CWatM model simulation can be set to multi-process and speed up the simulation?

[Defeng-Wu]

Discussed in https://github.com/iiasa/CWatM/discussions/9

^{Originally posted by Zhyu-1 December 12, 2020} Hi Mikhail and everyone, My name is Defeng Wu, I am a PhD candidate with China Agricultural University (CAU). I have been learning the CWatM model for half a year and it looks great. Now，I can simulate the hydrological process of the whole China from 1979 to 2018, but it takes almost one and a half hours to run for a year. CWatM model simulation can be set to multi-process and speed up the simulation? Can you help and tell me how to speed up the simulation?

[mikhailsmilovic]

Guten Tag @Defeng-Wu! Awesome to hear about the China model 🌏 For a China model at 5 arcminutes (~10 km), this is ~96,000 cells. The settings NoRoutingSteps is the number of sub-daily steps and can be reduced for faster runs, especially for setting up and testing.

Loose recommendations: At 30 arcminutes, NoRoutingSteps = 10 At 5 arcminutes, NoRoutingSteps = 24

Sub-daily steps are used in the modules for routing, soil, and lakes & reservoirs. Let us assume water with a speed of ~1 m/s = ~86.4 km/day. At ~10km resolution, discharge leaving a cell needs ~8 substeps to reach a cell ~86.4 km away (say, 8 cells away). At ~1 m/s with 24 substeps, at least 66% of the water leaving cells has the "substep time" to reach cells 8 cells away.

Reducing the number of routing steps may mean capturing the high flows less well. It can be interesting to test different NoRoutingSteps for a wetter period in your model.

For multi-processing, many parts of the model depend on previous modules. For calibration, we use a tool that runs several setups of the model in parallel.

[Defeng-Wu]

Thank you very much for this helpful details. Then i will try, and look forward to the good news.