Selected work

Representative Papers

HeatCast: A Benchmark for Neighborhood-Scale LST Forecasting across 124 U.S. Cities

Aug 07, 2026

This study addresses the lack of a unified, high-resolution, and open-access benchmark for urban land surface temperature (LST) prediction. To this end, it establishes the first Landsat-based 30-meter monthly LST forecasting benchmark covering 124 U.S. cities from 2013 to 2025, integrating multi-channel remote sensing data—including surface reflectance, spectral indices, albedo, and Local Climate Zone (LCZ) labels—and introducing a fixed temporal split with LCZ-stratified evaluation. Experiments using CNN+LSTM and Earthformer models demonstrate that Earthformer achieves an RMSE of 7.74 K for next-month LST prediction; notably, using only non-LST input channels yields an RMSE of 7.72 K, substantially outperforming approaches relying solely on historical LST or RGB data, thereby confirming the efficacy of multi-source non-LST information for accurate LST forecasting.

0 citationsRead paper

Are Pretrained Image Matchers Good Enough for SAR-Optical Satellite Registration?

Apr 11, 2026

This study addresses the critical challenge of cross-modal registration between optical and synthetic aperture radar (SAR) images in remote sensing for disaster response, systematically evaluating 24 pre-trained matchers under a zero-shot setting. By leveraging tile-based large-image inference, robust geometric filtering, and control-point-based metrics, the work demonstrates that explicit cross-modal training is not essential—foundation model features such as DINOv2 exhibit sufficient modality invariance to partially substitute for supervised signals. Notably, deployment protocols exert a far greater influence on accuracy than model choice itself. Experiments show that RoMa achieves an average error of 3.0 pixels on SpaceNet9 without any cross-modal training, while MatchAnything-ELoFTR attains 3.4 pixels; further refinement of deployment protocols reduces registration error by up to 33-fold.

0 citationsRead paper

Recent publications

Latest Papers

HeatCast: A Benchmark for Neighborhood-Scale LST Forecasting across 124 U.S. Cities

Aug 07, 2026

This study addresses the lack of a unified, high-resolution, and open-access benchmark for urban land surface temperature (LST) prediction. To this end, it establishes the first Landsat-based 30-meter monthly LST forecasting benchmark covering 124 U.S. cities from 2013 to 2025, integrating multi-channel remote sensing data—including surface reflectance, spectral indices, albedo, and Local Climate Zone (LCZ) labels—and introducing a fixed temporal split with LCZ-stratified evaluation. Experiments using CNN+LSTM and Earthformer models demonstrate that Earthformer achieves an RMSE of 7.74 K for next-month LST prediction; notably, using only non-LST input channels yields an RMSE of 7.72 K, substantially outperforming approaches relying solely on historical LST or RGB data, thereby confirming the efficacy of multi-source non-LST information for accurate LST forecasting.

0 citationsRead paper

Are Pretrained Image Matchers Good Enough for SAR-Optical Satellite Registration?

Apr 11, 2026

This study addresses the critical challenge of cross-modal registration between optical and synthetic aperture radar (SAR) images in remote sensing for disaster response, systematically evaluating 24 pre-trained matchers under a zero-shot setting. By leveraging tile-based large-image inference, robust geometric filtering, and control-point-based metrics, the work demonstrates that explicit cross-modal training is not essential—foundation model features such as DINOv2 exhibit sufficient modality invariance to partially substitute for supervised signals. Notably, deployment protocols exert a far greater influence on accuracy than model choice itself. Experiments show that RoMa achieves an average error of 3.0 pixels on SpaceNet9 without any cross-modal training, while MatchAnything-ELoFTR attains 3.4 pixels; further refinement of deployment protocols reduces registration error by up to 33-fold.

0 citationsRead paper