Trustworthy AI knowledge systems for precision agriculture and agribusiness management: Detecting crop health using satellite imagery

Precision agriculture urgently requires trustworthy AI knowledge systems to translate complex remote sensing data into actionable insights for crop health management. This study develops a robust framework for detecting critical crop stress indicators—specifically water stress and nutrient deficiency—in seasonal rice and mustard plantations within Majuli, Assam, India a region highly vulnerable to flooding. Leveraging high-resolution LANDSAT 8/9 OLI/TIRS satellite data, we compute essential vegetation (NDVI) and land surface temperature (LST) indices. These multi-modal inputs are processed using a pre-trained U-Net++ deep learning architecture for precise pixel-level segmentation of stressed crop areas. Evaluation demonstrates exceptional performance, with segmentation accuracy reaching 96.1 % and an F1-score of 94.5 %, confirming the model's high reliability. Beyond technical accuracy, this work explicitly addresses the trustworthy AI Knowledge for precision agriculture through transparent data sourcing by standardized indices, model explainability by using established U-Net++ architectures, and validation against known flood-impact scenarios. Organizational integration is facilitated by generating diagnostic knowledge directly usable by farmers and agricultural extension services in vulnerable ecosystems like Majuli, enabling timely interventions for improved resilience and yield. This approach exemplifies the development of trustworthy AI knowledge systems for sustainable agricultural decision-making.