Baseline choice and performance implications for REDD