Statistical model to predict contour elevation using shuttle radar topographic mission digital elevation data

Abstract

This study is based on two geographical datasets, namely, Shuttle Radar Topographic Mission (SRTM) elevation and contour elevation. The SRTM data is available for all locations in Sri Lanka. It follows the shape of the actual ground but not the actual elevation of the surface. This is due to errors introduced during processing. The contour data is obtained from the actual ground survey data of contour maps and using ArcGIS software. The survey data is more reliable, but more expensive. Since it does not contain ground level variations, Sri Lanka does not have contour elevations at all locations. Measuring contour elevation for all locations is a costly procedure. Therefore, finding a method to evaluate the approximated value of contour elevation with a less costly method is essential. Thus the objective of this study is to find a statistical model to predict the contour elevation based on SRTM data. Both types of data are available only for four locations: Paddhiruppu, Kegalle, Badulla, and Katharagama, in Sri Lanka. According to the geography of Sri Lanka, three clusters are distinguishable by elevation. These are the Central Highlands, the Plains, and the Coastal belt. Since the data used in this study are for four different locations and these locations fall into three different clusters, three regression models are fitted for each cluster and the models are validated. Multiple, linear regression analysis is used to fit the models. The t- test is used to test the significance of parameters while the F- test is used to test the significance of the overall model. Residual analysis is carried out to test the normality, homoscedasticity and auto correlation of the residuals. The goodness of the fitted model is evaluated by the coefficient of determination   2 R . Approximately 99% of the variation is explained by the fitted models and 82% by the validated model. Thus, if the SRTM data value is known, by choosing the appropriate model based on its cluster, the approximated contour elevation could be predicted.