Monitor's Practice Detection Accuracy

To ensure reliable crop and practice detection, the accuracy of Monitor models is validated using a standardized process focused on the most commonly grown crops in each region. This targeted approach allows the model to be optimized for crops that have the greatest impact on agricultural planning and decision-making.

Accuracy is established by comparing the model’s predictions against ground truth data—verified information about which crops are actually growing in specific fields at a given time. This ground truth data typically comes from field surveys, farmer reports, or government records.

Rather than attempting to validate every field across a region, we focus on a representative subset of fields where reliable and accurate ground truth data is available. These selected fields serve as benchmarks for evaluating model performance. By assessing how often the model correctly identifies crop types in these known areas, we can quantify accuracy using standard metrics such as precision, recall, and F1 score, which give us an overall measure of accuracy.

This validation process is repeated in all regions where Monitor is available.

North America

Europe

France

Germany, Belgium, Netherlands

Poland

Romania

Baltics

Ukraine, Hungary

Australia 

*Due to limited ground truth datasets available to train and validate CropID models, accuracy and performance stats are based on aggregated agreement with government-reported datasets. At this time, we are not able to publish performance stats for Australia.