|
 |
There are kriging methods that are more powerful interpolators than ordinary kriging; however, they were not utilized in the current work due to the number of assumptions required. Ordinary kriging was applied because it requires the least assumptions about the underlying structure of the data. Interpolation performance could be potentially improved with the use of stronger geostatistical techniques. This would require more extensive pre-modeling exploration of the data to determine a better understanding of the underlying characteristics of the data. With a more thorough characterization of the nature of the data, more complex geostatistical algorithms can be employed.
As previously discussed (see Discussion), due to the anisotropic variation in both elevation and precipitation data, it is expected that accounting for this characteristic would increase interpolation performance. Anisotropic models were not evaluated in the current work but it is likely that the application of such models would result in a potentially significant decrease in the amount of error associated with the estimated surface. Furthermore, as also discussed in greater detail previously, an investigation on the effects of the application of an anisotropic spatial filter to elevation data may be worthwhile.
In the face of sparse spatial data for the making of an accurate climate map, Daly et al. (2002) propose an expert system approach in addition to PRISM. The accumulation of expert knowledge in the spatial patterns of climate and their relationships with geographic features will help to enhance, control and make better parameter estimation for the numerical based statistical analysis. The knowledge framework based on such a system coupled with traditional statistical methods is suggested by the authors to be more effective conventional precipitation modeling.
 |
back to
top |
 |