masked results in smileRandomForest

Question

I used the deprecated randomForest function to classify land use using about 60,000 training points in a ee.FeatureClass with data from about 130 ee.Image bands. Somehow the old classifier always gave 'wall to wall' predictions, where in spite of missing data (e.g., cloudy periods in the Landsat record), the classifier still produced a result. I'm trying to reproduce those predictions using smileRandomForest, but much of the classification result is masked, especially during periods with fewer Landsat satellites in operation. Is there some explanation for why this is the case?

Answer 1

I used the .unmask() function on my input bands, allowing smileRandomForest to create 'wall to wall' predictions that are quite good and do well to reproduce what I found using the deprecated randomForest function.

classified_img = annual_stack.unmask().classify(trained_model)