Spatially distributed estimates of meteorological data are becoming increasingly important as inputs to spatially explicit landscape, regional, and global models. Accurate estimates of meteorological values such as temperature, precipitation, and evapotranspiration are required for a number of landscape scale models, including those of regeneration, growth, and mortality in forest ecosystems.
Given a set of meteorological data, researchers are confronted with a variety of stochastic and deterministic interpolation methods to estimate meteorological variables at unsampled locations. Depending on the spatial attributes of the data, accuracies may vary widely among different spatial interpolation methods. The choice of spatial interpolator is especially important in mountainous regions where data are sparse and variables may change over short spatial scales.
While there have been comparisons of interpolation methods, few research efforts have been directed towards comparing the effectiveness of different spatial interpolators in predicting temperature. Due to the additional effort kriging and cokriging entails, it was decided to compare the effectiveness of kriging and cokriging in estimating mean, maximum, and minimum temperature at unsampled locations with less computationally intensive interpolation techniques such as inverse distance weighted averaging, cubic splining, the fitting of a trend surface, polynomial regression, and the lapse rate method.
Eight interpolation techniques (inverse distance squared, optimal inverse distance, cubic splining, trend surface analysis, regression, kriging, cokriging, and the lapse rate method) were compared in their ability to predict temperature at unsampled locations. Temperature data for two regions, two scales (minimum and maximum temperatures) and three temporal scales (10 year mean, seasonal, and daily) were prepared and the eight methods were compared on the basis of bias, MAE, and MSE. In addition, summary statistics of interpolated mean, minimum, and maximum temperatures were recorded to determine how well the interpolated data represented the original temperature values. This dissertation provides evidence that certain apriori data characteristics such as temperature range, temperature variance, and temperature correlation with elevation may influence interpolator choice. The dissertation results also indicate that spatial scale and the relative spatial density and distribution of sampling stations may influence interpolator choice.
These results should be of interest to scientists studying global warming. The MAEs associated with interpolation techniques which did not use ancillary information were far greater than the 0.5°F to 1.0°F estimate of global warming over the past 100 years. The use of regression techniques which utilize the relationship between temperature and elevation as ancillary information offers significant improvement over the current inverse distance weighting methods. The dissertation also shows that when station elevations are not representative of regional elevations, bias occurs. In Region 2, stations were underrepresented for higher elevations. Interpolation techniques which did not use elevation as ancillary information were biased 1.0°F to 3.0°F above techniques which used elevation. While it is unclear what the extent of this effect is on a global scale, one would suspect the use of distance weighting techniques would bias global estimates upwards.
These dissertation results should also be of interest to scientists who use kriging and cokriging to interpolate irregularly spaced data onto a rectangular grid. The results indicate that when data are isotropic, less subjective methods, such as optimal inverse distance, have lower MAE values. The semivariogram fitting methodology outlined in this dissertation demonstrates how to fit semivariograms iteratively using an indicative goodness of fit (IGF) metric. Semivariogram fitting using an IGF is less subjective and more accurate than traditional fit-by-sight methods. Despite its mathematical elegance, kriging and cokriging did not perform better than many other less computationally intensive methods. In addition, when there is a more intensely sample covariate which is highly correlated, polynomial regression gave far better results than kriging or cokriging.
The results of this dissertation should also be of interest to users of geographic information systems (GISs). Because climatic data such as temperature is sampled from an irregular network, a number of interpolation techniques can be used to convert the data to a regular grid for use in visualization, models and GISs. This dissertation shows that the choice of spatial interpolator can influence the resulting data accuracy. In addition, data attributes influence the choice of interpolator. What is dissertation shows, is that through preliminary data analyses, an interpolator may be chosen which yields the most accurate grid for input into a GIS.
It should be noted that this dissertation has wider ranging applications beyond the three examples mentioned above. The results should be of interest in any field where point data is interpolated onto a regular grid. Additional application areas include, but are not limited to, medical imaging, scientific visualization, weather forecasting, ecological modeling, forestry, petroleum exploration, and hydrological modeling. / Ph. D.
| Identifer | oai:union.ndltd.org:VTETD/oai:vtechworks.lib.vt.edu:10919/38139 |
| Date | 06 June 2008 |
| Creators | Collins, Fred C. |
| Contributors | Forestry, Buhyoff, Gregory J., Campbell, James B. Jr., Smith, James L., Bolstad, Paul V., Gregoire, Timothy G. |
| Publisher | Virginia Tech |
| Source Sets | Virginia Tech Theses and Dissertation |
| Language | English |
| Detected Language | English |
| Type | Dissertation, Text |
| Format | xiv, 234 leaves, BTD, application/pdf, application/pdf |
| Rights | In Copyright, http://rightsstatements.org/vocab/InC/1.0/ |
| Relation | OCLC# 35799247, LD5655.V856_1995.C655.pdf |
Page generated in 0.0022 seconds