Temperature data reconstruction at high spatiotemporal resolution

Across 88 cities, 506 ZIP codes, and more than 3,000,000 buildings, Los Angeles County has just 24 national air temperature stations. This monitoring network is far too sparse, with even larger gaps in other parts of the U.S. and globally, limiting our ability to monitor temperatures within each neighborhood and to understand their health impacts. How can we get good enough temperature data?

Papers:
Resolution revolution: A physics-guided deep learning framework for spatiotemporal temperature reconstruction (ICCV 2025 Workshop SEA)
Daily land surface temperature reconstruction in Landsat cross-track areas (arXiv preprint 2025)
Hourly air temperature mapping at 2 km resolution (dataset, working paper 2025)

Temperature variation and health

El Monte and Westchester are two neighborhoods in Los Angeles. On some days, compared to Westchester, El Monte sees higher temperature during the day and lower temperature at night.

It is surprising that research on temperature and health often overlooks variations within a single day. Inland Los Angeles can swing from 95°F in the day to 70°F at night in summer, while New York typically stays within a narrower 75–85°F range. How daily temperature swings shape human health risks, and how urbanization reshapes these temperature patterns?

Papers:
Racial and ethnic minorities disproportionately exposed to daily temperature variation in the U.S. (PNAS Nexus 2024)
Socioeconomic status, greenspace and respiratory disease under short-term temperature variations (Social Science & Medicine 2024)
Diurnal temperature range, neighborhood environment and mortality in Los Angeles (Urban Climate 2023)

Crop mapping and accurate crop yield estimation

Agricultural regions are often rainy and cloudy, limiting the use of satellite data. Collecting field samples remains challenging, especially in regions with small, decentralized farmlands and scarce data. Often, we have no knowledge of all the crop types cultivated in the farmland, and we need to distinguish crops from non-food grass and trees. How do we achieve accurate crop yield estimates amid these challenges?

Papers:
Few-shot hyperspectral image classification with unknown classes using multitask deep learning (TGRS 2021)
Crop mapping using full-year dual-polarized SAR data in rainy and cloudy regions (JSTARS 2021)
Multitask deep learning with spectral knowledge for hyperspectral image classification (GRSL 2020)