A Diffusion-Based Framework for Terrain-Aware Remote Sensing Image Reconstruction

TL;DR

SatelliteMaker presents a diffusion-based framework for terrain-aware remote sensing image reconstruction that robustly fills large missing regions while preserving spatial, spectral, and temporal coherence. By incorporating Low-Rank Adaptation (LoRA) for lightweight fine-tuning, Digital Elevation Model (DEM) conditioning via ControlNet, and a VGG-Adapter with distribution and style losses, the method achieves state-of-the-art performance on Landsat-8 and EarthNet2021 benchmarks. The approach demonstrates strong generalization across geographies and times, with extensive ablations confirming the value of the VGG-Adapter and the resilience of diffusion-based restoration to increasing missing data. The work offers a scalable, terrain-consistent solution for quantitative remote sensing tasks, enabling more reliable environmental monitoring and analysis in cloud-covered or data-sparse scenarios.

Abstract

Remote sensing imagery is essential for environmental monitoring, agricultural management, and disaster response. However, data loss due to cloud cover, sensor failures, or incomplete acquisition-especially in high-resolution and high-frequency tasks-severely limits satellite imagery's effectiveness. Traditional interpolation methods struggle with large missing areas and complex structures. Remote sensing imagery consists of multiple bands, each with distinct meanings, and ensuring consistency across bands is critical to avoid anomalies in the combined images. This paper proposes SatelliteMaker, a diffusion-based method that reconstructs missing data across varying levels of data loss while maintaining spatial, spectral, and temporal consistency. We also propose Digital Elevation Model (DEM) as a conditioning input and use tailored prompts to generate realistic images, making diffusion models applicable to quantitative remote sensing tasks. Additionally, we propose a VGG-Adapter module based on Distribution Loss, which reduces distribution discrepancy and ensures style consistency. Extensive experiments show that SatelliteMaker achieves state-of-the-art performance across multiple tasks.

Figures (6)

• Figure 1: The model's training and inference pipeline consists of two main stages. We design the training phase to develop a diffusion model specifically for remote sensing imagery. A guidance mechanism is incorporated to control the generation process, ensuring stable and accurate image outputs, which are essential for subsequent quantitative remote sensing analysis. In the inference phase, we employ an adaptation module to ensure that reconstructed regions align with the target domain.
• Figure 2: Global distribution of selected regions for Landsat imagery. The imagery was downloaded using Google Earth Engine (GEE) from various countries, ensuring diverse geographical coverage for comprehensive analysis.
• Figure 3: Comparison for Task-1, addressing missing data in specific regions over a fixed time period.
• Figure 4: Comparison for Task-2 using the EarthNet2021 dataset with selected missing data days. The above images are the results after brightness adjustment and gamma correction, with a coefficient of 1.2. The original image was used for calculating the evaluation metrics.
• Figure 5: Ablation study on the impact of adding the VGG-Adapter.