Soil erosion susceptibility mapping in Bangladesh

Generating high-resolution daily soil moisture by using spatial downscaling techniques: a comparison of six machine learning algorithms

Machine learning approach to predict susceptible growth regions of Moringa peregrina (Forssk)

Introducing two Random Forest based methods for cloud detection in remote sensing images

Context Correct diagnosis of the state of soil compaction is a challenge in continuous no-tillage (NT). Aims and methods The aim of this study was to evaluate the performance of four machine learning algorithms to diagnose the state of soil compaction (NT and occasional tillage – OT). For these purposes, data from a field experiment conducted in a clayey Typic Hapludox with mechanical (chiselling and subsoiling) and chemical (gypsum and limestone) methods for mitigation of soil compaction were used. To diagnose the state of soil compaction, soil physical properties [soil bulk density, penetration resistance, macroporosity (MAC), microporosity (MIC), air capacity (AC), available water content, relative field capacity and total porosity (TP)] in addition to crop yield (Rel_Yield) were used as predictor variables for Classification and Regression Trees (CART), Random Forest (RF), Artificial Neural Network (ANN) and Support Vector Machine (SVM) algorithms. Key results The most important variables for predicting the state of soil compaction were Rel_Yield and soil porosity (MAC, TP, MIC and AC). The machine learning algorithms had satisfactory performance in diagnosing which sites were compacted and which were not. The decision tree algorithms (CART and RF) performed better than ANN and SVM, reaching accuracy = 0.90, Kappa index = 0.76 and sensitivity = 0.83. Conclusions and implications The machine learning algorithm approach proved to be an efficient tool in diagnosing soil compaction in continuous NT, improving decision-making concerning the use of OT.

Savannah grazing lawns are a key food resource for large herbivores such as blue wildebeest (Connochaetes taurinus), hippopotamus (Hippopotamus amphibius) and white rhino (Ceratotherium simum), and impact herbivore densities, movement and recruitment rates. They also exert a strong influence on fire behaviour including frequency, intensity and spread. Thus, variation in grazing lawn cover can have a profound impact on broader savannah ecosystem dynamics. However, knowledge of their present cover and distribution is limited. Importantly, we lack a robust, broad-scale approach for detecting and monitoring grazing lawns, which is critical to enhancing understanding of the ecology of these vital grassland systems. We selected two sites in the Lower Sabie and Satara regions of Kruger National Park, South Africa with mesic and semiarid conditions, respectively. Using spectral and texture features derived from WorldView-3 imagery, we (i) parameterised and assessed the quality of Random Forest (RF), Support Vector Machines (SVM), Classification and Regression Trees (CART) and Multilayer Perceptron (MLP) models for general discrimination of plant functional types (PFTs) within a sub-area of the Lower Sabie landscape, and (ii) compared model performance for probabilistic mapping of grazing lawns in the broader Lower Sabie and Satara landscapes. Further, we used spatial metrics to analyse spatial patterns in grazing lawn distribution in both landscapes along a gradient of distance from waterbodies. All machine learning models achieved high F-scores (F1) and overall accuracy (OA) scores in general savannah PFTs classification, with RF (F1 = 95.73±0.004%, OA = 94.16±0.004%), SVM (F1 = 95.64±0.002%, OA = 94.02±0.002%) and MLP (F1 = 95.71±0.003%, OA = 94.27±0.003%) forming a cluster of the better performing models and marginally outperforming CART (F1 = 92.74±0.006%, OA = 90.93±0.003%). Grazing lawn detection accuracy followed a similar trend within the Lower Sabie landscape, with RF, SVM, MLP and CART achieving F-scores of 0.89, 0.93, 0.94 and 0.81, respectively. Transferring models to the Satara landscape however resulted in relatively lower but high grazing lawn detection accuracies across models (RF = 0.87, SVM = 0.88, MLP = 0.85 and CART = 0.75). Results from spatial pattern analysis revealed a relatively higher proportion of grazing lawn cover under semiarid savannah conditions (Satara) compared to the mesic savannah landscape (Lower Sabie). Additionally, the results show strong negative correlation between grazing lawn spatial structure (fractional cover, patch size and connectivity) and distance from waterbodies, with larger and contiguous grazing lawn patches occurring in close proximity to waterbodies in both landscapes. The proposed machine learning approach provides a novel and robust workflow for accurate and consistent landscape-scale monitoring of grazing lawns, while our findings and research outputs provide timely information critical for understanding habitat heterogeneity in southern African savannahs.

Poyang Lake is not only a globally important stopover for migratory birds and a habitat for fish, but also one of the main sand mining areas in China. However, sand mining activities significantly affect the suspended sediment concentration in Poyang Lake, thereby impacting water quality, altering lake bed topography, and potentially disturbing the living environment of plants and animals. Therefore, understanding the suspended sediment concentration and the spatiotemporal patterns of sand mining activities in Poyang Lake holds vital significance for effective lake management and biodiversity conservation. Utilizing surface reflectance (SR) data derived from Landsat satellite imagery sourced via Google Earth Engine spanning from 1989 to 2018, coupled with SSC data recorded by daily observation at Hukou hydrological stations, we conducted a comparative analysis of five distinct methods: Linear Regression (LR), Support Vector Machine (SVM), Random Forest (RF), Classification and Regression Trees (CART), and Back Propagation Neural Network (BPNN). Moreover, the applicability of machine learning methods was analyzed and compared across four scenarios (#01 Landsat 5, #02 Landsat 7, #03 Landsat 8, and #04, Landsat 5, 7, and 8). Using the optimal SSC estimation model, we investigated the spatial and temporal trends of sand mining activities in Poyang Lake from 1989 to 2021, along with their potential impact on suspended sediment levels in the water body. The results showed that the prediction accuracy of machine learning is better than that of linear regression models, and the RF model has the best performance. The RF model generated R<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$^{2} \gt 0.9$ </tex-math></inline-formula> for four scenarios (#01 Landsat 5, #02 Landsat 7, #03 Landsat 8, and #04 Landsat 5, 7, and 8) and showed little to no overfitting. The generated SSC map can clearly show the distribution of SSC in Poyang Lake, unveiling the impact of sand mining activities on suspended sediment, especially around sand dredgers where SSC can exceed 0.15 g/L. Sand mining activities in Poyang Lake emerged after 2000 and gradually shifted southward and expanded, reaching a peak in 2016. Fortunately, under government regulation, illegal sand mining has been effectively controlled and is currently concentrated near Songmenshan Island. Despite the fact that sand mining has been brought under control, there remains a necessity for heightened oversight to prevent any impact on national nature reserves. The Random Forest (RF) model demonstrates significant potential in utilizing Landsat satellite data to predict SSC in Poyang Lake, as well as to monitor sand mining activities in the area.

To construct and validate the best predictive model for 28-day death risk in patients with septic shock based on different supervised machine learning algorithms. The patients with septic shock meeting the Sepsis-3 criteria were selected from Medical Information Mart for Intensive Care-IV v2.0 (MIMIC-IV v2.0). According to the principle of random allocation, 70% of these patients were used as the training set, and 30% as the validation set. Relevant predictive variables were extracted from three aspects: demographic characteristics and basic vital signs, serum indicators within 24 hours of intensive care unit (ICU) admission and complications possibly affecting indicators, functional scoring and advanced life support. The predictive efficacy of models constructed using five mainstream machine learning algorithms including decision tree classification and regression tree (CART), random forest (RF), support vector machine (SVM), linear regression (LR), and super learner [SL; combined CART, RF and extreme gradient boosting (XGBoost)] for 28-day death in patients with septic shock was compared, and the best algorithm model was selected. The optimal predictive variables were determined by intersecting the results from LASSO regression, RF, and XGBoost algorithms, and a predictive model was constructed. The predictive efficacy of the model was validated by drawing receiver operator characteristic curve (ROC curve), the accuracy of the model was assessed using calibration curves, and the practicality of the model was verified through decision curve analysis (DCA). A total of 3 295 patients with septic shock were included, with 2 164 surviving and 1 131 dying within 28 days, resulting in a mortality of 34.32%. Of these, 2 307 were in the training set (with 792 deaths within 28 days, a mortality of 34.33%), and 988 in the validation set (with 339 deaths within 28 days, a mortality of 34.31%). Five machine learning models were established based on the training set data. After including variables at three aspects, the area under the ROC curve (AUC) of RF, SVM, and LR machine learning algorithm models for predicting 28-day death in septic shock patients in the validation set was 0.823 [95% confidence interval (95%CI) was 0.795-0.849], 0.823 (95%CI was 0.796-0.849), and 0.810 (95%CI was 0.782-0.838), respectively, which were higher than that of the CART algorithm model (AUC = 0.750, 95%CI was 0.717-0.782) and SL algorithm model (AUC = 0.756, 95%CI was 0.724-0.789). Thus above three algorithm models were determined to be the best algorithm models. After integrating variables from three aspects, 16 optimal predictive variables were identified through intersection by LASSO regression, RF, and XGBoost algorithms, including the highest pH value, the highest albumin (Alb), the highest body temperature, the lowest lactic acid (Lac), the highest Lac, the highest serum creatinine (SCr), the highest Ca2+, the lowest hemoglobin (Hb), the lowest white blood cell count (WBC), age, simplified acute physiology score III (SAPS III), the highest WBC, acute physiology score III (APS III), the lowest Na+, body mass index (BMI), and the shortest activated partial thromboplastin time (APTT) within 24 hours of ICU admission. ROC curve analysis showed that the Logistic regression model constructed with above 16 optimal predictive variables was the best predictive model, with an AUC of 0.806 (95%CI was 0.778-0.835) in the validation set. The calibration curve and DCA curve showed that this model had high accuracy and the highest net benefit could reach 0.3, which was significantly outperforming traditional models based on single functional score [APS III score, SAPS III score, and sequential organ failure assessment (SOFA) score] with AUC (95%CI) of 0.746 (0.715-0.778), 0.765 (0.734-0.796), and 0.625 (0.589-0.661), respectively. The Logistic regression model, constructed using 16 optimal predictive variables including pH value, Alb, body temperature, Lac, SCr, Ca2+, Hb, WBC, SAPS III score, APS III score, Na+, BMI, and APTT, is identified as the best predictive model for the 28-day death risk in patients with septic shock. Its performance is stable, with high discriminative ability and accuracy.

BackgroundLumbar drainage is widely used in the clinic; however, forecasting lumbar drainage-related meningitis (LDRM) is limited. We aimed to establish prediction models using supervised machine learning (ML) algorithms.MethodsWe utilized a cohort of 273 eligible lumbar drainage cases. Data were preprocessed and split into training and testing sets. Optimal hyper-parameters were archived by 10-fold cross-validation and grid search. The support vector machine (SVM), random forest (RF), and artificial neural network (ANN) were adopted for model training. The area under the operating characteristic curve (AUROC) and precision-recall curve (AUPRC), true positive ratio (TPR), true negative ratio (TNR), specificity, sensitivity, accuracy, and kappa coefficient were used for model evaluation. All trained models were internally validated. The importance of features was also analyzed.ResultsIn the training set, all the models had AUROC exceeding 0.8. SVM and the RF models had an AUPRC of more than 0.6, but the ANN model had an unexpectedly low AUPRC (0.380). The RF and ANN models revealed similar TPR, whereas the ANN model had a higher TNR and demonstrated better specificity, sensitivity, accuracy, and kappa efficiency. In the testing set, most performance indicators of established models decreased. However, the RF and AVM models maintained adequate AUROC (0.828 vs. 0.719) and AUPRC (0.413 vs. 0.520), and the RF model also had better TPR, specificity, sensitivity, accuracy, and kappa efficiency. Site leakage showed the most considerable mean decrease in accuracy.ConclusionsThe RF and SVM models could predict LDRM, in which the RF model owned the best performance, and site leakage was the most meaningful predictor.

Soil cadmium (Cd) contamination has attracted a great deal of attention because of its detrimental effects on animals and humans. This study aimed to develop and compare the performances of stepwise linear regression (SLR), classification and regression tree (CART) and random forest (RF) models in the prediction and mapping of the spatial distribution of soil Cd and to identify likely sources of Cd accumulation in Fuyang County, eastern China. Soil Cd data from 276 topsoil (0–20 cm) samples were collected and randomly divided into calibration (222 samples) and validation datasets (54 samples). Auxiliary data, including detailed land use information, soil organic matter, soil pH, and topographic data, were incorporated into the models to simulate the soil Cd concentrations and further identify the main factors influencing soil Cd variation. The predictive models for soil Cd concentration exhibited acceptable overall accuracies (72.22% for SLR, 70.37% for CART, and 75.93% for RF). The SLR model exhibited the largest predicted deviation, with a mean error (ME) of 0.074 mg/kg, a mean absolute error (MAE) of 0.160 mg/kg, and a root mean squared error (RMSE) of 0.274 mg/kg, and the RF model produced the results closest to the observed values, with an ME of 0.002 mg/kg, an MAE of 0.132 mg/kg, and an RMSE of 0.198 mg/kg. The RF model also exhibited the greatest R2 value (0.772). The CART model predictions closely followed, with ME, MAE, RMSE, and R2 values of 0.013 mg/kg, 0.154 mg/kg, 0.230 mg/kg and 0.644, respectively. The three prediction maps generally exhibited similar and realistic spatial patterns of soil Cd contamination. The heavily Cd-affected areas were primarily located in the alluvial valley plain of the Fuchun River and its tributaries because of the dramatic industrialization and urbanization processes that have occurred there. The most important variable for explaining high levels of soil Cd accumulation was the presence of metal smelting industries. The good performance of the RF model was attributable to its ability to handle the non-linear and hierarchical relationships between soil Cd and environmental variables. These results confirm that the RF approach is promising for the prediction and spatial distribution mapping of soil Cd at the regional scale.

Groundwater is considered one of the most valuable fresh water resources. The main objective of this study was to produce groundwater spring potential maps in the Koohrang Watershed, Chaharmahal-e-Bakhtiari Province, Iran, using three machine learning models: boosted regression tree (BRT), classification and regression tree (CART), and random forest (RF). Thirteen hydrological-geological-physiographical (HGP) factors that influence locations of springs were considered in this research. These factors include slope degree, slope aspect, altitude, topographic wetness index (TWI), slope length (LS), plan curvature, profile curvature, distance to rivers, distance to faults, lithology, land use, drainage density, and fault density. Subsequently, groundwater spring potential was modeled and mapped using CART, RF, and BRT algorithms. The predicted results from the three models were validated using the receiver operating characteristics curve (ROC). From 864 springs identified, 605 (≈70 %) locations were used for the spring potential mapping, while the remaining 259 (≈30 %) springs were used for the model validation. The area under the curve (AUC) for the BRT model was calculated as 0.8103 and for CART and RF the AUC were 0.7870 and 0.7119, respectively. Therefore, it was concluded that the BRT model produced the best prediction results while predicting locations of springs followed by CART and RF models, respectively. Geospatially integrated BRT, CART, and RF methods proved to be useful in generating the spring potential map (SPM) with reasonable accuracy.

In order to accurately predict soil properties, various machine learning (ML) approaches and hybrid models constructed by integrating ML into regression kriging framework were used to predict and map arable land topsoil pH in Henan province, central China. Random forest (RF), cubist (Cu), support vector machine, artificial neural network, multiple linear regression, classification and regression trees (CART) and their hybrid models were compared for pH accuracy prediction. Among all standalone ML models, RF had the best predictive performance, in terms of the metrics employed in this study, followed by Cu, and CART was the worst. Compared with their ML counterparts, hybrid models could improve the accuracy of topsoil pH prediction to various extents. The accuracy improvement of the hybrid models constructed based on the simple ML was much greater than that based on the complex ensemble ML. Except for artificial neural network kriging , there was no significant difference between different hybrid models in the predicted results of topsoil pH. The outputs from the best predictive models showed that weak acidic soils and weak alkaline soils were the dominant arable soils in the study region, accounting for more than 30% and more than 50% of the total arable land area respectively, the topsoil pH of arable land in the north of the study area is generally higher than that in the south. Boruta variable selection revealed that altitude, climatic covariates closely related to soil moisture availability and some soil properties were the most critical factors affecting and controlling the topsoil pH of arable land.

Different modelling approaches for predicting titanium dioxide nanoparticles mobility in intact soil media

Landslide susceptibility and building exposure assessment using machine learning models and geospatial analysis techniques

India is ranked 120th among 122 countries globally in WaterAid&amp;#8217;s water quality index. Regular water quality monitoring is essential to determine which inland water bodies are experiencing depreciating water quality. Long-term trends have been obtained using satellite remote sensing, necessitating multiple image analysis. The computational burden of processing numerous satellite images can be reduced using Google Earth Engine&amp;#8217;s (GEE) cloud computing capabilities. Thorough research was conducted to determine the global spatio-temporal and biochemical factors impacting surface water quality (WQ). The public availability of geospatial datasets and free access to cloud-based geo-computing platforms such as GEE are widely used for spatio-temporal mapping, global surface water monitoring, water quality parametric variation, and real-time forecasting. Many researchers have recently focused on improving data mining and machine learning algorithms to accurately deal with image classification and predictive problems for crop identification and monitoring. In the present study, we propose to define band spectral ratios, spectral band equations, and empirical models for water quality parameters. A few neural network models will be used to (i) query and pre-process satellite earth observations that coincide with the study area, (ii) extract the spectra, and (iii) use spectral band wavelength charts, time-series charts, spatial-distribution maps, and the development of an online dashboard application to visualize the results graphically upon using integrated Landsat (8,9), Sentinel-2A/B and PlanetScope satellite data for the pre-monsoon and post-monsoon seasons in 2023 in the pan-India region. MODIS-TERRA provides LST spatial-temporal monitoring. In this, we have assessed and compared the performance of CART, SVM, and RF ML algorithms. We found that RF outperforms CART and SVM algorithms in the GEE platform with PlanetScope data (80.71% overall accuracy (OA) with Kappa 0.89) and also with the integration of PlanetScope and Sentinel-2A/B data (OA = 85.53%, Kappa 0.91). But CART outperforms RF and SVM algorithms with Sentiel-2A/B data (OA = 81.59%, Kappa 0.85). The SAM technique, spectral feature fitting, continuum band removal, and other band-spectral ratio techniques are employed for quantitative hyperspectral data analysis. Specifically, the performance metrics of XGBoost and SGD for both Chl-a (R^2 = 0.818) and Turbidity (R^2 = 0.815) models exhibited robust accuracy. We have also developed a Google API-based JavaScript code that can be tested under complex coastal shores, challenging inundation and variable climatic conditions. Our method provides the end-to-end cloud computing workflow shown in this research, considering cost and computational efficiency for timely information delivery.Keywords: Classification and Regression Trees (CART), Chlorophyll-a (conc), Color dissolved organic matter (CDOM), Composite water management index (CWMI), Environmental Mapping and Analysis Program (EnMAP), Land Surface Temperature (LST), Machine Learning (ML), Random Forest (RF), Spectral Angle Mapper (SAM), Stochastic Gradient Descent (SGD), Support Vector Machine (SVM), Total Suspended Solids (TSS), Water Quality (WQ), XGBoost (Extreme Gradient Boosting).

Introduction: No accurate diagnostic tests exist for exocrine pancreatic insufficiency (EPI). As a result, EPI is misdiagnosed, including in chronic pancreatitis (CP), and there is an unmet need for clinical tools to help clinicians evaluate patients with CP at risk for EPI. The aim of this study was to generate a simple clinical tool to predict EPI in patients with CP. Methods: 49 variables from medical records of CP patients with or without EPI were entered into a Classification and Regression Tree (CART) model using the unimputed full analysis set and 4 other models (Logistic Regression with Least Absolute Shrinkage and Selection Operator regularization, Support Vector Machine, Random Forest, and Gradient Boosting Machine) using imputed full analysis set. EPI misclassification rate (mRate) served as primary metric to train the 5 prediction models. Each model’s generalizability was assessed using 5-fold cross-validation and the model that was best suited for clinician use based on mRate and other characteristics was selected. Results: Records of 274 patients with CP from 6 pancreatitis centers across the US were included. Among the 160 (58%)/114 (42%) patients with EPI/without EPI, 48%/55% were female, mean age was 55/52 years, 76%/69% were white, 29%/52% were never smokers, and 39%/53% were never alcohol users. All 5 prediction models demonstrated similar EPI mRates. The CART model was selected. The final CART decision tree includes 10 variables (in order of importance): smoking, pancreatic atrophy, steatorrhea, alcohol use, hemoglobin, serum vitamin D, pancreatic duct contour change, serum albumin, pancreatic parenchymal calcifications, and use of opioid analgesics. The mRate without and with 5-fold cross-validation of the CART was 0.153 (training error) and 0.314 (prediction error), and the area under the receiver operating characteristic curve was 0.889 and 0.682, respectively (Figure 1). Sensitivity/specificity without/with 5-fold cross validation was 0.888/0.789 and 0.794/0.535, respectively. Conclusion: A parsimonious set of 10 variables associated with EPI in CP was identified. The CART decision tree demonstrated good prediction performance for EPI and was selected as the most practical screening tool to aid clinicians in the assessment of EPI in patients with CP. Further external validation of the EPI screening tool is planned.Figure 1.: AUC-ROC of the CART model trained on the full analysis set and after 5-fold cross-validation. AUC, area under the curve; CART, classification and regression tree; CV, cross-validation; ROC, receiver operating characteristic.