Spatial prediction of soil salinity: Remote sensing and machine learning approach

Abstract

Soil salinity significantly affects agricultural productivity and land sustainability, necessitating efficient monitoring and predictive strategies. This study leverages remote sensing data and machine learning algorithms to spatially predict soil salinity across the Puerto Vallarta region, covering an area of 716.03 km². To enhance prediction accuracy, a diverse set of indices were employed, including Salinity Indices (SI 1, SI 2, SI 3, and SI 11), Intensity Indices (INT 1 and INT 2), Brightness Index (BI), Soil-Adjusted Vegetation Index (SAVI), Enhanced Vegetation Index (EVI), Normalized Difference Vegetation Index (NDVI), and the ratio of spectral bands (B2 and B5). These indices deliver appreciated understandings into soil and vegetation properties, allowing accurate classification and salinity assessment. Soil salinity was categorized into five classes: Very Low (16.54%), Low (23.16%), Moderate (21.97%), High (19.95%), and Very High (18.38%). The machine learning models employed for this study included Random Forest (RF), Artificial Neural Network (ANN), and Support Vector Machine (SVM). Among these, RF exhibited the highest prediction accuracy (92.1%), followed by ANN (89.5%) and Support Vector Machine (86.2%). These results underscore the potential of RF as a robust tool for analyzing spatially complex environmental datasets. Integrating remote sensing parameters and machine learning algorithms demonstrates an effective approach for mapping and managing soil salinity. The findings support sustainable land management by pinpointing critical salinity zones that need urgent intervention. Furthermore, the methodology established in this learning can be practical to other regions facing parallel challenges. This research highlights the significance of data-driven approaches in environmental monitoring and resource management.