Any solution for "Optimal parameters not found"?

[rivagao]

Hello, I am trying to calculate the range value as in fit_model.percentile_scale(0.95) for a fractional cover map in a 2D raster image.
the fraction cover used in this application is the bare ground cover in percentage. each calculation includes 40 by 40 cells. I am trying to determine the bare ground size of these 40 by 40 images.

the question is I am not sure how to determine what covariance model should be used here, of all the models I tried out, the Linear model gave me the best results, however, it could only make sense in dim=1. I have also tried exponential. stable and spherical models, but these models usually show an error message: Optimal parameters not found: the maximum number of function evaluations is exceeded. result in a nodata output that is not optimal.

in addition, I have tried the method to determine which model fits the experimental variogram the best using the r square. but different cells have a different best fit. the problem is that I need to determine one best-fit model for all 300,000 calculations.

any suggestions or feedback on this issue would be very appreciated.

[MuellerSeb]

Hey @rivagao ,

For your first problem: you can increase the maximal number of function evaluations in the fitting routine by setting max_eval:

model.fit_variogram(bin_center, gamma, max_eval=1000)

By default, this value is 100 * n where n is the number of parameters. See the documentation

Maybe this is a hint, that we need to increase the default value here.

For your second problem: This is hard and there is no standard way. you could try to accumulate the r2 scores for each model for all cells and then select the one with the highest overall score.

Hope that helps!

Cheers, Sebastian