Interpolating missing land cover data using stochastic spatial random forests for improved change detection

Forest cover requires large scale and frequent monitoring as an indicator of biodiversity progress towards United Nations World Bank Sustainable Development Goal 15. Measuring change in forest over time is essential task order to track preserve quality habitats for species around the world. Due prohibitive expense impracticality mass field data collection monitor at regular intervals, satellite images are a key source globally. A challenge working with missing due clouds. Existing methods interpolating based on past images, such compositing, effective stable land but can be inaccurate dynamic substantially changing landscapes. Here we present adaptation our recent stochastic spatial random (SS-RF) method, which combines observed from prior image modelled estimates current produce interpolated values associated probabilities those values. Results show SS-RF method accurately detected simulated under both clear felling (0.83 average overall accuracy) tree thinning (0.85 accuracy). Our more than offering 39% 12% increases accuracy simulations respectively. However, when natural fluctuation occurs there minimal cover, compositing has equivalent or accurate performance method. Overall find that produces range clearing scenarios robust modelling noticeably