A brief update on our recent progress in downscaling the atmospheric variables. This work was performed to support the input variables for northeast regional ecosystem modeling group (NE-RESM.org).
We applied machine learning technique to downscale the GCM in reference to the Daymet variables (which represents the ground truth). Since the Daymet is only available in current scenario, our scheme will be more useful for providing the high resolution atmospheric variables for the future scenario. Moreover, since we have built the framework, this approach can be extended to continental USA. We have performed downscaling of Maximum temperature, minimum temperature and downwelling shortwave radiation. The shortwave radiation is the one that requires a lot of improvements... details to come out in a paper soon. This work was presented in Machine Learning Conference in NYC and received quite nice receptions from people who visited the poster. You can see some more pics here:
Thanks are due to Dr. Peter Thornton at Climate Change Science Institute / Environmental Sciences Division, Oak Ridge National Laboratory. I am grateful for his help in ingesting the daylength variable so that ISIMIP and Daymet could be converted to the same 24hr average,