Abstract
We present a first application of a fast super resolution convolutional
neural network (FSRCNN) approach for downscaling climate simulations.
Unlike other SR approaches, FSRCNN uses the same input feature
dimensions as the low resolution input. This allows it to have smaller
convolution layers, avoiding over-smoothing, and reduced computational
costs. We further adapt FSRCNN to feature additional convolution layers
after the deconvolution layer, we term FSRCNN-ESM. We use
high-resolution (0.25°) monthly averaged model output of five surface
variables over a part of North America from the US Department of
Energy’s Energy Exascale Earth System Model’s control simulation. These
high-resolution and corresponding coarsened low-resolution (1°) pairs of
images are used to train the FSRCNN-ESM and evaluate its use as a
downscaling approach. We find that FSRCNN-ESM outperforms FSRCNN and
other methods in reconstructing high resolution images producing finer
spatial scale features with better accuracy for surface temperature,
surface radiative fluxes and precipitation.