DCTII

Each indicator bar shows one component of the DCTII composite score. The bar length represents the value relative to a fixed physical reference bound for that metric. The composite score weights four sub-indicators asymmetrically: nighttime ΔT (55%), waste heat flux (25%), heat island extent (10%), and population exposure (10%).

ΔT Day / Night — surface temperature difference (°C) between the data center buffer zone and matched control sites. Nighttime ΔT is the strongest signal because convective mixing stops after sunset, making waste heat retention directly measurable. Day reference: 0–3°C; night reference: 0–2°C.

Waste Heat Flux — estimated thermal output in watts per square meter, derived from facility capacity, cooling efficiency, and a logistic ramp-up model that reflects how data centers gradually reach full utilization over approximately 8 years after activation (midpoint at 3 years, floor at 15% capacity). Reference: 0–250 W/m².

Heat Island Area — spatial extent (km²) where temperatures exceed one standard deviation above the control mean. Larger values indicate thermal influence spreading beyond the facility perimeter. Reference: 0–10 km².

Population Exposed — estimated number of residents within the elevated-temperature zone, calculated from regional census-based population density and the detected heat island area, with annual growth modeling. Reference: 0–50,000.

This chart plots the monthly surface temperature difference (ΔT in °C) between the data center treatment zone and its statistically matched control zone over the course of a year.

The orange line shows daytime ΔT (from satellite overpasses around 10:30 AM local time). The blue line shows nighttime ΔT (from ~10:30 PM overpasses). A positive value means the data center area is warmer than comparable surroundings.

Seasonal patterns matter: summer peaks indicate heat amplification during already-warm months, compounding stress on nearby communities. Persistent winter signals confirm year-round waste heat output regardless of climate.

If nighttime ΔT consistently exceeds daytime, the facility is trapping heat that does not dissipate overnight — a hallmark of urban heat island intensification driven by waste energy.

Enter a location and your planned data center specs, and the model estimates how much extra heat it would add to the surrounding area before you break ground. The DCTII Score (0–100) summarizes the overall heat impact — lower is better. ΔT Night is the most reliable number: it shows how many degrees warmer the ground around your site is expected to be at night compared to a similar area without a data center, driven mainly by the waste heat your cooling system releases. ΔT Day is derived from the night number using a climate-based formula and is less precise. Heat Area estimates how many square kilometers would feel that extra warmth, and Waste Heat Flux measures how much heat energy your facility pushes into the surrounding air per square meter. The Night ΔT Drivers chart shows which factors matter most — red bars push the score up (more heat), green bars push it down (less heat). Features like sensible heat flux and impervious surface fraction typically dominate.

The model was trained on 10 years of satellite thermal data from 42 real data centers across six regions: Phoenix, Houston, Northern Virginia, Central Texas, Toronto, and Montreal. It uses a machine learning method called LightGBM — a type of algorithm that learns patterns by building hundreds of small decision trees, each one correcting the mistakes of the last. It learned that air-cooled facilities in hot dry climates produce the most heat impact, while tower-cooled facilities in humid or cold climates produce less. When you submit a new location, the model compares your site’s features — vegetation cover, sealed surface fraction, elevation, population density, and your DC specs — to everything it learned from those 42 sites and returns its best estimate with a confidence range shown as the shaded band on each bar.

Limitations you must know before using these results. This model was trained on sites in six specific North American regions, so it works most reliably for locations in or near the US Sun Belt, Mid-Atlantic, and southern Canada. Predictions for cities far outside these regions — such as the US Pacific Northwest, Mexico, or the Caribbean — are less reliable and the yellow warning banner will appear when the tool detects your location is outside its comfort zone. Daytime temperature predictions are estimated indirectly and carry roughly twice the uncertainty of nighttime predictions. Two Phoenix sites in our training data were actually cooler than their surroundings due to heavy irrigation — if your site has unusually high vegetation in a desert climate, the green leaf warning will appear and you should treat predictions as approximate. The model does not account for local wind patterns, nearby water bodies, or future urban development. Results should be used as a planning indicator alongside site-specific environmental studies, not as a substitute for them.