g. pedogenetical differences, diverse soil types etc.). Given an existing spectral library with labeled data and a new unlabeled region (i.e. where no soil samples are analytically measured) the question then becomes how to best develop a model which can more accurately predict the soil properties of the unlabeled region.In this paper, a machine learning technique leveraging on the capabilities of semi-supervised learning which exploits the predictors’ distribution of the unlabeled dataset and of active learning which expertly selects a small set of data from the unlabeled dataset as a spiking subset in order to develop a more robust model is proposed. The semi-supervised learning approach is the Laplacian Support Vector Regression following the manifold regularization framework. As far as the active learning component is concerned, the pool-based approach is utilized as it best matches with the aforementioned use-case scenario, which iteratively selects a subset of data from the unlabeled region to spike the calibration set. As a query strategy, a novel machine learning–based strategy is proposed herein to best identify the spiking subset at each iteration. The experimental analysis was conducted using data from the Land Use and Coverage Area Frame Survey of 2009 which covered most of the then member-states of the European Union, and in particular by focusing on the mineral cropland soil samples from 5 different countries. The statistical analysis conducted ascertained the efficacy of our approach when compared to the current state-of-the-art in soil spectroscopy.

Improved prediction of soil properties with multi-target stacked generalisation on EDXRF spectra

Energy dispersive X-ray fluorescence (EDXRF) is one of the most quick, environmentally friendly and least expensive spectroscopic analytical methodologies for assessing soil quality parameters. However, challenges in EDXRF spectral data analysis still demand more efficient methods. One possible solution is using Machine Learning (ML), particularly Multi-target Regression (MTR) methods, which predict multiple parameters taking advantage of inter-correlated parameters. In this study, we proposed the Multi-target Stacked Generalisation (MTSG), a novel MTR method relying on learning from different regressors in stacking structure for a boosted outcome. We compared MTSG and 5 MTR methods for predicting 10 parameters of soil fertility. Random Forest and Support Vector Regression (SVR) were used as learning algorithms embedded into each MTR method. Results showed the superiority of MTR methods over the Single-target Regression (the traditional ML method), reducing the predictive error for 5 parameters. Particularly, MTSG obtained the lowest error for phosphorus, total organic carbon and cation exchange capacity. When observing the relative performance of SVR with a radial kernel, the prediction of base saturation percentage was improved by 19%. Finally, the proposed method was able to reduce the average error from 0.67 (single-target) to 0.64 analysing all targets, representing a global improvement of 4.48%.

Mapping soil properties with soil-environmental covariates using geostatistics and multivariate statistics

The spatial modelling of soil properties provides us with essential and useful information relevant to soil fertility management and environmental protection. The study aims to investigate the ability of empirical Bayesian kriging and principal component analysis, multiple linear regressions with environmental covariates in the modelling of soil properties distribution. For this study, thirty (n = 30) soil samples were obtained at 0–30 cm depth and nine (9) soil-environmental covariates derived from the digital elevation model (Shutter Radar Topography Mission at 30 m resolution) in southeastern Nigeria. The summary statistics revealed high sand content (> 70%) which revealed that the soils of the humid tropics developed on the coastal plain parent material are coarse-textured. Pearson correlation matrix revealed a significant but weak correlation between soil properties and soil-environmental variables. Using empirical Bayesian kriging interpolation, the cross-validation results revealed an acceptable prediction for magnesium, potassium, phosphorus, pH and total nitrogen (R2 > 0.5 with RMSE closer to 0). The principal component analysis reveals that principal component 1 to principal component 5 could interpret 78.1% of the total variability of soil properties. Modelling each soil property using multiple linear regression with the derived soil-environmental covariates, the study noted that only magnesium gave the best model fit with 50.9% of the soil-environmental covariates explaining its variability, while other soil properties presented unacceptable models. Therefore, to improve soil property prediction through multiple linear regression, more observation points are recommended to interpret better the performance of multiple linear regression over flat terrain system.

Agriculture environment monitoring: rapid soil fertility evaluation by means of near infrared spectroscopy

In agricultural practices, the development of environmental controlling programs is essential in maintaining soil fertility. It should be monitored constantly in real time situation by predicting soil properties such as nutrients and mineral contents. The main purpose of this presented paper is employ the near infrared spectroscopy (NIRS) technique in determining nitrogen (N) and magnesium (Mg) content of agricultural soil rapidly and simultaneously. Near infrared spectra data of soil samples were obtained by acquiring absorbance data in wavelength range from 1000 to 2500 nm using a portable sensing device NIR with a photodiode array detector. Spectra data were enhanced using mean normalization approach to improve prediction accuracy. Further, prediction models were developed by means of partial least square regression (PLSR) followed by cross validation. The results showed that both soil fertility properties (N and Mg) can be predicted by NIRS with maximum correlation coefficient (r) were 0.93 and 0.91 for N and Mg respectively. It may conclude that NIRS technique was able to be used in agriculture environment monitoring especialy to predict soil fertility properties.

Prediction of soil calcium carbonate with soil visible-near-infrared reflection (Vis-NIR) spectral in Shaanxi province, China: soil groups vs. spectral groups

ABSTRACT Soil visible-near-infrared reflection (Vis-NIR; 350 to 2500 nm) spectra is a comprehensive reflection of various soil physical and chemical properties, resulting in significant differences of hyperspectral characteristics and diversity of soil property prediction models. Calcium carbonate regulates several soil properties that were widely used to describe soil types and to quantify vulnerability to erosion. Therefore, spectral-based determination of soil calcium carbonate content is essential for agricultural management and environmental evaluation. Based on their reflectance similarity, the soil Vis-NIR spectra data itself can be classified into different spectral groups to predict soil calcium carbonate. To obtain the best hyperspectral quantitative prediction model for soil calcium carbonate based on soil group spectral data and spectral group data, a total of 246 soil samples were collected from seven soil types in Shaanxi province, China, and the soil Vis-NIR spectra were measured. All the soil spectral data were smoothed using Savitzky-Golay and continuum removal method. In this paper, we combined spectral angle cosine (SAC) algorithm with spectral correlation coefficient (SCC) algorithm to classify soil spectra. The results indicated that the coefficient of correlations between measured and predicted soil properties for seven prediction models based on random regression forest (RFR) were low (Ratio of performance to deviation, RPD < 2.00). When soil spectra were classified into three types based on the SAC-SCC spectral similarity measurement method, accuracy of all the prediction models increased significantly (R 2 ≥ 0.96; RPD > 2.00) compared with the individual soil type models. These results provide a novel method for soil spectral classification and soil attribute estimation.

Integration of multispectral and hyperspectral data to map magnetic susceptibility and soil attributes at depth: A novel framework

The understanding of attributes and magnetic susceptibility (χ) at soil surface, mainly subsurface, is crucial due to their role to identify climate changes, soil degradation, soil classification systems, soil fertility, and pedogenesis. The integration of proximal sensing (PS) and remote sensing (RS) data sources could increase the efficiency of Digital Soil Mapping. Nevertheless, products of this integration need to be evaluated in hybrid, stochastic, and deterministic models to predict soil attributes and χ at surface and subsurface. This study investigates the PS and RS integration by applying four deterministic (e.g. Bayesian Regularised Neural Network, Generalised Linear Model, Random Forest and Cubist) and hybrid models (e.g. Regression Kriging of residuals of the best-fitted model) to create a new environmental variable, the Best Synthetic Soil Image (BSSI), at three soil depths (e.g. 0 – 20, 40 – 60 and 80 – 100 cm) that quantitatively represent the soil spectral signature. We also used the BSSI in a comparison with bare soil surface (e.g. SYSI - Synthetic Soil Image) to predict soil attributes and χ by performing the deterministic and hybrid models. We hypothesize that the BSSI, which integrates PS and RS data, enhances soil modelling predictions at subsurface by selecting the best model approach. The BSSI demonstrated original and valuable contribution to increase the predictive model power at deeper layers, while SYSI was effective at upper layers. The PS and RS integration helped to identify the main soil patterns horizontally and vertically, which traditional soil surveys have not been capable of representing.

Semi-supervised DNN regression on airborne hyperspectral imagery for improved spatial soil properties prediction

A number of algorithms have been developed for soil organic matter (SOM) or soil heavy metal detection in airborne hyperspectral imagery with high spatial and spectral resolutions. However, to achieve improved land management, the problems of the inconsistent features and low accuracy still need to be solved. In this paper, we propose a novel regression model to estimate the concentrations of SOM, arsenic (As), and chromium (Cr) in soil. Firstly, a hyperspectral unmixing technique is utilized to extract the bare soil pixels. We then combine the absorption depth feature after continuum removal, the original absorption feature, the band ratio feature, and the first-order differential feature, to form a set of features for parameter inversion. To solve the over-fitting problem caused by the small number of samples and the weak expression problem, the semi-supervised deep neural network regression (Semi-DNNR) model is introduced. The experimental were conducted using several datasets collected by HyMap, which is an airborne hyperspectral imaging sensor in VNIR-SWIR spectral range in Yitong county, Jilin province, China. The proposed Semi-DNNR model shows a good performance in this study, with the prediction Rp2 values for SOM, As, and Cr being 0.71, 0.82, and 0.63, respectively. After the spatial distribution map of the soil components of the study area was overlaid with the stream network, which was obtained from the digital elevation model (DEM). It was found that snowmelt, the melting of frozen soil, and surface rainfall can transport SOM to low-lying areas. A similar phenomenon was also observed for As, due to SOM adsorption and dissolved organic matter (DOM) complexation. A comparison of the proposed method with both feature selection methods (competitive adaptive reweighted sampling (CARS), genetic algorithm (GA)) and regression methods (partial least squares regression (PLSR), support vector regression (SVR)) shows that the proposed feature selection method is more robust than the CARS and GA methods. The proposed Semi-DNNR model was found to be at least 18.80% higher in prediction accuracy for As than the SVR or PLSR methods, at least 25.71% higher for Cr, and at least 19.73% higher for SOM.