MARRMoT
MARRMoT copied to clipboard
wknoben
Low computing efficiency
In R, Xinanjiang model calibration only cost about 40s.
But in MARRMoT, the calibration is quite low efficiency. With the same forings, half hour passed, still no output.
Just wondering which part delays the computing efficiency? ODE or optimization function? Any idea?
Currently we are not looking to redevelop MARRMoT in a new language. We are aware that it is not the most computationally efficient tool and we refer to the discussion section (Sect. 5.3.2) in Knoben et al. (2019) for a brief discussion on this.
I will leave this issue open, in case someone decides to pick-up the challenge of further speeding up MARRMoT by translating this in a different language. Thanks for the suggestion.
Reference: Knoben, W.J.M., Freer, J.E., Fowler, K.J.A., Peel, M.C., Woods, R.A., 2019. Modular Assessment of Rainfall-Runoff Models Toolbox (MARRMoT) v1.2: an open-source, extendable framework providing implementations of 46 conceptual hydrologic models as continuous state-space formulations. Geoscientific Model Development 12, 2463–2480. https://doi.org/10.5194/gmd-12-2463-2019
Dear @ltrotter ,
Thanks for your response. I have finished translating Xinanjiang model to Julia.
The Julia version costs 4-5s in the calibration mode, (in comparison 40s in Rcpp, 30 mins in MATLAB). The carefully designed Julia MARRMoT is about 8 times faster than the cpp (i.e., 5s vs. 40s), and 300 times faster than MATLAB MARRMoT. https://mp.weixin.qq.com/s/Ah8mjXBUhtpIZbvxicnDFA
In the Julia version, all module are same as the MATLAB MARRMoT, except for ODE. I closed the ODE, as ODE is the speed killer. When closed ODE, MARRMoT may be 30 times faster, meanwhile the KGE only decrease a bit (about 0.05).
Regards Dongdong
References
Rcpp Xinanjiang, https://github.com/rpkgs/VIC5/blob/master/src/XAJ.cpp, https://github.com/CUG-hydro/hydroTools/blob/master/R/calib_XAJ.R#L55
@ltrotter What do you think the strategy of cloing ODE in every simulation step's water balance solving?