A Multiscale Machine Learning Approach to Study Extreme Fire-weather Regimes in Santa Barbara, California, USA

Abstract:

Santa Barbara County is prone to fast spreading wildfires that pose high risk to the coastal communities. The most extreme fire-weather regime in the region is associated with Sundowner winds, which are late afternoon-to-early morning episodes of offshore, northerly gusty downslope winds on the southern slopes of the Santa Ynez Mountains. Sundowner winds show a high degree of spatiotemporal variations, which makes forecasting these fire-weather conditions very difficult. Being able to identify regimes of extreme fire-weather situations can bring substantial benefits to medium-range weather forecasts and emergency management operations.

This work employs a multi-scale approach to link the local climatology of Sundowner winds with large-scale circulation patterns. A 30-yr high spatiotemporal downscaling dataset performed with the WRF model (1 km horizontal grid spacing, hourly, 1 September 1987 to 31 August 2017) is used to identify Sundowner winds regimes. This data set is also used to calculate the Canadian fire weather index. The ERA-5 reanalysis is used to identify large-scale circulation patterns leading and during extreme fire-weather conditions associated with Sundowner winds. Cluster and self-organized maps analyses are used to link large-scale modes of variability, synoptic patterns and high-resolution extreme fire-weather conditions in Santa Barbara.