So here comes some nice parts of working with Linux. I am trying to figure out running several WRF simulations in parallel (I am not talking about a MPI run of WRF) on the school cluster. From the simple run experience, we just need to compile the code, run the model. But what if I want to have multiple runs at the same time? I do not want to go into the “building-running-deleting” cycle, and I would like to figure out how to have several sets of namelists along with only one wps.exe/wrf.exe to do the simulation. This would make it handy when we need to recompile wrf.exe or wps.exe. I am writing down what I have tested, but there is no guarantee that this is correct. Continue reading
CUDA is an exciting tool for accelerating current models. Here I am applying it to a simple water balance model.
What a water balance model does can be simply described by pouring water into/extracting water out from a water bucket. Here “pouring water into the bucket” can be in the form of precipitation, or irrigation, etc. “extracting water out from bucket” can happen in the forms of evaporation, transpiration, runoff process, etc. This makes the calculation framework ideal at CUDA core scale, assuming that I am now involving too complicated presentation of these processes. It is notable that modern hydrological models do contain sophisticated math presentation of these processes, but for now I am not bothering my CUDA threads with them.