A conservative adaptive wavelet method for the shallow water equations on the sphere

We introduce an innovative wavelet-based approach to dynamically adjust the local grid resolution to maintain a uniform specified error tolerance. Extending the work of Dubos and Kevlahan (2013), a wavelet multi-scale approximation is used to make dynamically adaptive the TRiSK model (Ringler et al. 2010) for the rotating shallow water equations on the sphere. This paper focuses on the challenges encountered when extending the adaptive wavelet method to the sphere and ensuring an efficient parallel implementation using MPI. The wavelet method is implemented in Fortran95 with an emphasis on computational efficiency and scales well up to O(10^2) processors for load-unbalanced scenarios and up to at least O(10^3) processors for load-balanced scenarios. The method is verified using standard smooth test cases (Williamson et al. 1992) and a nonlinear test case proposed by (Galewsky te al. 2004). The dynamical grid adaption provides compression ratios of up to 50 times in a challenging homogenous turbulence test case. The adaptive code is about three times slower per active grid point than the equivalent non-adaptive TRiSK code and about four times slower per active grid point than an equivalent spectral code. This computationally efficient adaptive dynamical core could serve as the foundation on which to build a complete climate or weather model.