← BACK
MHacks 2025 · NASA SMAP

Soil Moisture Prediction

Predicting soil moisture for Southeast Michigan farmland from satellite observations and above-ground features — built solo in a weekend.

0.84
R² (validation)
0.02
RMSE cm³/cm³
RF
Random Forest
SMAP
+ STF-SSM data

SoilMoisturePrediction was my first solo hackathon project, built at MHacks 2025 to merge my two fields — environmental engineering and computer science — into something I had the domain knowledge to ship. The core question: how well can above-ground features like precipitation and humidity, paired with satellite data, predict what is happening in the soil below?

The model ingests NASA SMAP satellite soil-moisture retrievals and the STF-SSMspatiotemporal fusion dataset, blended with local conditions across Southeast Michigan farmland. Most of the work was preprocessing — aligning resolutions, filling gaps, and engineering features — before training a Random Forest regressor.

On held-out validation it reached R² = 0.84 with an RMSE of 0.02 cm³/cm³, accurate enough to be useful for irrigation-scheduling decisions.

The soil-moisture signal feeds into BLUELab Smart Irrigation, complementing in-field sensors so the irrigation controller has a satellite-scale view of moisture trends alongside ground-truth readings.

Pythonscikit-learnNASA SMAPSTF-SSMRandom ForestRemote SensingETL
Acknowledgments
Design and feedback from Artem Tikhonov— NASA researcher and MHacks judge.