Open Earth Monitor — Global Workshop 2024

Providing land cover estimates with both correct pixel-level class predictions and regional class area estimates is important for many monitoring and accounting purposes but rarely achieved by current land monitoring efforts. We present a framework that uses class probabilities predicted by machine learning to guarantee that the mapped proportion of each class matches independent area estimates.

IMP navigates the bias of area estimates and land cover models and uses this to make classifications that not only match area estimates, but are also more accurate than maps created with highest likelihood classification. This is especially the case for large general-purpose models trained on data whose class proportions are not representative of the mapped area, which means that this algorithm can be used to localize such models for more accurate mapping of individual countries.

We will discuss the workings and benefits of this methods, as well as its potential for further uses, such as validating area estimates themselves, or using it directly for land cover area estimation through time and space without the need for additional sampling.