Prepare the satellite inputs
Remove cloudy pixels. Use six visible and infrared bands to measure vegetation, moisture and built-surface patterns, including nearby areas.
INPUT: REFLECTED LIGHTYOUTH FOR EARTH
Find urban heat.
Focus cooling efforts.
Paved school surroundings, shopping areas and streets with little vegetation can trap surface heat. Explore San José’s satellite evidence to find places worth investigating for shade and greening.
Explore the heat map ↗154.8 km²
of San José studied using
real Landsat observations
A machine-learning study of surface heat, tested on later satellite images.
SAN JOSÉ, CALIFORNIA 37.33° N, 121.89° W
N ↑
CoolerHotter
01 / FROM SATELLITE DATA TO A TESTED MODEL
Learn heat patterns
from reflected light.
Landsat is NASA and the U.S. Geological Survey’s Earth-observing satellite program. Its images capture reflected sunlight and heat emitted by the ground.
Learn from measured heat
Pair earlier light readings with known surface temperatures. Our model learns patterns connecting reflected light with relative surface heat. Open the technical details to see its settings.
TRAINING ANSWER: THERMAL DATAPredict, then check
Estimate heat in two unseen 2025 images using reflected light alone. Compare the estimates with the satellite’s withheld thermal readings.
OUTPUT: °C ABOVE OR BELOW THE IMAGE MEDIANTRAIN THE MODEL2022—20248 images with light and heat measurements
→
TEST ON LATER DATES20252 images excluded from training
What the ML tests
Can reflected light recover surface-heat patterns? Thermal readings are evaluation targets, not input features. The observed map is available directly for this date; ML is not required to view it.
Model settings & evaluation design
ExtraTrees uses 220 trees, minimum leaf size 3, six reflectance bands, four spectral indices, and 90/270 m neighborhood context. Eight earlier images train the model; two later images test it. A separate spatial check withholds 600 m blocks.
02 / MEASURED PERFORMANCE
Tested on later dates.
Better than a simple baseline.
Surface-temperature prediction error
RMSE in °C (lower is better)Same temperature difference everywhereOur model
Baseline: predict the scene median at every location.
—lower average error
across the two test dates
across the two test dates
Did it find the hottest places?
Among locations flagged in the hottest fifth:
A match means a location was also in the satellite-measured hottest fifth. These results come from the later 2025 images.
Separate spatial check: compare with a simpler model
This check holds out 600 m blocks in the training-year images. It is separate from the two later test dates above.
Loading spatial comparison…
Real Landsat observations with a separate heat-band targetSurface heat differences, rather than air temperature
03 / EXPLORE YOUR SURROUNDINGS
Familiar places. A closer look at heat.
Choose a place or click the map. Compare the surrounding 300 m, then explore which hot, low-green areas qualify for field review.
Loading satellite observations and places…
04 / FROM EVIDENCE TO ACTION
From satellite evidence
to field review.
Use hot, low-green areas as a starting point for investigating urban shade and greening. A map is the beginning of a site assessment.
16.2 km²
flagged for review
10.5% of the mapped area
154.8 km² studied → 16.2 km² flagged
Source-grid area from the original analysis. Custom controls never alter this archived total. No cooling benefit has been measured.
A DOCUMENTED LOCAL COMPARISON
San Pedro Square & the park vicinity
Loading the satellite comparison…
Current field observations: not collected. The satellite image was acquired on August 31, 2025; it does not establish current site conditions.
Proposed follow-up: check shade, paved surfaces, planting space and local priorities at each location. A cooler observed area can still be informative; the comparison does not prove vegetation caused the difference.
Replaces the current comparison. Use “Undo selection change” on the map to restore it.
YOUR PLANNING OUTPUT
Field-review brief
Add one to three areas on the map to prepare a brief. Explore the map ↗