Kge calculation

[dancabrol]

Hi,

Thanks a lot for the project and all the related examples - they're very helpful.

I have a question about how the KGE (Kling-Gupta Efficiency) is calculated in the evaluation code.

Currently and if I'm not mistaken, the code calculates KGE by comparing all samples at each individual lead time. For example, if we have 100 samples, for lead time t=24:

• It takes 100 observed values at t=24
• It takes 100 predicted values at t=24
• Calculates KGE using these 100 pairs

I'm concerned this approach might not be ideal for evaluating hydrological forecasts because:

1. It doesn't evaluate temporal dynamics within each watershed's forecast
2. The correlation component of KGE is calculated across different forecast at a single time point, rather than across time points for a single forecast
3. We lose information about how well the model captures temporal patterns like flow peaks or recessions

Would it make more sense to calculate KGE using the temporal sequence for each watershed? For example, evaluating the first 24 hours of forecast against observations for each watershed separately?

I'd appreciate your thoughts on the rationale behind the current implementation.

Best regards
Dan