Terrain-type-specific Evaluation of Korea Multi-purpose Satellite-3 Stereo Imagery Using Hybrid Digital Elevation Model Refinement Method

Sep 14, 2026

Generating accurate digital elevation models (DEMs) from high-resolution satellite stereo imagery enables reliable topographic information, yet elevation errors vary strongly by terrain. A terrain-adaptive DEM refinement framework is proposed, combining U-Net for initial reconstruction, morphological filtering for nonground separation, low-rank matrix factorization for systematic bias elimination, and U-Net++ for boundary and detail restoration. Validation with KOMPSAT-3 stereo data over U.S. and Canadian sites demonstrates terrain-dependent improvements, with urban RMSE reductions exceeding 40–50%.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages14

Reading Time35 min

AccessFree to download

1. Introduction

Terrain dependence of DEM errors

Limitations of conventional DEM workflows

Deep learning for dense prediction and reconstruction

Role of low-rank plus sparse decomposition

Sensors and Materials, Vol. 38, No. 7 (2026) 4277–4290 4277 MYU Tokyo S & M 4564 Terrain-type-specific Evaluation of Korea Multi-purpose Satellite-3 Stereo Imagery Using Hybrid Digital Elevation Model Refinement Method Jae Myeong Kim, 1 Ji Yoon Kim,2* Kwan Young Oh,3 and Kwang Jae Lee3 1College of Interdisciplinary Studies, Department of Urban Planning Engineering, Seokyeong University, 124 Seogyeong-ro, Seongbuk-gu, Seoul 02713, Republic of Korea 2Department of Technology and Management, Carbon Ace, 484 Tongil-ro, Seodaemun-gu, Seoul 03628, Republic of Korea 3Satellite Application Center, Korea Aerospace Research Institute, 169-84 Gwahak-ro, Yuseong-gu, Daejeon 34133, Republic of Korea (Received November 28, 2025; accepted July 7, 2026) Keywords: satellite imagery, DEM refinement, U-Net++, low-rank matrix factorization, deep learning, terrain-stratified DEM evaluation Generating digital elevation models (DEMs) using high-resolution satellite imagery is an efficient method for acquiring topographic information over vast areas. However, the accuracy of satellite-derived DEMs is affected by terrain-specific error mechanisms: steep topography amplifies geometric distortions, dense urban blocks create severe occlusion artifacts, vegetated nonurban land introduces canopy-related overestimation, and low-texture flat surfaces degrade stereo correspondence. Despite these well-known terrain dependences, in most refinement studies, only area-averaged accuracy metrics, which can obscure substantially different residual error structures across landscape types, are reported. To address this deficiency, in this study, we introduce a terrain-adaptive DEM refinement framework that integrates U-Net++ deep learning with low-rank matrix factorization and evaluates its effectiveness across four distinct terrain categories: mountainous, flat, urban, and nonurban areas. The framework operates through initial digital surface model reconstruction via U-Net, morphological filtering for nonground separation, low-rank decomposition for systematic bias elimination, and U-Net++ refinement for boundary and detail restoration. Validation experiments using Korea Multipurpose Satellite-3 (KOMPSAT-3) stereo imagery over San Francisco, USA, and multiple Canadian sites reveal terrain-dependent improvement patterns. Urban areas exhibited the most dramatic gains, with root mean square error (RMSE) reductions exceeding 40–50% relative to three commercial photogrammetric software packages. Mountainous regions showed consistent improvement despite inherently higher baseline errors, while flat and nonurban terrains demonstrated stable sub-3 m RMSE convergence. Results of cross-site comparisons between the US and Canadian test areas suggest that the framework exhibits consistent terrain-specific improvement patterns under the conditions examined in this study. * Corresponding author: e-mail: [aliasgis@gmail.com](aliasgis@gmail.com)[ ](aliasgis@gmail.com)[https://doi.org/10.18494/SAM6193](https://doi.org/10.18494/SAM6193) ISSN 0914-4935 © MYU K.K. [https://myukk.org/](https://myukk.org/) 1. Introduction Accurate digital elevation models (DEMs) derived from high-resolution optical satellite imagery are fundamental to topographic mapping, infrastructure planning, and environmental analysis. Although satellite stereo imagery provides a cost-effective pathway for constructing DEMs over extensive areas, the resulting elevation products exhibit error characteristics that are strongly dependent on the underlying terrain type. In urban environments, tall structures create severe occlusion and shadow effects, leading to matching failures and spurious elevation spikes . In nonurban and forested regions, vegetation canopy introduces systematic overestimation of true ground elevation . Mountainous terrain amplifies error sources owing to steep slopes that distort the baseto-height ratio, introducing both random noise and systematic biases, whereas

FAQs

Why do satellite-derived DEM errors depend on terrain type?

Terrain-specific error mechanisms arise from steep topography, urban occlusion and shadowing, vegetation canopy overestimation in nonurban/forested areas, and low-texture flat surfaces that weaken stereo correspondence.

What components are integrated in the proposed terrain-adaptive DEM refinement framework?

The framework uses U-Net for initial digital surface model reconstruction, morphological filtering to separate nonground areas, low-rank matrix factorization to remove systematic bias, and U-Net++ refinement to restore boundaries and details.

How do results vary across the four terrain categories?

Urban areas show the largest RMSE reductions (exceeding 40–50% versus three commercial software packages). Mountainous regions improve consistently, while flat and nonurban terrains converge to stable sub-3 m RMSE.

Sep 14, 2026

Uploaded by นรินทร์