Investigating spatial error structures in continuous raster data

摘要: The objective of this study is to investigate spatial structures error in the assessment continuous raster data. use conventional diagnostics often overlooks possible variation because such report only average or deviation between predicted and reference values. In respect, work uses a moving window (kernel) approach generate geographically weighted (GW) versions mean signed deviation, absolute root squared quantify their variations. Such computes local from data by its distance centre kernel allows map surfaces each type error. addition, GW correlation analysis values provides an alternative view These are applied two earth observation case studies. results reveal important unusual clusters can be identified through Monte Carlo permutation tests. first demonstrates fractional impervious surface area datasets generated four different models for Jakarta metropolitan area, Indonesia. where perform differently similarly, found areas under-prediction urban core, with larger errors peri-urban areas. second remotely sensed aboveground biomass Yucatan Peninsula, Mexico. mapping means compare accuracy these locally. discussion considers relative nature error, determining size issues around interpretation diagnostic measures. Investigating hidden informative descriptions