Data Methods: Heat Waves

Using Cal-Adapt's open source Python library climakitae, input data was retrieved. Climate model simulations were retrieved for historical and future global warming levels (GWL): 0.8, 1.5, 2.0, 2.5, and 3.0 °C. This retrieval process pulls the 30 year time window around the year (+/- 15 years) in which the global warming level is reached. A 30 year window is used to look at typical conditions over a long period centered on the global warming level.

Inland water was masked from the data to discard artificially high values over water that are not representative of the variable.

For percentile-based thresholds, a per-gridcell historical temperature threshold was calculated using the 0.8 GWL for each simulation. This value represents an extreme heat threshold unique to the historical conditions at each location.

This step differs between the two metrics. Each was calculated for each maximum temperature threshold (e.g. 95 °F, 100 °F, 95th percentile), simulation, and GWL. Absolute thresholds range from 85 to 115 °F and percentile-based thresholds from the 90th to the 99th percentile.

Heat wave frequency

For each threshold and streak duration (from 3 to 14 days) pair, the number of streaks of consecutive days exceeding the threshold in each year was counted. Streaks were non-overlapping (any single day can only belong to one streak, if any) but may be consecutive. For example, if the 3-day duration is selected, a 6-day streak that exceeds the temperature threshold would be counted as two 3-day streaks.

Heat wave length

For each threshold, the average length of streaks of consecutive days exceeding the threshold per year was calculated. Only streaks with a length of at least 3 days are included in this average. A streak that runs from one year into the next is split, and each part is counted in its own year.

The following metrics were calculated for each simulation across the 30-year period corresponding to each GWL: median, 10th percentile, 90th percentile. This yields a single value per metric per simulation.

For each GWL above the reference period (1.5 °C and higher) and for each metric (median, 10th percentile, 90th percentile), the historical baseline value was subtracted from the future GWL value. This difference represents the projected change in heat wave frequency or length relative to the historical period.

Next, we aggregated the gridded data by spatial unit (e.g. county, city, etc) by taking a median of all the grid cells within the boundary. A median is used to capture the typical number or length of heat waves within the area of interest without introducing bias from any anomalous grid cells.

Finally, the data was aggregated across simulations using a mean. The final data product is a table for each county containing the following variables:

Absolute variables

• Annual average frequency of heat waves, with its 10th and 90th percentile (for each GWL, threshold, and duration)
• Annual average length of heat wave, with its 10th and 90th percentile (for each GWL and threshold)

Relative change variables

• Change from reference period: annual average frequency of heat waves, with its 10th and 90th percentile (for each GWL, threshold, and duration)
• Change from reference period: annual average length of heat wave, with its 10th and 90th percentile (for each GWL and threshold)