In the last couple of month, I spend most of my research time calibrating my model. Calibration is very time consuming, but still does not lead directly to publications. So what does calibration mean, why I’m calibrating my model and what for?
First of all, there are different types of models in hydrology, more physical based and more -let’s say- statistical models. On one side the perfect physical model would not need to be calibrated, since it includes all physical processes. But up to now, and probably not even in the future there will be the one and only perfect model, which is useable for every catchment around the world. Since we don’t have such a model we need to tune our models a bit, so that the model output comes as close to reality as possible. Statistical models on the other side do not include physical processes, they just concentrate on the model output and rely on calibration. And of course there are many models in between statistical and physical approaches.
It is also important to mention that often even quite physical based models are becoming more statistic because of data scarcity. So all models need to be calibrated to tune some model parameter to get a better and more realistic model performance. In hydrology we often like to calibrate our models to discharge. So we look how much water is running in the river in reality and try to get the model to reproduce the results. But with a physical based model we can also check for other parameters. Does, for example, the soil moisture show a realistic behavior? Are the runoff processes realistic for my specific catchment? How does the actual runoff curve look like? Or is only the mean value approximated correctly? Questions like this you have to keep in mind while calibrating a hydrological model. So step by step and sometimes with the help of algorithms you balance all the different hydrological processes in the model while turning different screws. Since there are always plenty of possibilities to tune one screw it is quite hard to decide when the model is properly calibrated. And then, when you are so confident that you now have a more or less good calibrated model, which took so much of your time and energy, you are just at the starting point for your research…
You have a calibrated model, great! But in your paper all this long, long work will only be a little, little part, just a few sentences. It is just that you need a calibrated model before you start to do something with it. But when you think you are at an end with all your thousands model runs, you will see while you are validating the model, that still something is not as good as hoped, so you might start again to run more and more model simulations to get a better performance and have an even more realistic model setup.
So sorry for only this few thoughts, but I have to start a new calibration run with my WaSiM model (Schulla, 2013) setup for the Raab catchment and the shuffled complexity evolution algorithm (Duan etal., 1992) ;-).
Duan, Q., Sorooshian, S., Gupta, H. V., Gupta, V. (1992). Effective and efficient global optimization for conceptual rainfall-runoff models. Water Resources Research, 28(4), 1015–1031. http://doi.org/10.1029/91WR02985.
Schulla, J. (2013). Model Description WaSiM, Technical Report, pp. 324 (www.wasim.ch).