Predicting Crop Yields from Satellite imagery

Using Machine Learning, we developed a method to predict sugar cane crop yields in Brazil for the 13 different processing mills days before the bi-weekly official crop yield report.

Project Objectives:

    / [pdf]

What/How:

Below see how we are analyzing the outputs – the NDVI shows the diff we see between multiple dates – which can indicate crop harvest. To get this, we have to align images, do the NDVI calculation (non-trivial), determine the area, and then use historical crop yield actuals to determine what we think this area represents in terms of yield AND THEN build a model so we can look at other harvested areas and predict what the yield is like to be. Here we are looking at < 20 square miles – but we are doing this on the order of 13,000 square km. Trickiest part? Cloud coverage.

Example output:

    / [pdf]

OneShare - Send Screenshots/Links etc from Phone to your Email/Laptop in 1 tap.

Ramble