microsoft/farmvibes-ai

What it solves

FarmVibes.AI provides a framework for developing geospatial machine learning models specifically for agriculture and sustainability. It addresses the difficulty of obtaining deep insights—such as carbon footprint estimation and crop growth rates—when using single geospatial datasets in isolation by enabling the fusion of multiple data sources.

How it works

The system operates through three primary components:

1. Dataset Preparation: It uses a configurable workflow (defined as a directed acyclic graph) to ingest and pre-process data from various sources, including Sentinel 1 and 2 satellite imagery, NOAA weather data, USGS elevation maps, and drone imagery.
2. Model Sample Notebook Library: A collection of notebooks allows users to tune existing models for specific regions or seasons, covering tasks like harvest date detection and micro-climate prediction.
3. Inference Engine: This component combines data connectors and pre-processing steps with tuned models to perform inference over specific areas of interest and time ranges, updating results as new data becomes available.

Who it’s for

Data scientists and researchers focusing on remote sensing, earth observation, and agricultural sustainability who need to build robust, multi-modal geospatial models.

Highlights

• Multi-modal Fusion: Combines RGB, SAR, multispectral satellite imagery, weather data, and elevation maps.
• Configurable Workflows: Uses DAG-based workflows to manage data downloading and preparation.
• Extensible Data Ingestion: Supports a wide range of public datasets (Sentinel, NAIP, NOAA) and custom rasterized data.
• Local Cluster Support: Runs as a local cluster using Docker images, interactable via REST API or a Python client.