Satellite estimation of forest carbon using regression models

Periodic monitoring of forest carbon is important, since forest cover is changing rapidly in many parts of the world, and becomes a major source of terrestrial carbon emission that may be one of the main drivers of global climate change. Regression is often used to estimate forest variables (including carbon) using satellite sensor data though a low coefficient of determination (R 2) is apparent and this research was designed to investigate both traditional and alternate regression approaches to increase the magnitude of R 2. The study area was located in southeastern Bangladesh. Data from Landsat Enhanced Thematic Mapper Plus (ETM+) and ground‐based forest survey were used. This research explored the use of dummy variables in regression models to increase R 2, while the dummies were set from the optimal stratification of forestland. The finding will heighten the accuracy of forest attribute estimation and help to understand terrestrial carbon dynamics and global climate change.