@Scott,
EBK or EBK regression prediction typically outperforms other interpolation techniques, may be, apart from some other ML techniques. However, EBK and EBKRP can't work here using ArcGIS since I guess the minimum number of samples it can handle is 20. If you have covariates with noise, you can try co-kriging since. Adding covariates may help improve the prediction. If not, advise from SteveLynch is most appropriate. Try as many as may appear reasonable, than compare the error stats.