https://twitter.com/Hydrology_WSL/status/1081816260558622720

]]>Recently I encountered an issue with the KGE when I tried to assess the quality of modelling lake levels. I had already seen in some graphs that the model showed a substantial bias. However, the Beta (bias) component of the KGE still came up as 1.01, 0.99, for even the worst periods. It took me some time to realize that this is because lake levels are on an interval scale (no true zero); The ratio between modelled mean levels of 419 or 418 m.A.S.L. is close to one, but the bias of 1m is large if the dynamic range is only ~5m or so.

Reflecting on the physical meaning of the bias term as an estimation of the water balance, I now like how beta is implemented in the python package spotpy as sum(simulation)/sum(observation) instead of means. Comparing the means of two series of water levels makes sense, but comparing the sum of those series does not.

Clearly, I would have been less surprised if I had read your technical note directly after this post came out đ

]]>Dear Andy, thank you for sharing your thoughts. I understand that the use of transformations such as the logarithm or the inverse of flows can be aggressive and constrain a lot the models. However, in my experience, it seems that models that are calibrated using the log or the inverted flows in the objective function are better to reproduce the lowest part of the flow duration curve than models calibrated using the square root (maybe for the wrong reason, I don’t know…). Regarding the unbounded aspect of the KGE and KGE’, I agree that it can represent an issue for parameter calibration. I can add, that one of my colleagues observed that it is also an issue when applying classical sensitivity analysis techniques (i.e. Sobol or Morris).

]]>You can use these tools for calculating of drought indices:

https://agrimetsoft.com/MDM.aspx

https://agrimetsoft.com/KBDIS.aspx

https://agrimetsoft.com/RDIT.aspx

https://agrimetsoft.com/drought%20monitor.aspx

Cheers

]]>