Using calibrated EBK model for forward modeling (no input features)

af2k24 — Tue, 11 Jun 2024 20:36:16 GMT

I am wanting to use Empirical Bayesian Kriging for forward modeling. Since EBK is a geostatistical model that is calibrated based on data from input features, my thinking is that it should also be possible to create a prediction surface using different explanatory raster layers than the ones initially used to calibrate the model. In my case, I have feature data from climate stations and explanatory rasters that all represent a historical climate. I used this historical data to calibrate an EBK model and create a prediction surface. Now, with the calibrated model, is it possible to provide the same explanatory raster variables, but for future time periods, in order to create a second prediction surface? In this forward modeling scenario, no input features can be provided (as climate station data from the future doesn't exist, of course). I suppose this would be possible to implement from scratch, but I'm hoping there is a tricky way to do it, perhaps by editing .xml files, or the like, and running the EBK ArcPy function.

I have a screenshot that shows my calibration inputs. But, the geostatistical wizard always asks for input features. To further illustrate, I am wanting to create a second surface without providing an Input Dataset, and using Explanatory Variables called "accm_2070_2099_1.tif", "tmax_2070_2099_1.tif", and "srad_2070_2099_1.tif".

Re: Using calibrated EBK model for forward modeling (no input features)

EricKrause — Wed, 12 Jun 2024 15:41:03 GMT

Hi @af2k24,

I don't believe there is any way to do this. If I'm understanding correctly, you're wanting to use the coefficients from the original model and use them for the new rasters. However, EBK Regression Prediction will rebuild the coefficients for any new rasters, estimated from the input features that you provide; you can't save the coefficients are reuse them, unfortunately. The only thing that comes to mind is trying to forecast the input point values to 2070-2099 and use them along with the forecasted rasters to get a prediction surface for 2070-2099. Though that is obviously easier said than done, especially if you do not have historical point data to build a forecast model.

-Eric

topic Re: Using calibrated EBK model for forward modeling (no input features) in ArcGIS GeoStatistical Analyst Questions

Using calibrated EBK model for forward modeling (no input features)

Re: Using calibrated EBK model for forward modeling (no input features)