Remote-sensing derived landcover products, such as those provided by MODIS, are valuable tools for a wide variety of analytic or predictive applications. These products may include a wide range of landcover classifications, including anthropogenic classes such as cropland and urban development. However, certain applications may not be interested in these anthropogenic classes and may instead prefer to use the corresponding natural landcover class that would be present if a given area was never subjected to human development. In this study, we aim to use classification algorithms to reclassify human-developed landcover classes to the natural landcover class that most closely matches the site's climatic character in order to control for the effect of human development. We develop and compare Random Forest and Deep Neural Network algorithms to classify landcover class by annual climate variables, then reclassify human developed land cover classes to natural land cover classes. We found similar overall performance between Random Forest and DNN models, with overall predictive accuracy of 62.6% and 64.2% respectively. Predictive accuracy of particular landcover classes varied significantly within and between models, with some classes predicted well by one model but not the other.