Developing a Reliable Long-Term Record of Fire-Weather Monitoring Indices from a 200-yr Atmospheric Reanalysis Dataset

Our goal is to develop a reliable long-term, historical record of fire danger indices based on fire-weather variables such as near-surface temperature, wind speed, humidity, and precipitation from the 200-year NOAA-CIRES-DOE 20th Century Reanalysis dataset version 3 (20CRv3). Despite incorporating only surface and sea level pressure observations for its weather reconstruction, we show that local 3-hourly, daily-averaged, and weekly-averaged values of these variables from the 20CRv3 agree well with those in the more comprehensive and higher-resolution yet shorter ERA5 reanalysis, which incorporates satellite, aircraft, weather balloon, and surface meteorological observations. For the overlapping 1940-2015 period, this agreement is illustrated over North America for all months, with an emphasis on May to October, corresponding to the main western North America fire season. Local rank correlations of the 20CRv3 and ERA5 values are >0.80 for 2-meter air temperature, 10-meter wind speed, and 2-meter vapor pressure deficit. Another particularly appealing feature of using the 20CRv3 dataset is that, because it is an 80-member ensemble reanalysis, it provides time-dependent uncertainty estimates. These uncertainty estimates are found to be reliable. They indicate how far back in time the indices can be trusted before the uncertainties become too large due to fewer and less accurate available observations. Using the 20CRv3 and other observational data, the variability and trends of weather-based fire danger indices, including evaporative demand, are examined over North America and compared to known large wildfire occurrences from 1806 to present.