Soil Particle Prediction using Spatial Ordinary Logistic Regression

Purpose: This study embarks on a novel journey to create a predictive model for soil texture. We utilize the Spatial Ordinary Logistic Regression model to estimate soil particles in the topsoil with unprecedented accuracy. This involves employing Geographically Weighted Ordinary Logistic Regression to analyze and map the spatial distribution of these particles based on primary data collected from the field.Design/methodology/approach: This study distinguishes itself by adopting a meticulous approach to gathering soil particulate and geospatial data from various random locations. This method is crucial in addressing the complexity of modelling soil texture, an essential aspect of soil management. The study analyzes soil texture, a combination of sand, silt, and clay, using Digital Elevation Model (DEM) data. By leveraging topographical variations, the study predicts soil texture, employing Geographically Weighted Ordinary Logistic Regression for areas without direct observations. This approach significantly enhances both understanding and prediction in soil science.Findings: The proposed model will be cross-validated to ensure precision. Aimed at aiding land and resource management, this study focuses on examining spatial variations in topsoil particle sizes and their influencing factors. The Geographic Weighted Ordinary Logistic Regression (GWOLR) model, designed for estimating soil particle sizes using a fixed bi-square weight, demonstrated superior effectiveness with a 90% accuracy rate compared to the standard model's 88%. Further findings show that all topographical predictors exhibit significant spatial autocorrelation (Moran’s I, p < 0.05), justifying the spatial approach. The GWOLR model also provides localized parameter estimates, revealing spatial heterogeneity in the influence of terrain features. Specifically, vertical curvature and slope positively associate with sandy textures, while lower northness aspects correlate with higher clay and silt presence. Spatial prediction maps generated from the model align closely with actual field data, affirming its practical value in precision agriculture, land-use planning, and targeted soil conservation strategies.Practical Implications: In 2023, soil particle size data were gathered from the Kalikonto Watershed Area in Batu City, East Java, Indonesia. This data, divided into three categories, was analyzed using the Geographically Weighted Ordinal Logistic Regression method, incorporating spatial factors.Originality/value: This study presents innovative methods for enhanced spatial analysis, notably the Geographically Weighted Ordinary Logistic Regression technique. This approach improves spatial and statistical data integration for analyzing geographic information, offering insights into how spatial variables influence soil properties. Focusing on estimating particle-size fractions in the soil's top layer, the research underscores the significance of soil attributes on plant growth and agricultural productivity. Furthermore, it provides new perspectives in the crucial field of soil property investigation.