Satellite imagery alone won't tell you when the corn is ready

What it does

FarmVibes.AI is a Dockerized toolkit for building multi-modal geospatial models focused on agriculture. It ingests and aligns disparate data sources—Sentinel-1/2, NAIP, USGS elevation, NOAA weather, private weather stations, and drone imagery—then feeds them into Jupyter-based training notebooks and an inference engine. The goal is to let data scientists detect harvest dates, estimate soil carbon, or forecast micro-climates without hand-rolling raster pipelines.

The interesting bit

Instead of training on isolated pixels, the framework expands the input context across modalities, space, and time. A workflow might fuse optical imagery with radar and elevation for a single location, then layer in historical observations from nearby fields. The whole data prep pipeline is expressed as a configurable directed acyclic graph of operators that output fused tensors or pandas arrays ready for model consumption.

Key highlights

• Ships with pre-built notebooks for harvest detection, carbon simulation via the COMET-Farm API, crop segmentation using the SpaceEye cloud-removal model, and micro-climate forecasting with DeepMC.
• Ingests public and private data sources—Sentinel, NAIP, NOAA, Ambient Weather, plus custom raster inputs—and keeps all generated data local.
• Workflows can be set to re-run inference automatically when upstream data (e.g., new satellite imagery) is updated.
• Exposes a local REST API and Python client for running the DAG-based workflows from Jupyter or other tooling.

Caveats

• The open-source release is strictly a local cluster; all data persists on your machine.
• Custom physical weather sensor integration is flagged as a future addition, not yet implemented.

Verdict

Worth a look if you work at the intersection of remote sensing and agronomy and want a structured starting point rather than a from-scratch raster pipeline. Purely general-purpose geospatial ML users may find the agriculture-specific tuning and local-only deployment limiting.