Development of a Remote Sensing-Based Soil Nutrient Monitoring System

Thesis Abstract

Soil nutrient management remains a critical challenge in sustainable agriculture, particularly due to the limited capacity of traditional soil testing methods to provide timely, spatially explicit information over large and inaccessible areas. This study aims to develop a remote sensing-based soil nutrient monitoring system that enhances the accuracy, efficiency, and accessibility of nutrient assessment in agricultural landscapes. The primary objectives include evaluating the correlation between multispectral satellite imagery and soil nutrient levels, designing predictive models for nutrient estimation, and developing an integrated spatial information system for decision support. Employing a mixed-methods research design, the study integrates quantitative analysis of spectral data with qualitative validation from field samples. The research population comprises agricultural plots within the Central Valley region, with a purposive sample of 300 georeferenced sites representative of different soil types and crop regimes. Data collection involved acquiring high-resolution multispectral imagery from Sentinel-2 satellites, supplemented by in situ soil sampling and laboratory analysis for key nutrients such as nitrogen (N), phosphorus (P), and potassium (K). The laboratory analyses utilized standard methods including Kjeldahl digestion for nitrogen, Bray-P extraction for phosphorus, and flame photometry for potassium. To ensure data validity and reliability, field sampling followed standardized protocols, and spectral data preprocessing included atmospheric correction and georeferencing. Analytical techniques employed encompass regression analysis to establish correlations between spectral indices and soil nutrient concentrations, and machine learning algorithms, notably support vector regression and random forest models, to develop predictive algorithms. Model performance was evaluated through cross-validation using metrics such as R-squared, root mean square error (RMSE), and mean absolute error (MAE). Geostatistical methods, including kriging, facilitated spatial interpolation of soil nutrients, and Geographic Information System (GIS) platforms integrated the outputs into an accessible spatial monitoring system. Expected findings indicate significant correlations between normalized difference vegetation index (NDVI) and soil nitrogen levels, while spectral reflectance at specific wavelengths predict phosphorus and potassium concentrations with high accuracy (R-squared > 0.75). The developed models are anticipated to predict spatial variability of soil nutrients with RMSE below 10%. The integration of remote sensing data with ground-truth samples is projected to significantly improve the temporal and spatial resolution of soil nutrient monitoring, surpassing traditional point sampling methods. This research makes an original contribution to soil science and precision agriculture by demonstrating that satellite-based spectral data can reliably estimate soil nutrient status across diverse agricultural settings, thus facilitating real-time decision-making. It advances the theoretical understanding of the relationship between spectral reflectance and soil chemical properties, supported by the application of machine learning and geostatistical techniques in a remote sensing context. The study also provides a practical decision-support system for farmers, agronomists, and policymakers, enabling targeted nutrient management practices that improve crop yields, reduce fertilizer use, and mitigate environmental impacts. The study concludes that remote sensing can serve as a vital tool in modern soil management, with recommendations emphasizing the integration of spectral data with other proximal sensing technologies for enhanced accuracy. Future research should focus on expanding the system's applicability to different climatic regions, incorporating multisensor data fusion, and exploring temporal dynamics of soil nutrients under changing land use and climate conditions. This work underscores the need for widespread adoption of ICT-driven approaches in sustainable soil resource management, ultimately contributing to increased agricultural productivity and environmental conservation.