SRTM/TopSAR高程數據比對與多尺度分析

Abstract

干涉合成孔徑雷達(InSAR)可進行地表三維地形之測繪，InSAR具有作業受限小、製圖速度快、作業單價低之優點。然而InSAR測量受地物性質、地形效應與植披影響很大。本文針對兩組InSAR DEM進行高程比對與多尺度分析，分別為太空梭雷達製圖任務(SRTM)與空載地形合成孔徑雷達(TopSAR)高程數據，以航空攝影測量所得DEM為參考數據進行比對，對InSAR數據的差值平均、精度與粗差進行分析。比對區域面積達2497平方公里，比對點數共3,017,889點。 比對結果顯示在多山地區SRTM之精度為9.451公尺，在平坦地區精度可達7.579公尺，均優於NASA公佈的精度規範。而空載TopSAR資料則包含大量粗差，在山區測量的品質不理想。若不經過粗差偵測和編修，無法達到NASA公佈的1~5公尺精度，在粗差移除後平坦地區的精度為5.603公尺，山區則為12.383公尺。 多尺度分析則使用以小波轉換與傅立葉轉換分析三組資料的尺度性質，探討不同尺度下三組資料的關連與差異，並分析不同尺度對於精度的影響，最後則將小波分析於粗差偵測與等高線縮編的應用上，顯示小波理論在高程資料縮編與不同尺度DEM分析的應用潛力。

InSAR technology takes advantages of cloud penetrating capability of microwaves, being claimed as a system for all weather and day and night. However, the accuracy of InSAR data was affected by terrain slope, canopy and land cover. The quality of InSAR DEM needs to be further inspected. This thesis focused on the DEM comparison and multiscale analysis for two InSAR DEM datasets. One is the 3 arc data from Shuttle Radar Topographic Mission (SRTM), and the other is from Topographic Synthetic Aperture Radar (TopSAR) mission. A DEM dataset obtained by aerial photogrammetry is used to validate the system bias, random error and blunders of the two InSAR DEMs. The aera size of the study aera is 2,497 km2 with a total of 3,017,889 pixels. Comparing with the photogramtric DEM, the RMSE of SRTM data is 9.451m in the mountainous area and 7.579m in the flat area. It is equal or better than the offical specification. But the TopSAR data that contained massive blunders was not as good as claimed in the specification of an accuracy 1~5m, especially in the mountainous aera. After removing the blunders, the RMSE of DEM difference is 5.603m in the flat area and 12.383m in the mountainous area. The frequency properties of three datasets were also studied in multiscale analysis with wavelets and Fourier transform, such as the relation and difference between three dataset in different scale, and the impact in accuracy in different scales. The application in contour genlization and blunder detection was also proved to be useful by applying wavelet and multiscale analysis to th DEMs.