A multi-agent model integrating Bayesian deep learning for drought monitoring using multi-source remote sensing data: application in STEM education

The rapid development of artificial intelligence, big data, and cloud computing is creating new opportunities for interdisciplinary, experiential, and problem-oriented innovation in STEM education. This study proposes a multi-agent model integrating Bayesian deep learning on the Google Earth Engine platform to support STEM teaching and learning through drought monitoring using multi-source remote sensing data. The model integrates meteorological conditions through the Standardized Precipitation Index and Standardized Precipitation Evapotranspiration Index, vegetation health through the Vegetation Health Index, and soil moisture data. Specialized agents perform data acquisition, preprocessing, index calculation, drought-state analysis, and probability fusion from multiple information sources using Bayesian inference to provide an integrated assessment of drought severity. The model is also designed as an intelligent learning environment that enables learners to engage with contemporary concepts in artificial intelligence, multi-agent systems, Bayesian deep learning, geospatial data analysis, and decision-making under uncertainty. Bayesian deep learning allows the model not only to predict drought conditions but also to quantify the reliability of its results, thereby improving AI explainability. By utilizing large-scale remote sensing data on Google Earth Engine, the study contributes a new approach to STEM education that connects artificial intelligence, data science, remote sensing, and sustainable development challenges in higher education.