Zimbabwe Weather Intelligence: An Evidence-Based Climate Research and One-Hour-Ahead Temperature Prediction System

Climate information plays an important role in understanding weather variability, assessing climate-related risks, and supporting informed decisions across climate-sensitive sectors such as agriculture, water management, and disaster preparedness. As part of the Zimbabwe Weather Intelligence initiative, MacroEdtech has developed an extended research framework integrating Generative Artificial Intelligence (GenAI) with machine-learning-based weather prediction. The project aims to make climate-related knowledge more accessible through natural-language interactions while providing near-term temperature prediction capabilities for four major Zimbabwean cities: Harare, Bulawayo, Mutare, and Gweru.The system combines a Generative AI-powered Climate Research Assistant with a one-hour-ahead temperature prediction interface. 

The research assistant uses Retrieval-Augmented Generation (RAG), FAISS vector search, LangChain, LangGraph, and Model Context Protocol (MCP) to retrieve relevant information from climate-related documents and generate evidence-based responses. Its knowledge base contains 217 indexed text chunks. The weather prediction component utilises a historical dataset of 385,728 records collected from Open-Meteo covering 2015–2025. An XGBoost-based prediction model achieved a reported Mean Absolute Error (MAE) of 0.3448. The application was initially implemented locally using Ollama and subsequently adapted for cloud-based access using the Gemini API, demonstrating different approaches to deploying AI-powered climate intelligence systems.

This research demonstrates how Generative AI, machine learning, information retrieval, and weather data analytics can be integrated into a practical climate intelligence platform. It also provides valuable insights into system architecture, model evaluation, cloud deployment, API integration, and the limitations associated with document coverage and external API availability. The project contributes to the broader exploration of AI-enabled climate research and accessible weather intelligence, particularly for regions where climate information is distributed across multiple sources.

This work is completed by Rosemary, Generative AI Research Assistant at MacroEdtech.

Github : https://github.com/roseTadiwa/zimbabwe-weather-intelligence-MACROEDTECH/tree/master