Global forest soil respiration and changes with 30 m resolution (2000-2020)

The datasets provided global forest soil respiration (Rs) and change with 30 m resolution, which were estimated by artificial neural network (ANN) models that were evaluated with a 10-fold cross-validation scheme . Global forest change datasets (V1.7; treecover2000, gain, and lossyear) were used to accurately define forest extent and changes from 2000 to 2019 because of their dynamics information and high resolution (30 m). The optimal ANN models reported 0.75 of r2 and 208.4 g C m−2 yr−1 of RMSE, The spatial patterns of global forest Rs maps showed fine detail (30 m resolution). Both analyzing uncertainty of estimates and comparing with other global Rs researches indicated that the presented datasets of global Rs and changes from 2000 to 2020 were reliable, which can provide an accurate benchmark for discussing carbon cycle and climate change in global-to-regional scales, even to a small area of forest (i.e. dozens ha), which used to be ignored in other global Rs researches.