Integrating Proximal Sensing, high-resolution Imagery, and Machine Learning for Field-Scale Soil Salinity Mapping in Semi-Arid Region

Soil salinity is a major environmental challenge that reduces agricultural productivity and degrades soil health, especially in arid and semi-arid regions. Conventional soil salinity assessment methods involve extensive manual labor and are time-consuming In this study, we explored alternative approaches by using a combination of proximal sensing data (i.e., electromagnetic (EM) induction instruments, EM 38-MK2) with two very high-resolution multi-spectral and -sources imagery (i.e., a UAV (Unnamed Aerial Vehicle) and PlanetScope (PS)), topographic attributes, and machine learning methods to achieve field-scale soil salinity mapping under data-scarce conditions. To do so, an initial set of 26 topsoil samples (0–5 cm) were collected from a saline field in the semi-arid area of Sehb El Masjoune in Southern Morocco. and their Electrical conductivity (EC, a proxy of salinity) was determined at the lab. Then, proximal sensed data from EM38 were collected along the same field and measured apparent soil electrical conductivity (ECa – dS/m) was correlated with measured topsoil EC. We used proximal sensing technology to generate 500 EC (electrical conductivity) observations for spatialization, thereby creating a robust dataset for training four machine learning models: partial least squares regression (PLSR), support vector machine (SVM), random forest (RF), and an ensemble (stacked) model. Among these models, the RF and ensemble approaches delivered the highest accuracy, with RF outperforming all others. Performance assessments indicated that PlanetScope data achieved R² = 0.91 and RMSE = 3.47, while UAV data showed R² = 0.89 and RMSE = 3.83. These findings underscore that integrating multisource data, even in data-scarce environments, enhances reliability and robustness in soil salinity mapping at the field scale. Our results highlight a cost-effective, high-precision strategy for characterizing saline and sodic soils, offering valuable insights for targeted reclamation and management interventions in arid and semi-arid regions. We conclude that the used approach not only contributes to the scientific understanding of soil salinity dynamics but also provides practical implications for sustainable land management and agricultural planning. The research highlights the potential of combining cutting-edge technology with environmental predictors to address critical global issues.