Spatial Random Forest (S-RF): A random forest approach for spatially interpolating missing land-cover data with multiple classes

Land-cover maps are important tools for monitoring large-scale environmental change and can be regularly updated using free satellite imagery data. A key challenge with constructing these is missing data in the images on which they based. To address this challenge, we created a Spatial Random Forest (S-RF) model that accurately interpolate based modest training set of observed image interest. We demonstrate approach effective only minimal number spatial covariates, namely latitude longitude. The motivation longitude our covariates available all whether or due to cloud cover. S-RF flexibly partition provide accurate estimates, easy incorporation additional improve estimation if available. effectiveness has been previously demonstrated prediction two land-cover classes an Australian case study. In paper, extend method more than classes. performance at interpolating multiple classes, study drawn from South America. results show best predicting three compared 5 10 other information needed as grows, particularly unbalanced. explore issues through sensitivity analysis: influence amount development performance. cover influential also found increasing beyond 100,000 observations had impact accuracy. Hence, relatively small required model, beneficial computation time.