Performance improvements to modern hydrological models via lookup table optimizations

Abstract

Distributed hydrological models predict the spatial variability in processes that govern observed mass and energy fluxes. A challenge associated with the use of these models is the computational burden associated with representing the Earth's (sub)surface via millions of computational elements. This burden is exacerbated as more complex process representations are included because their parameterizations involve computationally intensive mathematical functions. Lookup tables (LUTs) approximate a mathematical function by interpolating precomputed values of the function. Highly accurate approximations are possible for substantially reduced computational costs. In this work, a general methodology using the C++ LUT library FunC is applied to identify and replace computationally intensive mathematical function evaluations in the Canadian Hydrological Model (CHM). The use of LUTs introduces a pointwise relative error below 10−8 and provides a reduction in run time of almost 20%. This work shows how LUTs can be implemented with relatively little pain and yield significant computational savings for distributed hydrological models.