A number of studies have been conducted by using machine learning approaches to simulate carbon fluxes. We performed a meta-analysis of these net ecosystem exchange (NEE) simulations. Random forests and support vector machines performed better than other algorithms. Models with larger timescales had a lower accuracy. For different plant functional types (PFTs), there were significant differences in the predictors used and their effects on model accuracy.

Detailed climate data are needed to assess the impact of climate change on human and natural systems. The performance of two high-resolution regional climate models, ALARO-0 and REMO2015, was investigated over central Asia, a vulnerable region where detailed climate information is scarce. In certain subregions the produced climate data are suitable for impact studies, but bias adjustment is required for subregions where significant biases have been identified.

Some river basins are considered to be very similar because they have a similar background such as a transboundary, facing threats of human activities. But we still lack understanding of differences under their general similarities. Therefore, we proposed a framework based on a Bayesian network to group watersheds based on similarity levels and compare the causal and systematic differences within the group. We applied it to the Amu and Syr Darya River basin and discussed its universality.