Application of Model-based Geostatistics for Natural Hazards Identification

Annually much agricultural productions are lost by natural hazards such as flooding, extreme temperature, and drought. In this research we present a procedure to identify such hazards for rice cropping from remote sensing imagery. Extreme temperature is investigated by comparing actual temperatures with tolerance temperatures during the growing stage. Drought is investigated using the evapotranspiration ratio (ETa/ETp) and comparing it with the allowable range. Flooding is studied by finding sudden drops in NDVI values after an extreme rain fall. The procedures are applied to a rice growing area in Iran, 2006. The results, however, did not show any of these hazards to have occurred in the area during the study period. To study spatial effects of hazard, we next created a yield map using model-based geostatistics. We compared ordinary kriging (OK) with external drift kriging (KED). Smaller prediction errors exist when using KED with NDVI and Distance from city as external drift. Cross validation indicates that the straightforward OK that ignores the information from secondary data yields the largest prediction errors (1.36 compare to 0.061 for KED). The results suggest using spatially intensive ancillary data leads to increasing the accuracy of prediction of yield map.