Prediction Of Soil Properties Research Articles

Validation of digital maps derived from spatial disaggregation of legacy soil maps

Spatial disaggregation of soil map units involves downscaling existing information to produce new information at a finer scale than that of the original source. Currently, it is becoming a powerful tool to address the spatial distribution of soil information over large areas, where legacy soil polygon maps are the only source of soil information. Because of the high expense of additional resampling, few studies have sought to validate disaggregated soil maps using independent sampling. This study implemented spatial disaggregation approach to measure the quality of soil property predictions derived from disaggregated soil maps, using stratified simple random sampling of a study area of 6848 km2 (11 strata and 135 soil profiles). In a previous study, the existing legacy soil polygon map of Brittany (France) at 1:250,000 scale was spatially disaggregated at 50 m resolution using an algorithm called Disaggregation and Harmonisation of Soil Map Units Through Resampled Classification Trees (DSMART), which uses soil-landscape expert rules of soil distribution in space. By fitting equal-area spline functions, soil properties were then estimated at six depth intervals according to GlobalSoilMap specifications. To validate disaggregated soil maps, two approaches were developed according to the soil attribute nature (continuous or categorical). For categorical soil properties (soil parent material, soil drainage class, soil type and soil depth class), the overall strict purity (the degree to which all classification criterion are respected) by the most probable STU (Soil Typological Unit) map was estimated at 34%, while the overall average purity reached 70%. The overall partial soil-type purity reached 60%, the overall partial parent material purity reached 78% and the overall partial soil drainage class as well as soil depth class purities reached 65% and 78%, respectively. Continuous soil properties (clay content, fine silt content, coarse silt content, total silt content, fine sand content, coarse sand content, coarse fragments, Cation Exchange Capacity (CEC) and pH) were validated at two soil depth intervals (5–15 and 30–60 cm) using 260 soil samples. In general, soil property predictions were unbiased except for coarse fragments and CEC in the 5–15 cm layer. Validation statistics (R2, RMSE, RRMSE and ME) were better for the 30–60 cm layer except for soil particle-size distribution. Thus, differences in prediction accuracies among strata (the validation support) denote areas where more soil data or better soil prediction models are needed to improve the disaggregation process.

Accuracy of Estimating Soil Properties with Mid‐Infrared Spectroscopy: Implications of Different Chemometric Approaches and Software Packages Related to Calibration Sample Size

Core Ideas Selection of spectral regions was included in MIR predictions of soil properties. The R software outperformed a commercial chemometric software. SVMR and model averaging performed better than PLSR and artificial neural networks. With decreasing calibration sample size, the usefulness of SVMR over PLSR decreased. MIRS studies may focus more on representativeness, sample sizes, and variabilities. Different algorithms exist in various software programs for the estimation of soil properties using mid‐infrared (MIR) spectroscopy, with recommendations varying between different studies regarding which algorithm should be used. Objectives were to compare the performance of the commercial OPUS Quant 2 software, which uses partial least squares regression (PLSR) and a selection of spectral ranges, with the R software and to study the accuracy of different algorithms as a function of the information provided in the calibration. Contents of soil organic carbon (SOC), nitrogen, and texture for surface soils of an arable field were determined, and MIR were spectra recorded. Partial least squares regression used with either software was useful (ratio of performance to interquartile distance in the validation sample [RPIQV] >1.89) for an estimation of SOC, clay, and N contents but not for sand and silt. The wavenumber region selection concept used in OPUS was also implemented in R, and it proved useful for SOC (all algorithms) and total nitrogen (artificial neural networks, support vector machine regression [SVMR]) in the validation. Support vector machine regression generally slightly outperformed the other approaches and resulted in a successful estimation of sand content. The usefulness of SVMR over PLSR generally decreased with decreasing sample size used for the calibration (thus decreasing the information provided), and PLSR partly outperformed SVMR in the validation. Overall, this study indicates that there is no general superiority of a chemometric algorithm over PLSR independent of the information provided in the calibration sample.

Evaluating sampling efforts of standard laboratory analysis and mid-infrared spectroscopy for cost effective digital soil mapping at field scale

The performance of digital soil mapping (DSM) model is highly reliant on the intensity and spatial distribution of the input soil data points. Increasing the number of soil data points (i.e. samples) improves the accuracy of the prediction, but it also raises the sampling effort, including the time, money and labor required for field and laboratory analysis. Thus, optimizing the production of DSMs requires maximizing accuracy while minimizing cost. In this study, we evaluated a range of strategies for DSM of a farm field using high spatial resolution ancillary environmental data (e.g. unmanned aerial vehicle-UAV imagery) and compared sampling efforts of soil data generated from standard laboratory analysis (SLA) and mid-infrared spectroscopy (MIRS) at equivalent costs. We produced DSMs of a number of soil properties including sand, silt, clay, pH, salinity, organic matter, and total nitrogen. We employed Conditioned Latin Hypercube Sampling (cLHS) to generate a range of sampling efforts from the full SLA (n = 62) and MIRS (n = 308) datasets and contrasted the DSM outcomes modeled using kriging with external drift (KED). We found that the DSM outputs were most effective, in terms of accuracy and cost, at 50–60% of the full sampling effort. Although MIRS predictions of soil properties introduced a sizable amount of error, DSMs produced using the MIRS dataset were more accurate as compared to the outcomes of SLA datasets at equivalent sampling efforts. The prediction accuracy for DSMs varied across the soil properties with R2 ranging from 0.82 (for sand) to 0.45 (for total nitrogen) at the optimum sampling effort. The outcomes of the study highlight that spatially optimized sampling efforts and the use of the MIRS technique substantially improve the cost efficiency and accuracy of kriging-based DSM models for predicting a range of field scale soil properties.

Evaluating low-cost portable near infrared sensors for rapid analysis of soils from South Eastern Australia

Near infrared spectroscopy has been proposed as a rapid and cost-effective method for soil analysis. Full-range visible near infrared (Vis-NIR) spectrometers working in the 350–2500 nm wavelength are commonly used for research purposes, but now miniaturized spectrometers with limited NIR ranges have become available. This study aims to compare the accuracy of near infrared reflectance (NIR) instruments with different sizes and wavelength ranges on the prediction of soil properties. Soil samples were taken from southern New South Wales and northern Victoria, comprised of Chromosols, Dermosols, Kandosols, and Sodosols. In total, the dataset consists of 392 soil samples from the top 1 m. The study compared two research-grade Vis-NIR spectrometers (350–2500 nm) and two miniaturized NIR spectrometers with limited wavelengths (NeoSpectra operating at 1250–2500 nm and NIRVascan operating at 900–1700 nm). Cubist regression tree and Partial Least Squares Regression (PLSR) were used to build calibration models. The results showed that both modelling procedures are reliable for estimating soil properties. Cubist models gave greater validation r2 values for ten soil properties investigated for all four instruments. Based on the Cubist model, promising results were found in predictions of clay, sand, total carbon, CEC, pH, exchangeable Mg and Ca (r2: 0.43–0.81). The results are also comparable to published studies for all instruments. As expected, the research-grade spectrometers provided the best prediction accuracies (r2: 0.56–0.81). Results from NeoSpectra provided a comparable accuracy with the Vis-NIR spectrometers in prediction of soil pH, CEC and exchangeable Ca and Mg (r2 > 0.63–0.78). NeoSpectra produced a slightly less accurate prediction of total carbon, sand, and clay. While NIRVascan showed the lowest accuracy, it still can be used in prediction of soil texture and total carbon with reasonable accuracy (r2 clay: 0.73; sand: 0.63; total carbon: 0.73). This study demonstrates the potential of miniaturized spectrometers with reduced wavelength ranges as cheaper instruments in soil analysis.

Combined Application of Imaging Methods for Estimating Soil Physicochemical Properties

Traditional techniques and methods for evaluation and detection of soil samples are tedious, laborious, expensive, and time-consuming. In comparison, spectroscopic techniques have successfully overcome some of these disadvantages and can supplement or replace them. As soil is a complex media, it is difficult to assign specifically spectral features for physicochemical properties of soil. This study discusses the combined application of Attenuated Total Reflection Fourier Transform Infrared (ATR-FTIR), and Scanning Electron Microscopy coupled to EDX microprobe (SEM/EDX) methods for estimating soil physicochemical properties. For this purpose, 30 topsoil samples were measured according to standard methods of soil analysis. All spectra were collected in the mid-infrared (MIR) from 4000 to 650 cm-1. Chemometric methods were used in the analysis of multivariate data using the Quant2 package in OPUS 7 software and improved the prediction of soil properties using partial least squares regression (PLSR). The results showed that specific surface area (SSA MB-titration) had the best predictions for the soil properties in ATR-FTIR study with a RPD of 2.08, RMSE of 13.50 m2 g–1, and R2 of 0.77 (very good); followed by cation exchange capacity, organic carbon, SSA (BET), Al, clay, and Ca (good); saturation, sand, silt/clay, electrical conductivity, saturated hydraulic conductivity, K, silt, Fe, C, and Si (fair); O, Mg, pH, bulk density, porosity, total pore volume, and particle density (poor). Also, we found that SEM-EDX is able to perform a rapid simultaneous multielement analysis without any special soil sample preparation. Consequently, modern techniques, such as ATR-FTIR and SEM-EDX can be used for such studies according to the information needed and the time required for the sample preparation.

Removing Moisture Effect on Soil Reflectance Properties: A Case Study of Clay Content Prediction

Visible, near-infrared and shortwave-infrared (VNIR-SWIR) spectroscopy is an efficient approach for predicting soil properties because it reduces the time and cost of analyses. However, its advantages are hampered by the presence of soil moisture, which masks the major spectral absorptions of the soil and distorts the overall spectral shape. Hence, developing a procedure that skips the drying process for soil properties assessment directly from wet soil samples could save invaluable time. The goal of this study was twofold: proposing two approaches, partial least squares (PLS) and nearest neighbor spectral correction (NNSC), for dry spectral prediction and utilizing those spectra to demonstrate the ability to predict soil clay content. For these purposes, we measured 830 samples taken from eight common soil types in Israel that were sampled at 66 different locations. The dry spectrum accuracy was measured using the spectral angle mapper (SAM) and the average sum of deviations squared (ASDS), which resulted in low prediction errors of less than 8% and 14%, respectively. Later, our hypothesis was tested using the predicted dry soil spectra to predict the clay content, which resulted in R2 of 0.69 and 0.58 in the PLS and NNSC methods, respectively. Finally, our results were compared to those obtained by external parameter orthogonalization (EPO) and direct standardization (DS). This study demonstrates the ability to evaluate the dry spectral fingerprint of a wet soil sample, which can be utilized in various pedological aspects such as soil monitoring, soil classification, and soil properties assessment.

Desenvolvimento de uma função hidropedológica para umidade volumétrica saturada de solos brasileiros

Propriedades hidráulicas do solo geralmente são de difícil aquisição e demandam bastante tempo e recursos financeiros. Nesse contexto, surgem as Funções Hidropedológicas (FHP) que são equações que por meio de modelos matemáticos predizem propriedades do solo a partir de dados de sua composição. Neste artigo, foi desenvolvida uma FHP para a estimativa da umidade volumétrica saturada tomando como base um banco de dados de 84 amostras de solos representativos do Brasil. Para a sua avaliação, o Modelo Proposto foi comparado com outras sete FHP, e ao final mostrou-se a mais eficiente para o caso estudado, já que obteve o melhor desempenho geral nos quatro parâmetros estatísticos aplicados. Verificou-se que a origem dos solos utilizados para o desenvolvimento das equações é fator influenciador na eficiência: modelos aplicados a solos com características pedológicas semelhantes aos utilizados na sua criação tendem a obter melhores resultados.

Using time-series Sentinel-1 data for soil prediction on invaded coastal wetlands.

Coastal soils are particularly sensitive to nonnative species invasion. In this context, spatially explicit soil information is essential for improving the knowledge of the role of soil in changing environments, supporting coastal sustainable management. Synthetic-aperture radar (SAR) data provides an attractive opportunity to monitor soil because the acquisition of images is independent of weather and daylight. However, SAR has not been commonly used for soil prediction. In this study, we firstly investigated the temporal variation of vegetation canopy and the soil-vegetation relationship using Sentinel-1 data in an invaded coastal wetland. And then we built 3D models to predict soil properties at multiple depths. A total of 16 Sentinel-1 images were acquired in a growing season. A series of soil physicochemical properties were examined including soil bulk density, texture, organic/inorganic carbon, pH, salinity, total nitrogen, and C/N ratio, relating to three depth layers in the top 1-m depth. Our results showed that time-series Sentinel-1 data can capture temporal characteristics of vegetation, and VH/VV was more sensitive to the vegetation growth than VH and VV. The soil-vegetation relationship captured by time-series SAR data was beneficial to predict soil properties, especially for soil chemical properties. The models provided permissible prediction accuracy, with an average RPD of 0.99. We concluded that the prior understanding of the temporal variation of SAR data is essential for developing practical soil prediction strategy. Our results highlight that SAR has the potential to predict a diverse set of soil properties in coastal wetlands with dense vegetation cover.

Convolutional neural network for simultaneous prediction of several soil properties using visible/near-infrared, mid-infrared, and their combined spectra

No single instrument can characterize all soil properties because soil is a complex material. With the advancement of technology, laboratories have become equipped with various spectrometers. By fusing output from different spectrometers, better prediction outcomes are expected than using any single spectrometer alone. In this study, model performance from a single spectrometer (visible-near-infrared spectroscopy, vis-NIR or mid-infrared spectroscopy, MIR) was compared to the combined spectrometers (vis-NIR and MIR). We selected a total of 14,594 samples from the Kellogg Soil Survey Laboratory (KSSL) database that had both vis-NIR and MIR spectra along with measurements of sand, clay, total C (TC) content, organic C (OC) content, cation exchange capacity (CEC), and pH. The dataset was randomly split into 75% training (n = 10,946) and the remaining (n = 3,648) as a test set. Prediction models were constructed with partial least squares regression (PLSR) and Cubist tree model. Additionally, we explored the use of a deep learning model, the convolutional neural network (CNN). We investigated various ways to feed spectral data to the CNN, either as one-dimensional (1D) data (as a spectrum) or as two-dimensional (2D) data (as a spectrogram). Compared to the PLSR model, we found that the CNN model provides an average improvement prediction of 33–42% using vis-NIR and 30–43% using MIR spectral data input. The relative accuracy improvement of CNN, when compared to the Cubist regression tree model, ranged between 22 and 36% with vis-NIR and 16–27% with MIR spectral data input. Various methods to fuse the vis-NIR and MIR spectral data were explored. We compared the performance of spectral concatenation (for PLSR and Cubist model), two-channel input method, and outer product analysis (OPA) method (for CNN model). We found that the performance of two-channel 1D CNN model was the best (R2 = 0.95–0.98) followed closely by the OPA with CNN (R2 = 0.93–0.98), Cubist model with spectral concatenation (R2 = 0.91–0.97), two-channel 2D CNN model (R2 = 0.90–0.95) and PLSR with spectral concatenation (R2 = 0.87–0.95). Chemometric analysis of spectroscopy data relied on spectral pre-processing methods: such as spectral trimming, baseline correction, smoothing, and normalization before being fed into the model. CNN achieved higher performance than the PLSR and Cubist model without utilizing the pre-processed spectral data. We also found that the predictions using the CNN model retained similar correlations to the actual values in comparison to other models. By doing sensitivity analysis, we identified the important spectral wavelengths variables used by the CNN model to predict various soil properties. CNN is an effective model for modelling soil properties from a large spectral library.

Determining the spatial distribution of soil properties using the environmental covariates and multivariate statistical analysis: a case study in semi-arid regions of Iran

Natural soil-forming factors such as landforms, parent materials or biota lead to high variability in soil properties. However, there is not enough research quantifying which environmental factor(s) can be the most relevant to predicting soil properties at the catchment scale in semi-arid areas. Thus, this research aims to investigate the ability of multivariate statistical analyses to distinguish which soil properties follow a clear spatial pattern conditioned by specific environmental characteristics in a semi-arid region of Iran. To achieve this goal, we digitized parent materials and landforms by recent orthophotography. Also, we extracted ten topographical attributes and five remote sensing variables from a digital elevation model (DEM) and the Landsat Enhanced Thematic Mapper (ETM), respectively. These factors were contrasted for 334 soil samples (depth of 0–30 cm). Cluster analysis and soil maps reveal that Cluster 1 comprises of limestones, massive limestones and mixed deposits of conglomerates with low soil organic carbon (SOC) and clay contents, and Cluster 2 is composed of soils that originated from quaternary and early quaternary parent materials such as terraces, alluvial fans, lake deposits, and marls or conglomerates that register the highest SOC content and the lowest sand and silt contents. Further, it is confirmed that soils with the highest SOC and clay contents are located in wetlands, lagoons, alluvial fans and piedmonts, while soils with the lowest SOC and clay contents are located in dissected alluvial fans, eroded hills, rock outcrops and steep hills. The results of principal component analysis using the remote sensing data and topographical attributes identify five main components, which explain 73.3% of the total variability of soil properties. Environmental factors such as hillslope morphology and all of the remote sensing variables can largely explain SOC variability, but no significant correlation is found for soil texture and calcium carbonate equivalent contents. Therefore, we conclude that SOC can be considered as the best-predicted soil property in semi-arid regions.

Temporal stability of soil apparent electrical conductivity (ECa) in managed podzols

The spatial variability in soil physical and hydraulic properties for a managed podzol was assessed using soil apparent electrical conductivity (ECa). Two EMI sensors, the multi-coil (MC) and multi-frequency (MF), were adopted for measurement of ECa on a silage- corn experimental plot in western Newfoundland, Canada. Results demonstrated a significant relationship between the ECa mean relative differences (MRD) and the soil moisture content MRD (R2 = 0.33 to 0.70) for both MC and MF sensors. The difference in depth sensitivity between MC and MF sensors accounted for the variation (0.015 to 0.09) in ECa standard deviation of the relative differences. A significant linear relationship was found between the ECa MRD and sand (R2 = 0.35 and 0.53) or silt (R2 = 0.43), but not with clay (R2 = 0.06 and 0.16). The spatial variability of the ECa-based predictions (CV = 3.26 to 27.61) of soil properties was lower than the measured values (CV = 5.56 to 41.77). These results inferred that the temporal stability of ECa might be a suitable proxy to understand the spatial variability of soil physical and hydraulic properties in agricultural podzols.

An evolutionary fuzzy rule-based system applied to the prediction of soil organic carbon from soil spectral libraries

The recent efforts for obtaining vast soil spectral libraries covering a significant part of the spatial and compositional variability of soils have underscored the need for accurate and interpretable models. Herein, the application of an evolutionary Fuzzy Rule-based System (FRBS) named DECO3RUM (Differential Evolution based Cooperative and Competing learning of Compact Rule-based Models) for the prediction of soil properties from visible, near-infrared, and shortwave-infrared (VNIR–SWIR) laboratory spectral data obtained from the LUCAS topsoil database is investigated. FRBSs model the input–output relation with fuzzy logic statements, offering an enhanced interpretability degree for the experts over classical rule-based systems and other black box models. The proposed algorithm was statistically compared with other state of the art approaches and was found to outperform other global models, while being statistically similar with local approaches that offer lower interpretation capabilities.

Multi-source data integration for soil mapping using deep learning

Abstract. With the advances of new proximal soil sensing technologies, soil properties can be inferred by a variety of sensors, each having its distinct level of accuracy. This measurement error affects subsequent modelling and therefore must be integrated when calibrating a spatial prediction model. This paper introduces a deep learning model for contextual digital soil mapping (DSM) using uncertain measurements of the soil property. The deep learning model, called the convolutional neural network (CNN), has the advantage that it uses as input a local representation of environmental covariates to leverage the spatial information contained in the vicinity of a location. Spatial non-linear relationships between measured soil properties and neighbouring covariate pixel values are found by optimizing an objective function, which can be weighted with respect to a measurement error of soil observations. In addition, a single model can be trained to predict a soil property at different soil depths. This method is tested in mapping top- and subsoil organic carbon using laboratory-analysed and spectroscopically inferred measurements. Results show that the CNN significantly increased prediction accuracy as indicated by the coefficient of determination and concordance correlation coefficient, when compared to a conventional DSM technique. Deeper soil layer prediction error decreased, while preserving the interrelation between soil property and depths. The tests conducted suggest that the CNN benefits from using local contextual information up to 260 to 360 m. We conclude that the CNN is a flexible, effective and promising model to predict soil properties at multiple depths while accounting for contextual covariate information and measurement error.

Mapping parent material as part of a nested approach to soil mapping in the Arkansas River Valley

Soil mappers have traditionally relied on tacit knowledge and qualitative assessment of soil-landscape relationships to obtain the physiographic context necessary to predict soil distribution and spatial patterns. This assessment implicitly utilizes a nested hierarchical approach based on differences in the phenomenon scale of soil forming factors where climate, landscape, parent material, and topography are examined in sequence to create a model of soil-landscape relationships. Our objective was to predict parent material distribution using expert knowledge paired with quantitative digital terrain attributes as part of a nested approach to digital soil mapping. The study took place at an 890-hectare research farm in Logan County, Arkansas, which is part of the Arkansas River Valley. Two major groups of parent material are identifiable in the Arkansas River Valley: residual sandstone and shale on erosional uplands, and silty/clayey pedisediment in depositional areas. A 5-m digital elevation model was used to derive thirteen terrain attributes for the study site. Three of the terrain attributes, namely topographic position index, multi-resolution valley bottom flatness, and vertical distance to channel network, were utilized as part of a rule-based approach to model parent material distribution based on preliminary reconnaissance and expert knowledge of the area. The model was validated by sampling 20 locations using a conditioned Latin hypercube sampling (cLHS) design to evaluate the prediction accuracy. Seventy-five percent of cLHS samples were accurately predicted to be residuum or pedisediment. The resulting map also had 90% agreement with the National Cooperative Soil Survey map; however, the digital map was able to provide more spatially explicit information especially on inclusions. Incorporating parent material distribution as part of a nested hierarchical approach to digital soil mapping aids in constraining and predicting soil properties, enables a more straightforward examination of physiographic context and can ultimately lead to more accurate digital soil maps for land management.

A genetic algorithm‐based stacking algorithm for predicting soil organic matter from vis–NIR spectral data

It has been demonstrated that diffuse reflectance spectroscopy in the visible and near‐infrared (vis–NIR) can be exploited to predict chemical and physical soil properties. Immense soil spectral libraries (SSL) are being developed; therefore, more elaborate tools that capitalize on contemporary knowledge and techniques need to be established to provide accurate predictions. In this paper, we propose a novel genetic algorithm‐based stacking model that makes synergetic use of multiple models developed from different preprocessed spectral sources (termed L1 models). This is a form of ensemble learning where multiple hypotheses are combined to create a more robust and more accurate ensemble hypothesis. The genetic algorithm automatically defines the configuration of the stacked model, by selecting the best cooperating subset of the initial models. Our methodology was tested on the newly developed GEO‐CRADLE SSL to predict soil organic matter (SOM). Results showed that the accuracy of prediction of the proposed method ( R2 = 0.76, and ratio of performance to interquartile range [RPIQ] = 2.22) was better than the one attained by the best L1 model ( R2 = 0.65, RPIQ = 1.93). This approach can thus be effectively utilized to enhance the predictions of soil properties in small and large soil spectral libraries alike.Highlights A novel model stacking algorithm is proposed, combining spectral models from different sources and algorithms. Use of a genetic algorithm is examined, accounting for the large number of possible model permutations. The methodology was applied to the GEO‐CRADLE vis–NIR soil spectral library. Results indicate that model stacking creates more accurate and robust models.

Combining Environmental Factors and Lab VNIR Spectral Data to Predict SOM by Geospatial Techniques

Soil organic matter (SOM) is an important parameter related to soil nutrient and miscellaneous ecosystem services. This paper attempts to improve the performance of traditional partial least square regression (PLSR) model by considering the spatial autocorrelation and soil forming factors. Surface soil samples (n = 180) were collected from Honghu City located in the middle of Jianghan Plain, China. The visible and near infrared (VNIR) spectra and six environmental factors (elevation, land use types, roughness, relief amplitude, enhanced vegetation index, and land surface water index) were used as the auxiliary variables to construct the multiple linear regression (MLR), PLSR and geographically weighted regression (GWR) models. Results showed that: 1) the VNIR spectra can increase about 39.62% prediction accuracy than the environmental factors in predicting SOM; 2) the comprehensive variables of VNIR spectra and the environmental factors can improve about 5.78% and 44.90% relative to soil spectral models and soil environmental models, respectively; 3) the spatial model (GWR) can improve about 3.28% accuracy than MLR and PLSR. Our results suggest that the combination of spectral reflectance and the environmental variables can be used as the suitable auxiliary variables in predicting SOM, and GWR is a promising model for predicting soil properties.

Predicting soil types and soil properties with limited data in the Uasin Gishu Plateau, Kenya

Digital soil mapping (DSM) approaches can be used to create new soil maps or enhance existing maps, particularly in areas where only general soil maps are available. In this study, we utilized a knowledge-based inference soil mapping approach to develop a first-generation digital soil map for part of the Uasin Gishu Plateau in western Kenya. We calculated the following environmental covariates from the Shuttle Radar Topographic Mission (SRTM) 30 m digital elevation model (DEM): slope gradient, multi-resolution valley bottom flatness index (MrVBF), multi-resolution ridgetop flatness index (MrRTF), topographic position index (TPI), elevation, and profile curvature. These covariates were then used along with existing soil information and expert knowledge from soil scientists familiar with the area to produce new raster-based maps of soil types, effective soil depth, soil moisture storage capacity and soil drainage class. The soil type maps predicted using clustering analysis and fuzzy logic methods showed good agreement with field observations based on the overall accuracy values. The fuzzy logic map performed slightly better (kappa coefficient (k) = 0.68; overall accuracy = 0.76) than the map based on clustering analysis (k = 0.59; overall accuracy = 0.68). The accuracy for the effective soil depth fuzzy logic map was higher (R2 = 0.56; RMSE = 11; ME = 1.1) compared to the existing soil map (R2 = 0.34; RMSE = 27; ME = 8). Seven major soil types occur in the study area: Ferralsols, Nitisols, Gleysols, Luvisols, Acrisols, Cambisols and Regosols, according to the World Reference Base soil classification system. This study generated detailed and improved predictions of soil types and properties at 30 m grid resolution. These maps should be more useful for soil, crop and land use management decisions than existing maps.

Towards “New Soil Geography”: Challenges and Solutions. A Review

This paper provides a review of the current state of soil geography and budding directions for the development of pedogeographic research. We mention some new ideas in the frames of structural approach rooting in the classical concept of soil cover pattern and based on new concepts, such as pedodiversity assessment, graph theory, and geostatistical analysis of soil spatial variation. We note the significance of digital soil mapping in the development of the theory and practice of pedogeography and stress that digital soil mapping is a method that cannot replace soil geography as a scientific discipline. There is a need for deeper integration of mathematical methods in traditional soil geography. We stress that pedogeographical models are required for predicting soil properties and regimes even in digital agriculture. We discuss the necessity for adequate reflection of polygenetic soils in the soil mantle, and recommend using both indirect paleogeographic information and current remote and proximate sensing data. We also note the difficulties in predicting the spatial distribution of anthropogenically transformed soils using state factor theory; we discuss the possibilities of broader use of historical and economical geography data. In conclusion, we suggest developing “new soil geography” not only through integration of mathematical methods but also through closer integration with allied sciences.

Accurate and Precise Prediction of Soil Properties from a Large Mid-Infrared Spectral Library

Diffuse reflectance spectroscopy (DRS) is emerging as a rapid and cost-effective alternative to routine laboratory analysis for many soil properties. However, it has primarily been applied in project-specific contexts. Here, we provide an assessment of DRS spectroscopy at the scale of the continental United States by utilizing the large (n > 50,000) USDA National Soil Survey Center mid-infrared spectral library and associated soil characterization database. We tested and optimized several advanced statistical approaches for providing routine predictions of numerous soil properties relevant to studying carbon cycling. On independent validation sets, the machine learning algorithms Cubist and memory-based learner (MBL) both outperformed random forest (RF) and partial least squares regressions (PLSR) and produced excellent overall models with a mean R2 of 0.92 (mean ratio of performance to deviation = 6.5) across all 10 soil properties. We found that the use of root-mean-square error (RMSE) was misleading for understanding the actual uncertainty about any particular prediction; therefore, we developed routines to assess the prediction uncertainty for all models except Cubist. The MBL models produced much more precise predictions compared with global PLSR and RF. Finally, we present several techniques that can be used to flag predictions of new samples that may not be reliable because their spectra fall outside of the calibration set.

Simultaneous measurement of multiple soil properties through proximal sensor data fusion: A case study

In this research, proximal soil sensor data fusion was defined as a multifaceted process which integrates geospatially correlated data, or information, from multiple proximal soil sensors to accurately characterize the spatial complexity of soils. This has capability of providing improved understanding of soil heterogeneity for potential applications associated with crop production and natural resource management. To assess the potential of data fusion for the purpose of improving thematic soil mapping over the single sensor approach, data from multiple proximal soil sensors were combined to develop and validate predictive relationships with laboratory-measured soil physical and chemical properties. The work was conducted in an agricultural field with both mineral and organic soils. The integrated data included: topography records obtained using a real-time kinetic (RTK) global navigation satellite system (GNSS) receiver, apparent soil electrical conductivity (ECa) obtained using an electromagnetic induction sensor, and content of several naturally occurring radioisotopes detected using a mobile gamma-ray spectrometer. In addition, the soil profile data were collected using a commercial ruggedized multi-sensor platform carrying a visible and near-infrared (vis-NIR) optical sensor and a galvanic contact soil ECa sensor. The measurements were carried out at predefined field locations covering the entire study area identified from sensor measured a priori information on field elevation, ECa and gamma-ray count. The information was used to predict: soil organic matter (SOM), pH, lime buffer capacity (LBC), as well as concentration of phosphorus (P), potassium (K), calcium (Ca), magnesium (Mg), and aluminum (Al). Partial least squares regressions (PLSRs) were used to predict soil properties from individual sensors and different sensor combinations (sensor data fusion). By integrating the data from all of the proximal soil sensors, SOM, pH, LBC, Ca, Mg, and Al were predicted simultaneously with R2 > 0.5 (RPD > 1.50). Improved predictions were observed for most soil properties based on sensor data fusion than those based on individual sensors. After choosing the optimal sensor combination for each soil property, the predictive capability was compared using different data mining algorithms, including support vector machines (SVM), random forest (RF), multivariate adaptive regression splines (MARS), and regression trees (CART). Improved predictions for SOM, Ca, Mg, and Al were observed using SVM over PLSR. The predictive capability was followed by RF and MARS, with CART. Predictions of pH and LBC were only feasible using MARS and PLSR, respectively. In this field, it was not possible to predict extractable P and K using all tested sensor combinations or algorithms. With large variability in SOM, the field presents a special situation and thus, the result could be specific to the study site. Further research includes an extended number of experimental sites covering different geographic areas around Eastern Canada.