Daily Land Surface Temperature Reconstruction in Landsat Cross-Track Areas Using Deep Ensemble Learning With Uncertainty Quantification

TL;DR

This work tackles the challenge of obtaining daily, high-spatial-resolution land surface temperature (LST) data in urban areas where temperature heterogeneity is pronounced. It introduces DELAG, a two-stage deep ensemble framework that combines enhanced annual temperature cycles (ATC) with Gaussian-process residuals, integrating ERA5-derived areal means and Landsat cross-track observations to reconstruct daily LST and quantify uncertainty via ensemble predictions. Across New York City, London, and Hong Kong, DELAG achieves RMSEs of 0.73–1.62 K under varying cloud conditions and enables accurate indirect estimation of near-surface air temperature (RMSEs of 1.48–2.11 K), outperforming existing methods. The approach provides a practical path to high-resolution, all-weather urban temperature data with uncertainty-aware reconstructions, supporting climate, health, and urban planning applications; code and data are publicly available.

Abstract

Many real-world applications rely on land surface temperature (LST) data at high spatiotemporal resolution. In complex urban areas, LST exhibits significant variations, fluctuating dramatically within and across city blocks. Landsat provides high spatial resolution data at 100 meters but is limited by long revisit time, with cloud cover further disrupting data collection. Here, we propose DELAG, a deep ensemble learning method that integrates annual temperature cycles and Gaussian processes, to reconstruct Landsat LST in complex urban areas. Leveraging the cross-track characteristics and dual-satellite operation of Landsat since 2021, we further enhance data availability to 4 scenes every 16 days. We select New York City, London and Hong Kong from three different continents as study areas. Experiments show that DELAG successfully reconstructed LST in the three cities under clear-sky (RMSE = 0.73-0.96 K) and heavily-cloudy (RMSE = 0.84-1.62 K) situations, superior to existing methods. Additionally, DELAG can quantify uncertainty that enhances LST reconstruction reliability. We further tested the reconstructed LST to estimate near-surface air temperature, achieving results (RMSE = 1.48-2.11 K) comparable to those derived from clear-sky LST (RMSE = 1.63-2.02 K). The results demonstrate the successful reconstruction through DELAG and highlight the broader applications of LST reconstruction for estimating accurate air temperature. Our study thus provides a novel and practical method for Landsat LST reconstruction, particularly suited for complex urban areas within Landsat cross-track areas, taking one step toward addressing complex climate events at high spatiotemporal resolution. Code and data are available at https://skrisliu.com/delag

Figures (15)

• Figure 1: Spatial resolution comparison of satellite-derived land surface temperature products in New York City: MODIS (1 km), Landsat (30 m resampled from 100 m). The overlaid polygons are the New York City merged zip-code boundaries that have daily counts of emergency department visit data. The coarse-resolution LST data is not sufficient to capture spatial variations and urban heterogeneity. Landsat LST can capture temperature variations but is limited by temporal frequency and clouds.
• Figure 2: New York City is situated within the Landsat cross-track areas, completed being covered by two Landsat scenes (Path 014 Row 032, Path 013 and Row 032). This area, New York City and the surroundings, has complex urban landscape and high population density.
• Figure 3: Global areas covered by two or more Landsat scenes (in red) within a revisit period of 16 days
• Figure 4: Percentage of Landsat cross-track areas in global countries and regions. The top-20 are listed on the lower left. The ratio of Landsat coverage as a function of latitude is shown in the middle right.
• Figure 5: Meteorological stations of the three cities, overlapped with annual mean temperature and geographical unit boundaries. NYC: 7 stations over 135 merged zip codes; London: 6 stations over 983 Middle Super Output Area; Hong Kong: 23 stations over 452 District Council Ordinary Election Constituency Boundaries.