General Mills Optimized Stratified Sampling

“Soil Organic Carbon (SOC) sampling is a crucial yet costly procedure with immense applications in the field of environmental and agronomic research, management, and policy.”

An API Powered by Trailblazing Data Sources

We utilized state-of-the-art geospatial data sets obtained from NASA satellites such as LANDSAT-9 and SENTINEL-2, powered by a combination of Google Earth Engine (geemap) as a planetary-scale software for Earth science and data analysis.

Pic 1Pic 2Pic 3

In addition, we capitalized on various variables extracted from Gridded Soil Survey Geographic (gSSURGO), a high-definition national geodatabase of the United States of America. We then carefully pruned 82 features to most relevant to our cause.

Pic 4Pic 5Pic 6Pic 7Pic 8Pic 9Pic 10Pic 11

Pioneering Machine Learning Modelling

We designed and implemented a mathematically optimized Bayesian stratification and sampling algorithm, derived from unsupervised clustering techniques united with cutting-edge Conditioned Latin Hypercube Sampling (cLHS) methodology.

Pic 12Pic 13Pic 14

Harnessing the power of an advanced tree-based machine learning model, computations are rolled out and served in form of a scalable, generalizable and interpretable estimation system producing industry-standard insights (70% CI; 10% ME).