New benchmark dataset HeatCast enables neighborhood-scale LST forecasting

Researchers have introduced HeatCast, a new benchmark dataset for forecasting land surface temperature (LST) at a neighborhood scale across 124 U.S. cities. This dataset, derived from Landsat imagery, provides monthly LST data at 30m resolution, along with associated environmental factors like elevation and spectral indices, covering the period from 2013 to June 2025. The benchmark includes a fixed temporal split, LCZ-stratified metrics, and an evaluation harness. In testing, the Earthformer model achieved a 7.74 K RMSE for next-month LST forecasting, outperforming a CNN+LSTM model and demonstrating the potential of non-LST channels for accurate predictions. AI

IMPACT Enables more accurate neighborhood-scale climate and urban heat island effect modeling.

RANK_REASON The cluster describes a new benchmark dataset and evaluation for a specific research problem (LST forecasting), including a published paper and released code/data. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• arXiv
• Earthformer
• HeatCast
• Landsat program
• MIT
• U.S.

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →