Prediction Of Soil Properties Research Articles

Multi-Sensor Soil Probe and Machine Learning Modeling for Predicting Soil Properties.

We present a data-driven, in situ proximal multi-sensor digital soil mapping approach to develop digital twins for multiple agricultural fields. A novel Digital Soil CoreTM (DSC) Probe was engineered that contains seven sensors, each of a distinct modality, including sleeve friction, tip force, dielectric permittivity, electrical resistivity, soil imagery, acoustics, and visible and near-infrared spectroscopy. The DSC System integrates the DSC Probe, DSC software (v2023.10), and deployment equipment components to sense soil characteristics at a high vertical spatial resolution (mm scale) along in situ soil profiles up to a depth of 120 cm in about 60 s. The DSC Probe in situ proximal data are harmonized into a data cube providing vertical high-density knowledge associated with physical-chemical-biological soil conditions. In contrast, conventional ex situ soil samples derived from soil cores, soil pits, or surface samples analyzed using laboratory and other methods are bound by a substantially coarser spatial resolution and multiple compounding errors. Our objective was to investigate the effects of the mismatched scale between high-resolution in situ proximal sensor data and coarser-resolution ex situ soil laboratory measurements to develop soil prediction models. Our study was conducted in central California soil in almond orchards. We collected DSC sensor data and spatially co-located soil cores that were sliced into narrow layers for laboratory-based soil measurements. Partial Least Squares Regression (PLSR) cross-validation was used to compare the results of testing four data integration methods. Method A reduced the high-resolution sensor data to discrete values paired with layer-based soil laboratory measurements. Method B used stochastic distributions of sensor data paired with layer-based soil laboratory measurements. Method C allocated the same soil analytical data to each one of the high-resolution multi-sensor data within a soil layer. Method D linked the high-density multi-sensor soil data directly to crop responses (crop performance and behavior metrics), bypassing costly laboratory soil analysis. Overall, the soil models derived from Method C outperformed Methods A and B. Soil predictions derived using Method D were the most cost-effective for directly assessing soil-crop relationships, making this method well suited for industrial-scale precision agriculture applications.

Soil erodibility mapping using remote sensing and in situ soil data with random forest model in a mountainous catchment of Indian Himalayas.

Land degradation is accelerating in the Himalayan ecosystem, resulting in the loss of soil nutrients due to severe erosion. Soil erosion presents a significant environmental challenge, resulting in both on-site and off-site consequences, such as reduced soil productivity and siltation in reservoirs. Soil erodibility (K factor), an inherent soil property, determines the susceptibility of soils to erosion. Sampling across hilly and mountainous terrain pose challenges due to its complex landscape. Despite these challenges, it is essential to study K factor variations in different land use/land cover types to comprehend the threat of erosion. Digital soil mapping offers an opportunity to overcome this limitation by providing spatial predictions of soil properties. The objective of our study is to map the spatial distribution of soil erodibility using the Random Forest (RF) model, a machine learning method based on sampled in situ soil data and environmental covariates. We collected 556 surface soil samples from the mountainous catchment (Tehri dam catchment) using the stratified random sampling approach. The model performed satisfactorily in both training (r2 = 0.91; RMSE = 0.00185) and testing (r2 = 0.45; RMSE = 0.00318) phases. Subsequently, we generated a digital map with a resolution of 12.5m to depict the distribution of the K factor. Our analysis revealed that key environmental variables influencing the prediction of the K factor included geology, mean NDVI, and climatic factors. The average K factor value was estimated at 0.0304 and ranging from 0.0251 to 0.0400 t ha h ha-1MJ-1mm-1. A higher K factor was observed in the barren land (0.0344) primarily located in the higher and trans-Himalayan region of seasonally snow-covered areas. These areas typically feature young soils with weak soil formation and unstable soil aggregates. Subsequently cropland/cultivated soils (0.0307) exhibited higher K factor values due to the breakdown of soil aggregates by ploughing activities and exposing carbon to decomposition. The average K factor value of evergreen (0.0294) and deciduous (0.0295) forests were the lowest compared to other land use/land cover types indicating the role of forests in resisting soil erosion. By assessing and predicting soil erodibility, land planners and farmers can implement erosion control measures to protect soil health, prevent sedimentation in water bodies, and sustain agricultural productivity in the Himalayas.

Using convolutional neural network combined with multi-scale channel attention module to predict soil properties from visible and near-infrared spectral data

With the increasing demand for precision agriculture and sustainable land management, it is crucial to quickly and accurately predict soil properties. However, there are still challenges in predicting large-scale soil properties. This study proposes a new convolutional neural network architecture (MCA-CNN) combined with visible and near-infrared spectroscopy for predicting soil properties. The architecture introduces a multi-scale channel attention mechanism, aiming to more effectively capture complex features in the spectrum. The performance of the MCA-CNN model was tested on a large-scale dataset and compared with partial least squares (PLS), extreme gradient boosting tree (XGBoost), and conventional convolutional neural network (CNN) models. The predictive performance of different models in organic carbon (OC), nitrogen (N), pH, Clay, Salt, and Sand was analyzed. The results showed that the MCA-CNN model had the best predictive performance on all soil properties, achieving the lowest root mean square error (RMSE) and the highest coefficient of determination (R2), the R2 of OC, N, pH, Clay, Silt, and Sand are 0.95, 0.93, 0.95, 0.80, 0.58, 0.64, respectively. The RMSE of OC, N, pH, Clay, Silt, and Sand are 16.60 g kg−1, 0.96 g kg−1, 0.31, 4.89 %, 8.27 %, and 11.61 %, respectively. Further research has found that soil types (mineral soil and organic soil) affect the predictive performance of OC. It is necessary to distinguish between mineral soil and organic soil when predicting soil OC content. In addition, the effectiveness of the model transfer strategy was explored for situations with small soil sample sizes. By fine-tuning the pre-trained model, the performance of the model can be significantly improved, which provides a feasible solution for predicting soil properties in situations of data scarcity. In summary, this study not only confirms the potential application of deep learning in the field of soil science, but also provides new technological avenues for future soil management and agricultural practices.

Self-supervised learning of Vision Transformers for digital soil mapping using visual data

In arid environments, prospecting cultivable land is challenging due to harsh climatic conditions and vast, hard-to-access areas. However, the soil is often bare, with little vegetation cover, making it easy to observe from above. Hence, remote sensing can drastically reduce costs to explore these areas. For the past few years, deep learning has extended remote sensing analysis, first with Convolutional Neural Networks (CNNs), then with Vision Transformers (ViTs). The main drawback of deep learning methods is their reliance on large calibration datasets, as data collection is a cumbersome and costly task, particularly in drylands. However, recent studies demonstrate that ViTs can be trained in a self-supervised manner to take advantage of large amounts of unlabelled data to pre-train models. These backbone models can then be finetuned to learn a supervised regression model with few labelled data.In our study, we trained ViTs in a self-supervised way with a 9500 km2 satellite image of dry-lands in Saudi Arabia with a spatial resolution of 1.5 m per pixel. The resulting models were used to extract features describing the bare soil and predict soil attributes (pH H2O, pH KCl, Si composition). Using only RGB data, we can accurately predict these soil properties and achieve, for instance, an RMSE of 0.40 ± 0.03 when predicting alkaline soil pH. We also assess the effectiveness of adding additional covariates, such as elevation. The pretrained models can as well be used as visual features extractors. These features can be used to automatically generate a clustered map of an area or as input of random forests models, providing a versatile way to generate maps with limited labelled data and input variables.

Interpreting and evaluating digital soil mapping prediction uncertainty: A case study using texture from SoilGrids

Soil information is critical for a wide range of land resource and environmental decisions. These decisions will be compromised when the soil information quality is unsatisfactory. Thus, users of soil information need to understand and consider the uncertainty of the available soil information and be able to judge whether it is fit for purpose. The uncertainty information provided with the SoilGrids2.0 product was examined in a case study. We hypothesised that the soil property predictions for the Netherlands (NL) might be less uncertain than those of New Zealand (NZ) because there were more relevant training data for NL than for NZ. The study objectives were to: 1)understand whether the provided uncertainty information is correct for both countries; 2)explore spatial patterns and relationships in the prediction error and uncertainty information using quantitative tools and new graphical analyses; 3)analyse whether these patterns and relations can be explained; and 4)explore how the uncertainty information and insights derived from graphical analyses might assist an end user to determine whether a map is suitable for their purpose. The study focused on soil texture.Independent datasets showed that the SoilGrids 2.0 uncertainty information was too optimistic for sand and too pessimistic for clay for both countries. The graphical analyses confirmed the initial assumption that NL predictions were more accurate than those for NZ, but they also indicated that some locations in NL have high uncertainty. The graphical analyses allowed only a limited identification of the four sources of uncertainty in digital soil maps, but were quite insightful in helping us to better understand the reliability of the information. A set of recommendations was developed for both producers and consumers of digital soil mapping (DSM) products. This includes the provision of a summary map of accuracy classes. We suggest that more research and educational effort is needed to ensure that digital soil maps are used appropriately.

Towards data-driven tropical forest restoration: Uncovering spatial variation, interactions and historical management effects on nutrients along soil depth gradients

Data scarcity hinders global conservation initiatives, and there is a pressing demand for spatially detailed soil and species data to restore human-altered tropical forests. We, therefore, aimed to generate foundational soil environment and habitat suitability data and high-resolution soil maps to aid restoration efforts in a critical ecosystem of the threatened Indo-Burma Biodiversity Hotspot region, i.e., Tarap Hill Reserve (THR) in Bangladesh. Using multiple soil depths and vegetation data, we answered three major questions. (QI) How do spatial distribution and the relationships between soil physicochemical properties (i.e., pH, sand, silt, and clay percentages, organic carbon, and nutrients – N, P, K, Ca, Mg, Fe, and Zn) vary from surface to deeper soils (top 1 m)? (QII) How do different forest management interventions, i.e., old-growth forests (OGF), mixed plantations (MXP), and mono-specific plantations (MOP), influence soil properties, nutrients, and carbon in different soil depths? (QIII) Which spatial interpolation methods are best suited for making more accurate soil property predictions at different depths? Our analyses reveal decreasing availability of critical nutrients like N, P, Mg, and Fe from surface to subsurface soils, while pH, soil organic carbon, and clay content increased with depth. Several soil properties showed significant interactions, although the strength of the interactions changed from surface to deeper soils. Besides, forest management interventions significantly influenced soil functionality by having higher nutrient availability and soil organic carbon in OGF than MXP and MOP. Predictive performances of the deterministic and geostatistical interpolation methods varied for different soil properties in different soil depths, and soil maps revealed substantial heterogeneity in the distribution of soil properties across space and along depths. This study represents a pioneering step in data-driven tropical forest restoration, and our novel findings and high-resolution soil maps could guide future studies focusing on species habitat preferences, restoration ecology, and spatial conservation planning in the Indo-Burma Biodiversity Hotspot region and elsewhere in the tropics.

Comparison of two gamma-ray datasets measured with different methods and assessment of their performance to predict soil properties

ABSTRACT Terrain mobility is directly dependent on soil properties, i.e. stoniness, soil particle size and peat depth. Gamma-ray spectrometry has been widely used in soil mapping as gamma-rays can penetrate up to 50 cm of soil. The method has proven to be effective in gathering information on various soil properties, such as clay and silt content, organic carbon content, soil pH, and soil stoniness. The intensity of gamma radiation decreases with increasing soil moisture content, which is important when assessing the load-bearing capacity of forest terrain, which affects the quality and planning of mechanized wood harvesting. In this study, we compared two gamma-ray datasets that were acquired from the same plots but measured with different methods: airborne gamma-ray spectrometry and a handheld gamma-ray spectrometer. Regression analyses were used to examine whether the gamma-ray datasets have significant similarities. Linear discriminant analysis was then used to study the performance of the two gamma-ray datasets when predicting different soil properties (stoniness index classes (SIC), soil depth, and peat depth). Our correlation analysis produced R2 values of 0.18–0.29 and root mean square error (RMSE%) values of 0.48–0.55. Prediction of SIC was 67.1% accurate with ground gamma and 45.9% accurate with air gamma. For soil depth predictions, the accuracies were 64.7–70.0% for ground gamma and 61.7–63.5% for air gamma. Our findings showed that ground gamma-ray spectrometry is better at predicting soil properties than air gamma-ray spectrometry.

Feasibility of Vis-NIR spectroscopy approach to predict soil biological attributes in arid land soils.

Visible and near-infrared (Vis-NIR) reflectance spectroscopy has recently emerged as an efficient and cost-effective tool for monitoring soil parameters and provides an extensive array of measurements swiftly. This study sought to predict fundamental biological attributes of calcareous soils using spectral reflectance data in the Vis-NIR range through the application of partial least square regression (PLSR) and stepwise multiple linear regression (SMLR) techniques. The objective was to derive spectrotransfer functions (STFs) to predict selected soil biological attributes. A total of 97 composite samples were collected from three distinct agricultural land uses, i.e., sugarcane, wheat, and date palm, in the Khuzestan Province, Iran. The samples were analyzed using both standard laboratory analysis and proximal sensing approach within the Vis-NIR range (400-2500 nm). Biological status was evaluated by determining soil enzyme activities linked to nutrient cycling including acid phosphatase (ACP), alkaline phosphatase (ALP), dehydrogenase (DEH), soil microbial respiration (SMR), microbial biomass phosphorus (Pmic), and microbial biomass carbon (Cmic). The results indicated that the developed PLSR models exhibited superior predictive performance in most biological parameters compared to the STFs, although the differences were not significant. Specifically, the STFs acceptably accurately predicted ACP, ALP, DEH, SMR, Pmic, and Cmic with R2val (val = validation dataset) values of 0.68, 0.67, 0.65, 0.65, 0.76, and 0.72, respectively. These findings confirm the potential of Vis-NIR spectroscopy and the effectiveness of the associated STFs as a rapid and reliable technique for assessing biological soil quality. Overall, in the context of predicting soil properties using spectroscopy-based approaches, emphasis must be placed on developing straightforward, easily deployable, and pragmatic STFs.

Estimation of top soil properties by Sentinel-2 imaging

ABSTRACT This study evaluated the feasibility of using free multispectral remote sensing data from Sentinel-2A satellites to predict soil properties in Northern Karnataka, India. Sentinel-2A images were downloaded for selected sites, covering Vertisol, Ultisol, and Alfisol soils. Multiple linear regression (MLR) models incorporated four Sentinel-2 bands and six spectral indices (NDVI, GNDVI, SAVI, TVI, EVI, and BI) as independent variables, with soil properties as dependent variables. Surface samples (0–15 cm depth) were collected from March to May 2022. The analysis showed significant correlations between individual bands and soil properties, with variations in Organic Carbon (OC) compared to sand, silt, clay, and pH. Sand positively correlated with all spectral indices, while silt, clay, and pH were negatively correlated. The red and Near-Infrared (NIR) bands showed a non-significant relationship with OC. No significant correlation was found between EVI and the soil properties. Strong regression coefficients were observed between Sentinel-2 predictions and laboratory measurements: sand (r² = 0.63), silt (r² = 0.73), clay (r² = 0.59), and pH (r² = 0.59). These results demonstrate the potential of Sentinel-2 data for predicting soil properties, offering a valuable tool for managing unsampled agricultural fields.

A Novel Transformer-CNN Approach for Predicting Soil Properties from LUCAS Vis-NIR Spectral Data

Soil, a non-renewable resource, requires continuous monitoring to prevent degradation and support sustainable agriculture. Visible-near-infrared (Vis-NIR) spectroscopy is a rapid and cost-effective method for predicting soil properties. While traditional machine learning methods are commonly used for modeling Vis-NIR spectral data, large datasets may benefit more from advanced deep learning techniques. In this study, based on the large soil spectral library LUCAS, we aimed to enhance regression model performance in soil property estimation by combining Transformer and convolutional neural network (CNN) techniques to predict 11 soil properties (clay, silt, pH in CaCl2, pH in H2O, CEC, OC, CaCO3, N, P, and K). The Transformer-CNN model accurately predicted most soil properties, outperforming other methods (partial least squares regression (PLSR), random forest regression (RFR), support vector machine regression (SVR), Long Short-Term Memory (LSTM), ResNet18) with a 10–24 percentage point improvement in the coefficient of determination (R2). The Transformer-CNN model excelled in predicting pH in CaCl2, pH in H2O, OC, CaCO3, and N (R2 = 0.94–0.96, RPD > 3) and performed well for clay, sand, CEC, P, and K (R2 = 0.77–0.85, 2 < RPD < 3). This study demonstrates the potential of Transformer-CNN in enhancing soil property prediction, although future work should aim to optimize computational efficiency and explore a wider range of applications to ensure its utility in different agricultural settings.

An improved digital soil mapping approach to predict total N by combining machine learning algorithms and open environmental data

Digital Soil Mapping (DSM) is fundamental for soil monitoring, as it is limited and strategic for human activities. The availability of high temporal and spatial resolution data and robust algorithms is essential to map and predict soil properties and characteristics with adequate accuracy, especially at a time when the scientific community, legislators and land managers are increasingly interested in the protection and rational management of soil.Proximity and remote sensing, efficient data sampling and open public environmental data allow the use of innovative tools to create spatial databases and digital soil maps with high spatial and temporal accuracy. Applying machine learning (ML) to soil data prediction can improve the accuracy of maps, especially at scales where geostatistics may be inefficient. The aim of this research was to map the nitrogen (N) levels in the soils of the Nurra sub-region (north-western Sardinia, Italy), testing the performance of the Ranger, Random Forest Regression (RFR) and Support Vector Regression (SVR) models, using only open source and open access data. According to the literature, the models include soil chemical-physical characteristics, environmental and topographic parameters as independent variables. Our results showed that predictive models are reliable tools for mapping N in soils, with an accuracy in line with the literature. The average accuracy of the models is high (R2 = 0.76) and the highest accuracy in predicting N content in surface horizons was obtained with RFR (R2 = 0.79; RMSE = 0.32; MAE = 0.18). Among the predictors, SOM has the highest importance. Our results show that predictive models are reliable tools in mapping N in soils, with an accuracy in line with the literature. The results obtained could encourage the integration of this type of approach in the policy and decision-making process carried out at regional scale for land management.

PS-MTL-LUCAS: A partially shared multi-task learning model for simultaneously predicting multiple soil properties

Soil acts as a foundation for human survival and social development and soil quality has a great effect on the growth of agricultural products. Visible/near-infrared spectroscopy has been acknowledged as a rapid and non-destructive method for predicting soil properties, and multi-task learning is a preferable approach to analyze the correlation between the spectroscopy data and soil properties. However, current multi-task learning models with the soft parameter sharing structure extremely rely on the task relatedness. To tackle this limitation, we proposed PS-MTL-LUCAS, a multi-task learning with a partially shared structure in this study. An additional shared layer was utilized to learn the general informative representations and interact with each task-specific layer. The partially shared structure ensured the maximum information flow between layers, thereby boosting the prediction performance. Also, the SHapley Addictive exPlanations (SHAP) algorithm was adopted to extract the feature wavelengths of each soil property. PS-MTL-LUCAS was validated on the LUCAS topsoil dataset (2009), and the experimental result suggested that PS-MTL-LUCAS dominated state-of-the-art models by achieving the determination of coefficient at 0.945, 0.936, 0.413, 0.624, 0.837, 0.952, and 0.956 for pH, N, P, K, CEC, OC, and CaCO3, respectively. In summary, this study highlighted the use of the soil spectroscopy and multi-task learning techniques in the soil property prediction task and provided a very promising approach for soil studies.

Advancing soil property prediction with encoder-decoder structures integrating traditional deep learning methods in Vis-NIR spectroscopy

The technology for estimating soil properties using visible and near-infrared spectroscopy has been maturing, with corresponding advances and breakthroughs in deep learning models. In this study, based on the large soil spectral library LUCAS, we explore the potential of encoder-decoder structures to improve convolutional neural network regression predictions. By introducing an encoder-decoder structure into the feature channels of a six-layer CNN model (TRNN model), we significantly enhanced the performance of shallow CNN models and successfully carried out regression predictions for seven soil properties. We employed IntegratedGradients, DeepLift, GradientShap, and DeepLiftShap methods to interpret the output of the TRNN model. Our TRNN model, built on raw spectra, demonstrated high accuracy in predicting multiple soil properties, outperforming residual architectures, LSTMs, various CNN architectures, and other traditional machine learning methods proposed in previous studies. We also investigated the impact of multi-task output structures (TRNN 1-M and TRNN M−M) and single-task output structures (TRNN 1-1) on model performance. For the TRNN model with an encoder-decoder structure, multi-task output structures resulted in a reduction in performance. The TRNN showed outstanding results in regression analysis of the seven soil properties selected in this study (cation exchange capacity, organic carbon content, calcium carbonate content, pH, clay content, silt content, and sand content), with R2 values exceeding 0.93 for all seven properties. Different soil characteristics correspond to different wavelengths, with multiple characteristic peaks commonly observed. This research convincingly demonstrates the enormous potential of combining large model architectures with traditional deep learning approaches for predicting soil properties, which could significantly advance precision agriculture.

Predicting the soil bulk density using a new spectral PTF based on intact samples

Sample collection and measurement of soil bulk density (BD) are often labor-intensive and expensive in large regions. Conversely, soil spectra are easy to measure and facilitate BD prediction. However, the literature suggests that the damage to the physical structure of soil during scanning spectra on the ground and/or sieved samples might hinder the capacity of spectral technology to accurately predict BD. In addition, because some soil properties that have high correlations with BD, such as the soil organic matter (SOM), are routinely measured and available in most soil databases, coupling them with soil spectra may improve BD prediction compared to using soil properties or spectra. Therefore, in this study, we propose a novel spectral pedo-transfer function (spectral PTF) that couples the measured visible and near-infrared spectra of soils on intact samples and other soil properties to accurately predict the BD (BD = f (soil spectra, soil properties)), which is different from the traditional PTF that uses only soil properties (BD = f (soil properties)) or spectra alone (BD = f (soil spectra)). In this study, we collected topsoil (0–20 cm) and subsoil (20–40 cm) samples from 586 sites in Northeast China, covering a large area of 1.09 million km2 characterized by black soils with high SOM contents. Five routinely measured soil properties were selected: SOM, moisture content (MC), Sand, Silt, and Clay, and various spectral PTFs with one, two, and three soil properties were calibrated using the partial least square regression. The cross-validation results show that the traditional PTF can only predict BD for subsoil with an R2 of 0.51 and an RMSE of 0.11 g·cm−3 when using SOM + MC + Silt or SOM + MC. Compared to subsoil, topsoil and all layers (topsoil + subsoil) had a lower BD prediction accuracy, and a saturation effect was observed for BD values above 1.5 g·cm−3. Unexpectedly, the soil spectra did not provide a higher BD prediction accuracy than traditional PTFs, although the spectra were measured on intact samples. However, adding soil properties to the spectral PTF improved the prediction accuracy and saturation effect for high BD values. The optimal spectral PTF with a single soil property (MC) showed an acceptable BD prediction performance with R2≥0.49, RPD>1.4, and RPIQ>1.7 regardless of whether the sample was topsoil, subsoil, or all layers. Furthermore, the spectral PTF with two or three soil properties yielded a slightly better prediction performance and a more stable prediction among different combinations of soil properties. These results indicate that soil properties and spectra are irreplaceable for BD prediction. Our study demonstrates the potential of spectral PTFs for the accurate prediction of BD and offers insights into the prediction of other soil properties using soil spectra.

Effects of Sample Handling on Visible and Near-Infrared Soil Reflectance

Soil reflectance is a cumulative attribute determined by interactions between light (photons) and the physical, chemical, and biological properties of soil. Soil properties such as organic matter, moisture, mineral oxide contents, soil texture, and surface roughness all influence soil reflectance at unique wavelengths. Many standard sample preparation techniques are designed to alter soil properties, so as to homogenize samples to improve the consistency of reflectance data collected. This study aims to quantify the effects of one standard sample preparation activity, drying, and repeated wetting–drying cycles on visible–near-infrared soil reflectance, by collecting reflectance data from nine samples which were dried then wetted for a total of three times. This study demonstrates the major, permanent effects of a drying and wetting cycle on soil reflectance, and then presents a model which can be used to correct for these effects. These results have direct implications for remote sensing activities, soil libraries, and soil spectral libraries. These results show that without correction, data from soil spectral libraries, spectral data collected from stored samples, and spectral data transferred between studies, have limited utility for characterizing soils as they exist in the field. Soil spectral data collected in the lab requires correction before it may be used to predict soil properties in the field, as standard sample handling procedures change intrinsic soil properties and introduce systematic error into these data.

A new SPT-CPT correlation by statistical analysis based on a database from Anhui, China

ABSTRACT The standard penetration test (SPT) and cone penetration test (CPT) are the 2 most commonly used in situ tests in geotechnical investigations. In this study, various existing SPT–CPT correlations were assessed based on the database of Anhui Province, China, using data from 802 pairs of SPT–CPT measurements. The results showed that the existing correlations from other regions were not suitable for the geological conditions in Anhui Province. A new SPT–CPT conversion model based on the S curve was proposed through statistical analysis. This model introduced the soil behaviour index (Ic ) to SPT–CPT correlations to involve the influence of soil types on data interpretation. Comparative analysis demonstrated that the proposed correlation had the highest accuracy compared to correlations found in the literature, with the highest R2 value and lowest RMSE. Also, to validate the proposed model further, an additional set of investigation data was collected in Wuhu, Anhui, to verify the correlation. The correlation was successfully applied to predict a range of soil properties using the transformed N60 values. The validation analysis confirmed the highest accuracy of the correlation and its ability to effectively predict soil properties, thereby assisting in geotechnical design.

Multi-Scale Spatial Attention-Based Multi-Channel 2D Convolutional Network for Soil Property Prediction.

Visible near-infrared spectroscopy (VNIR) is extensively researched for obtaining soil property information due to its rapid, cost-effective, and environmentally friendly advantages. Despite its widespread application and significant achievements in soil property analysis, current soil prediction models continue to suffer from low accuracy. To address this issue, we propose a convolutional neural network model that can achieve high-precision soil property prediction by creating 2D multi-channel inputs and applying a multi-scale spatial attention mechanism. Initially, we explored two-dimensional multi-channel inputs for seven soil properties in the public LUCAS spectral dataset using the Gramian Angular Field (GAF) method and various preprocessing techniques. Subsequently, we developed a convolutional neural network model with a multi-scale spatial attention mechanism to improve the network's extraction of relevant spatial contextual information. Our proposed model showed superior performance in a statistical comparison with current state-of-the-art techniques. The RMSE (R²) values for various soil properties were as follows: organic carbon content (OC) of 19.083 (0.955), calcium carbonate content (CaCO3) of 24.901 (0.961), nitrogen content (N) of 0.969 (0.933), cation exchange capacity (CEC) of 6.52 (0.803), pH in H2O of 0.366 (0.927), clay content of 4.845 (0.86), and sand content of 12.069 (0.789). Our proposed model can effectively extract features from visible near-infrared spectroscopy data, contributing to the precise detection of soil properties.

Performance Evaluation and Comparison of Deep Neural Network Models for African Soil Properties Prediction

ABSTRACT The significance of soil in agriculture and human survival cannot be overstated. Given its limited supply, improving soil properties is imperative. This study addresses this challenge by predicting five major soil properties—organic carbon, pH values, Mehlich-3 extractable calcium, Mehlich-3 extractable phosphorus, and sand content—utilizing mid-infrared absorbance measurements from the African Soil Information Service (AfSIS) dataset, covering non-desert regions of Africa. With a pressing need for a reliable and efficient model to predict African soil properties using spectral measurements, this study fills a crucial gap, addressing the scarcity of functional soil property databases in Africa. The developed model eliminates costly soil sample preparation and lengthy chemical analysis, applicable in both onsite and laboratory settings for determining soil functional properties. By employing stacked autoencoders for feature dimensionality reduction and combining discrete wavelet transform with two feature selection methods (stepwise regression and random forest) to build robust multi-layer perceptron (MLP) models, the study offers a comprehensive approach. Evaluation metrics including root mean square error (RMSE), scatter index (SI), Variance Accounted For (VAF), Nash-Sutcliffe Model Efficiency (NSE), and Correlation Coefficient (R) demonstrate the superior performance of the MLP model informed by stacked autoencoder-selected features, outperforming models informed by wavelet-transformed features and partial least square regression. This best-performing autoencoder-based model presents a valuable tool for soil scientists tasked with modeling African soil properties.

Geotechnical Site Characterizations Using a Bayesian-Optimized Multi-Output Gaussian Process

Geotechnical site characterizations aim to determine site-specific subsurface profiles and provide a comprehensive understanding of associated soil properties, which are important for geotechnical engineering design. Traditional methods often neglect the inherent cross-correlations among different soil properties, leading to high bias in site characterization interpretations. This paper introduces a novel data-driven site characterization (DDSC) method that employs the Bayesian-optimized multi-output Gaussian process (BO-MOGP) to capture both the spatial correlations across different site locations and the cross-correlations among various soil properties. By considering the dual-correlation feature, the proposed BO-MOGP method enhances the accuracy of predictions of soil properties by leveraging information as much as possible across multiple soil properties. The superiority of the proposed method is demonstrated through a simulated example and the case study of a Taipei construction site. These examples illustrate that the proposed BO-MOGP method outperforms traditional methods that fail to consider both types of correlations, as evidenced by the reduced prediction uncertainty and the accurate identification of cross-correlations. Furthermore, the ability of the proposed BO-MOGP method to generate conditional random fields supports its effectiveness in geotechnical site characterizations.

Real-time inference of compacted soil properties using deflection tests: An AI-driven solution informed by unsaturated soil mechanics principles

This paper presents a novel Artificial Intelligence (AI)-driven tool designed to convert deflection test results into crucial soil parameters essential for quality assurance in compaction projects. The accurate determination of these parameters, such as density and void ratio, is imperative for ensuring the structural integrity of infrastructures constructed on such soils. Moreover, it facilitates the utilization of modern non-destructive equipment in compaction endeavors. The determination of these parameters is notably challenging in unsaturated soils owing to the intricate interplay among factors such as suction, moisture content, void ratio, and resulting deflection.This paper presents a pioneering tool to address these challenges. By integrating unsaturated soil mechanics with advanced AI techniques, particularly reinforcement learning, the tool leverages a diverse array of inputs, including in-situ data, experimental observations, and physics-based modeling. This integration enables dynamic adaptation to changing field conditions and ensuring the tool’s real-time adaptability and predictive accuracy.Field trials validated the tool’s efficacy in predicting soil properties accurately without direct measurements of moisture content or suction, variables often unmeasured in practical soil compaction projects. This unique capability underscores significant advancements in real-time assessment of unsaturated soils, illustrating the transformative potential of AI in geotechnical engineering and unsaturated soil mechanics.