Temperatura notturna al suolo

Night Land Surface Temperature — RF Downscaling

Overview

The Night LST layer provides high-resolution nighttime land surface temperature maps by downscaling MODIS thermal data (~1 km) to 25 m resolution using Random Forest machine learning. Nighttime LST is particularly informative for urban heat island (UHI) analysis because it reveals heat retention patterns that daytime measurements cannot capture.

The downscaling uses three auxiliary variables:

• NDVI from Sentinel-2 (10 m) — proxy for vegetation cooling effects
• Land Cover from ESA WorldCover (10 m) — categorical surface type influencing thermal behavior
• Elevation from SRTM (30 m) — altitude-driven temperature lapse rates

In addition to the LST map itself, the product generates:

• UTFVI (Urban Thermal Field Variance Index) — quantifies urban heat island intensity
• SUHI (Surface Urban Heat Island) analysis — temperature anomaly relative to rural surroundings

---

Calculation Methodology

Random Forest downscaling pipeline:

1. MODIS LST extraction — Nighttime LST values from MOD11A1 (Terra, ~22:30 local) and MYD11A1 (Aqua, ~01:30 local) are extracted for the analysis area
2. Quality filtering — Pixels with LST errors > 2 K or excessive cloud contamination are excluded
3. Feature preparation — NDVI, land cover, and elevation rasters are aligned to the MODIS grid
4. Model training — Random Forest learns the relationship between MODIS LST and the auxiliary features at 1 km resolution
5. High-resolution prediction — The trained model predicts LST at the 25 m resolution of the auxiliary features
6. UTFVI computation — UTFVI = (LST - LST_mean) / LST_mean, where LST_mean is the spatial mean
7. SUHI analysis — Temperature difference between urban and reference rural pixels

MODIS LST characteristics:

• MOD11A1 (Terra): ~22:30 local overpass
• MYD11A1 (Aqua): ~01:30 local overpass
• Both provide daily LST with ±1 K accuracy under clear-sky conditions

---