---

AgroMetLLM: Evapotranspiration & Agro-Advisory System

---

Overview

This project provides a AgroMetLLM: web-based agrometeorological advisory tool that combines weather data from Open-Meteo, advanced evapotranspiration (ET) calculations, and localized, farmer-friendly advice generated by local large language models (LLMs) running on a Raspberry Pi (or any compatible Linux host) via Ollama on Raspberry Pi 4B an edge-IoT device. Gradio powers the user interface, making it easy for rural extension agents, scientists, and growers to obtain actionable, context-aware advice for their fields.

---

Features

• Location Search & Selection: Search by district/city/village name (global coverage via Open-Meteo Geocoding API).

• Multi-Model LLM Integration: Switch between multiple local LLMs (e.g., TinyLlama, Qwen, etc.) hosted via Ollama; models are managed in memory for optimal performance.

• Custom Forecast Window: Choose a forecast period from 3 to 7 days.

• ET Model Comparison: Calculates and visualizes ET using multiple scientific methods: FAO ET₀, Hargreaves-Samani, Turc, Priestley-Taylor, and Makkink.

• Intuitive, Rural-Friendly Advisory: Local LLM generates simple, pointwise farm advice with examples, covering:

  1. Irrigation plan
  2. Crop/soil action
  3. Livestock/labour management
  4. Pest & disease watch
  5. Input-saving tip

• Data Logging: All results are logged in a local SQLite database for traceability and research.

• Interactive Outputs:

  • ET metrics table
  • Comparative bar chart
  • Model information
  • Farmer advisory (LLM output)

---

Requirements

• Hardware:

  • Raspberry Pi 4B (recommended) or compatible Linux system
  • Sufficient RAM for LLMs (ideally >4GB for sub-1B models)

• Software:

  • Python 3.8+
  • Ollama (for local LLM serving)
  • Gradio
  • pandas, numpy, matplotlib, requests, sqlite3

Install requirements:

requirements.txt:

requests
pandas
gradio
matplotlib
numpy

pip install -r requirements.txt

# Ollama: follow [Ollama install docs](https://ollama.com/)

---

Usage

1. Start Ollama and Load LLMs Make sure your desired LLMs are available in Ollama as per https://github.com/ollama/ollama

2. Run the Application

   python agromet_advisory.py

   The Gradio UI will start at: http://localhost:7860

3. How to Use:

   • Select the LLM model (from dropdown)
   • Enter your location name (city, village, district, etc.)
   • Set forecast days (3–7)
   • Click "Search Location" and select the desired location if multiple
   • Click "Generate ET Forecast & Local LLM Advisory"
   • Review the outputs: ET table, ET chart, model info, and actionable farm advice

---

Customization

• Add/Remove LLM Models: Pull new models via Ollama and restart the app.
• Change Advisory Prompt: Edit the ollama_advice() function to tune few-shot examples or language for your region.
• Database: The app stores all runs in agromet_et.sqlite—analyze or export data as needed.

---

Troubleshooting

• If you see errors about models not loading, ensure Ollama is running and models are pulled.
• For slow responses, use smaller LLMs (<1B parameters) on Raspberry Pi.
• Weather or location not found? Check for typos or try a nearby city/village.

---

Acknowledgements

• Open-Meteo API for global weather and geocoding
• Ollama for local LLM inference
• Gradio for rapid web UI prototyping

---

License

Apache-2.0 License (see LICENSE for details)

---