granite-geospatial-land-surface-temperature
| Property | Value |
|---|---|
| License | Apache-2.0 |
| Research Paper | Link to Paper |
| Author | IBM Granite |
| Architecture | SWIN Transformer with UperNet |
What is granite-geospatial-land-surface-temperature?
This is a sophisticated geospatial foundation model designed to predict land surface temperature (LST) with high precision. Developed by IBM Granite, it combines Harmonised Landsat Sentinel-2 (HLS L30) imagery with ECMWF Reanalysis v5 (ERA5-Land) climate data to generate accurate temperature predictions at 30-meter resolution. The model has been trained on data from 28 global cities across various hydroclimatic zones, covering the period 2013-2023.
Implementation Details
The model utilizes a Shifted Windowing (SWIN) Transformer architecture and builds upon the IBM Earth Observation Foundation Model "Prithvi-SWIN-L". It features an unfrozen pre-trained encoder and a UperNet regression decoder with an auxiliary 1-layer Convolution regression head.
- Input Requirements: 224x224 patches with 6 HLS bands (B02-B07) and ERA5 temperature data
- Output: High-resolution (30m) LST predictions
- Implementation Framework: Terratorch
Core Capabilities
- High-resolution land surface temperature prediction
- Temporal gap filling through "Tweening"
- Urban Heat Island (UHI) effect analysis
- Climate impact assessment
- Hourly temperature predictions at 30m resolution
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its ability to combine satellite imagery with climate data to produce high-resolution temperature maps, specifically designed to address urban heat challenges. Its temporal gap-filling capability through "Tweening" makes it particularly valuable for continuous monitoring.
Q: What are the recommended use cases?
The model is ideal for urban planning, climate research, heat stress assessment, and environmental monitoring. It's particularly useful for studying Urban Heat Island effects, energy demand planning, and climate adaptation strategies in cities.
