Fusion of multi-scale DEMs using a regularized super-resolution method

Abstract

The digital elevation model (DEM) is a significant digital representation of a terrain surface. Although a variety of DEM products are available, they often suffer from problems varying in spatial coverage, data resolution, and accuracy. However, the multi-source DEMs often contain supplementary information, which makes it possible to produce a higher-quality DEM through blending the multi-scale data. Inspired by super-resolution (SR) methods, we propose a regularized framework for the production of high-resolution (HR) DEM data with extended coverage. To deal with the registration error and the horizontal displacement among multi-scale measurements, robust data fidelity with weighted L 1 norm is employed to measure the conformance of the reconstructed HR data to the observed data. Furthermore, a slope-based Markov random field (MRF) regularization is used as the spatial regularization. The proposed method can simultaneously handle complex terrain features, noises, and data voids. Using the proposed method, we can reconstruct a seamless DEM data with the highest resolution among the input data, and an extensive spatial coverage. The experiments confirmed the effectiveness of the proposed method under different cases.

Fusion of high-resolution DEMs for water flow modeling

“…To address inhomogenity of available DEM products, several methods of fusing DEMs have been developed to obtain a complete DEM coverage with improved quality. Fusion approaches vary from simple techniques, such as weighted averaging of input DEMs based on height error maps [3], or terrain derivatives [4,5], to more complex techniques involving the use of sparse representations [6], frequency domain filtering [7], slope-based Markov random field regularization [8], or k-means clustering [9].…”

A Global Seamless DEM Based on Multi-Source Data Fusion (GSDEM-30): Product Generation and Evaluation

“…For SRTM1 data beyond the 50° latitude, the tiles are sampled at 2 arc-seconds along the longitudinal direction. To ensure the spatial consistency of the global data, we used the multi-scale data fusion method [26] to up-sample the high-latitude data between 50° and 60° to 1-arc×1-arc tiles. To reconstruct the high-latitude SRTM tiles, the input data include the original SRTM1 and the high-resolution ASTER GDEM and AW3D30 data.…”

“…A is the multiplication of the translation matrix k M , the sampling matrix k D and the cropping matrix k O , which can be described as k [26]. Besides, k n stands for the additive noises in the input DEMs, also including the model errors.…”

“…The second term is the regularization prior used to describe the spatial characteristics of the fused data. Here we employ the adaptive Markov-random field regularization to model the contextual correlations between neighboring elevation pixels [26]. For   c d  in equation (1), the finite-difference approximations to second-order derivatives in four directions ( =4  ) are employed to measure the spatial relationship between the neighboring pixels [50].…”

“…Based on the fusion methods, great efforts have been made to generate high-quality DEM products by merging the existing datasets [19,21,25].In this paper, we introduce a fused DEM dataset (82°S-82°N) mainly reconstructed from the 30-m unfilled SRTM1, AW3D30, ASTER GDEM v2 and ICESat GLAS data, which we refer to as GSDEM-30. The goal is to generate a global seamless DEM using the multi-source accuracy enhancement method and the regularized multi-scale fusion algorithm [22,26]. Considering the global data characteristics, the artificial neural network (ANN) based accuracy enhancement and void filling method were used for the void SRTM tiles between 50°S-50°N, while the multi-scale fusion algorithm was applied to the SRTM tiles sub-sampled to a 2-arc-second resolution along the longitude within 50-60°N and 50°-60°S.…”