An AI visual chatbot system for climate heat risk communication and decision-making in at-risk agricultural communities

Information on heat-related health risks that is based on meteorological data and forecasts is important for outdoor workers, farm managers, and many other stakeholders, but visualization and interpretation of climate data in a user-friendly way remains a challenge. This proof-of-concept research developed Sol, an artificial intelligence (AI) chatbot to conveniently interpret and communicate heat stress risks through smartphones and other devices. Sol is integrated into iHeatApp, a wet bulb globe temperature (WBGT) forecast visualization tool, to turn complex climate data into practical guidance with bilingual output in English and Spanish. The chatbot leverages an open source large language model (LLM) constrained by prompt engineering designs to reduce hallucinations and encourage domain-specific responses. The chatbot input may be either text or screenshot of the app's interface. Guided by our tutorial materials, we gathered farmworker feedback on the chatbot's ability to translate complex weather metrics into risk mitigation actions using focus groups of farmworkers from the Imperial Valley, California. When WBGT forecast values reach the “danger” category, the chatbot recommends wearing light, breathable clothing and increasing water breaks to prevent heat stress. Our app can help both farmworkers and their managers adapt work scheduling decisions to avoid the risks of peak heat and other health hazards and to mitigate labor productivity loss.