Open jzwart opened 3 years ago
Leaving flow and temperature weights all at 1 reduced RMSE for flow (6.1 RMSE compared to 6.8 for uncal), but inflated RMSE for temperature (~40 RMSE).
Currently trying out adjusting weight by log normalizing between 0 and 1. This gives more weight to low values (temperature values and low flow). See log normalizing (red) vs. linear (black) below
zoomed in x-axis
Log normalization still biases flow. Setting temp weight to 1 and flow weights to 0.05 still biases calibration towards flow. Probably best to calibrate flow and temp separately for now.
Leaving the weights of the different observation types (e.g. temperature obs, flow obs) causes bias towards high value observations (or residuals) in objective function. According to PESTPP4.2.4 manual