Soil salinization is a widespread form of land degradation that threatens crop productivity, soil health, and long-term agricultural sustainability worldwide. Traditional soil salinity monitoring relies heavily on field sampling and laboratory analysis, which are labor-intensive, time-consuming, and poorly suited for capturing spatial heterogeneity at fine scales. Remote sensing techniques provide broader coverage but often lack sufficient spatial resolution or fail to account for complex soil–environment interactions. Moreover, soil salinity is influenced by multiple interacting factors, including spectral characteristics, soil organic matter, and pH, making accurate estimation particularly difficult. Based on these challenges, it is necessary to conduct in-depth research on integrated, high-precision soil salinity estimation methods.
Researchers from the Institute of Agricultural Resources and Regional Planning, Chinese Academy of Agricultural Sciences reported this work in Journal of Remote Sensing, published (DOI: 10.34133/remotesensing.0805) on January 15, 2026. The study addresses a critical challenge in modern agriculture: how to reliably estimate soil salinity at the field scale using fast, non-destructive technologies. By combining UAV multispectral imagery with soil auxiliary data and advanced machine-learning techniques, the research offers a practical solution for monitoring saline soils, supporting precision farming, and improving land management decisions in salt-affected regions.
The study proposes a novel feature-optimized and performance-weighted ensemble learning framework that outperforms traditional machine-learning and ensemble models. By intelligently selecting the most sensitive spectral and soil variables and dynamically weighting multiple base learners, the model achieves higher accuracy and greater robustness than existing approaches. The method consistently delivered superior prediction performance across multiple evaluation metrics, including determination coefficient and prediction stability. Notably, the framework demonstrated low uncertainty in soil salinity estimation while maintaining fine spatial resolution. Compared with single models or conventional stacking methods, the new approach better captures nonlinear relationships and spatial variability, enabling reliable soil salinity mapping at the sub-meter scale.
High-resolution multispectral UAV data were collected over representative saline agricultural fields, capturing green, red, red-edge, and near-infrared bands. These spectral data were combined with laboratory-measured soil salinity, soil organic matter, and pH information. A hybrid embedded feature-selection strategy was applied to identify the most influential variables driving soil salinity variation. Among them, specific salinity indices derived from UAV imagery, together with soil organic matter and pH, emerged as the most sensitive predictors.
To further enhance performance, the study employed a performance-weighted ensemble learning strategy that evaluates multiple machine-learning models and assigns weights based on their predictive reliability. This approach reduced model redundancy and improved generalization. Validation results showed that the proposed framework achieved a coefficient of determination exceeding 0.75, alongside reduced prediction error and uncertainty compared with benchmark models. Spatial mapping revealed clear salinity gradients and localized hotspots, demonstrating the model’s ability to resolve fine-scale soil salinity patterns critical for precision agriculture applications.
“Our approach demonstrates how intelligent model integration can unlock the full potential of UAV data for soil monitoring,” said one of the study’s lead researchers. “By combining optimized feature selection with performance-based weighting, we achieved both higher accuracy and greater stability. This framework has strong potential to support real-world agricultural decision-making, especially in regions facing increasing salinization pressures.”
The research combined field soil sampling, laboratory chemical analysis, UAV multispectral data acquisition, and advanced machine-learning modeling. Soil samples were collected and analyzed for salinity, organic matter, and pH. UAV imagery was processed to extract spectral bands and salinity-related indices. Multiple machine-learning algorithms were trained and evaluated using repeated cross-validation. A hybrid feature-selection method and a weighted ensemble strategy were then applied to construct the final predictive model and generate high-resolution soil salinity maps.
This integrated UAV-based ensemble learning framework offers a promising pathway toward scalable, low-cost soil salinity monitoring. In the future, the approach could be extended to larger regions, different soil types, and other soil properties such as moisture or nutrient status. Its application may support precision irrigation, crop selection, and land reclamation strategies, contributing to sustainable agricultural systems under climate change. More broadly, the framework highlights how combining UAV technology with intelligent data analytics can reshape environmental monitoring and smart farming worldwide.
###
References
DOI
10.34133/remotesensing.0805
Original Source URL
https://spj.science.org/doi/10.34133/remotesensing.0805
Funding Information
This study was supported by the Cropland Degradation Monitoring Project.
About Journal of Remote Sensing
The Journal of Remote Sensing, an online-only Open Access journal published in association with AIR-CAS, promotes the theory, science, and technology of remote sensing, as well as interdisciplinary research within earth and information science.