Análise comparativa de métodos de interpolação em modelos digitais de elevação

Abstract

Digital elevation models (DEM) are the basic elements for the development of environmental or socio-spatial analyses. But the quality of these tridimensional models of the terrain is intimately related to its spatial resolution, in such way that the better the spatial resolution, the better the capability to represent the shapes in the surface of the imaged body. However, the products of better spatial resolution have high acquisition costs, in such way that the State does not always have the resources needed for their acquisition, be it on the sphere of urban/regional planning or for academic research. This scenario results in the usual use of lower spatial resolution DEM which are made available for free. Currently, DEM with better spatial resolution and 100% free are emerging, such as the products from ALOS-PALSAR, which delivers DEM with spatial resolution of 12.5m. One of the free ALOS-PALSAR DEM predecessors was the SRTM DEM, which was initially distributed with a spatial resolution of 90m, but Valeriano (2008; 2010; 2010a), after a series of refinement processes perfomed by means of geostatistics, was able to generate new products with spatial resolution of 30m. In this regard arises the question about the possibility of application of similar methodological processes on the ALOS-PALSAR DEM, in order to obtain products with spatial resolution higher than 12.5m. In this manner, the general aim of this research was to do a comparative analysis of several interpolation methods, by changing the spatial resolution of the ALOSPALSAR DEM to the spatial scales of 10m, 8m and 6m. In order to observe, qualitatively and statistically, the potentialities and losses during this process, the comparison parameter used was a DEM generated with LiDAR of spatial resolution of 1m. As general results, regarding the statistical analyses, the tests applied were the Pearson’s Coefficient Correlation (r), Coefficient of Determination, Mean Absolute Error (MAE), Mean Squared Error (MSE) and Root Mean Square Error (RMSE). A qualitative verification of the DEM was performed in order to do an analysis still more complete, where declivity and hypsometric maps were generated from the reinterpolated ALOS-PALSAR DEM in comparison with maps generated from LiDAR and terrain profiles with the central purpose of verification of the behavior of the dossel effect in the reinterpolated DEM and its influence in the capability of representation of the geometry of the terrain. At the end of the research, it was possible to conclude, from the statistical analysis and the qualitative assessment, that there are no significant enhancements with the process of interpolation for more detailed spatial resolutions, regardless of the interpolation method chosen, since it is impossible to create information out of nowhere. In this regard, much as the pixels size are reduced to generate the new DEM, through application of interpolation methods, the original source is made of data with spatial resolution of 12,5m. And even if there is no improvement, it is possible to measure the losses that occur throughout the process of enhancement of spatial resolution, which may aid researchers choose techniques that generate DEM with the smallest possible error, according to the use of these DEM. If the aim is to use the reinterpolated ALOS-PALSAR DEM for flood or inundation analysis, due to the low capability of reinterpolated DEM to model lowland areas, which are where these events take place, the studies will be prone to result in greater uncertainty. By contrast, if the aim of the studies is landslides, DEM has shown better results in representing steep surfaces where the occurrence of these events is concentrated. However, the choice of an interpolation method that underestimates less areas of higher declivity is necessary to improve the results. In this manner, should the researcher opt to use these techniques in the generation of a higher resolution DEM, it is necessary to be aware that the degree of uncertainty of the generated information will increase in order to know whether the errors are admissible in the final results of the research.