i am trying to interpolate the seabed sediment gravel content and i experience some problems in the cross-validation results, like the slope of the graph predicted vs. measured is not bigger than 0.6 with some hi errors. However, analyzing the layer obtained, this one seems good. I think that the problem it maybe related with the high number of zeros, 92 in 226 samples. I already transform the variable with square root, 4th root, log (x + 0.0001), logit (x + 0.0001) and normal score transformation included in ArcGIS, but the problem persists. I already make an interpolation for other variable quite similar to this using the logit, namely the fines content, and it went well. I´m asking why i could not make a good interpolation in the case of the gravel content? Maybe the large number of zeros? Or it could be related to the inherent variability to this variable vs. sampling resolution?
Someone could help me with this problem?
Thank you in advance.