Iterative Mapping of Probabilities (IMP): A data fusion framework for generating accurate land cover maps that match area statistics

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 propose a framework that uses class probabilities predicted by machine learning to guarantee that the mapped proportion of each class matches independent area estimates. We demonstrate the use of the framework and map the 8 primary LUCAS land cover classes in multiple European countries using CORINE training data. We validate the baseline highest likelihood class maps and the proportional class maps output by the proposed algorithm with LUCAS land cover observations and S2GLC validation points. Our results show that the framework and algorithms create maps that match area estimates, and that may also be more accurate than the baseline map. We also found that the algorithm does not need probabilities predicted by models that were trained on data with representative class proportions.