A new method based on global climate model pressure gradients was developed for identifying coastal high-wind fire weather conditions, such as the Santa Ana Occurrence (SAO). Application of this method for determining southern California Santa Ana wind occurrence resulted in a good correlation between derived large-scale SAOs and observed offshore winds during periods of low humidity. The projected change in the number of SAOs was analyzed using two global climate models, one a low temperature sensitivity and the other a middle-temperature sensitivity, both forced with low and high emission scenarios, for three future time periods. This initial analysis shows consistent shifts in SAO events from earlier (September–October) to later (November–December) in the season, suggesting that SAOs may significantly increase the extent of California coastal areas burned by wildfires, loss of life, and property.
 California coastal region wildfire weather conditions typically occur during the fall season (September to December) prior to the winter rains, and when an inland high pressure and an offshore low pressure set up a strong pressure gradient, resulting in heated air mass, and high offshore winds with low humidity. These conditions are known locally as Santa Ana winds in southern California and Diablo winds in northern California, but are more generally defined as foehns. Such weather conditions have a long history of being associated with high winds that spread fires. During foehn occurrences, hot downslope winds may exceed 30 ms−1, are warmed by adiabatic compression at a rate that can exceed 9°C km−1, and have very low relative humidity, making these conditions the most conducive for the spread of fires.
 One of the earliest records of a southern California foehn was in 1859 [Ryan, 1991, W. A. Tompkins, Santa Barbara's incredible hottest day, undated]. A ship operated by the U.S. Coast Survey was anchored near Santa Barbara; ship records indicated temperatures near 30°C in the late morning and no unusual weather. By 1 pm gusty northerly winds developed from the Santa Ynez peak, accompanied by a sharp rise in temperatures, and by 2 pm temperatures rose to a record setting 56°C with strong northerly winds. By 5 pm the ship thermometer indicated temperatures had dropped to 50°C, and by 8 pm the ship temperature was at 25°C.
 Fires linked to such windy, hot, and dry offshore flow conditions have been observed throughout coastal California and have resulted in significant loss of life and property, especially in regions where development has encroached on wilderness interfaces. California Santa Ana and Diablo winds impact hundreds of miles along the coastal mountains. The 1991 Oakland fire re-ignited and spread because of Diablo wind conditions. In 2003, the Cedar fire near San Diego spread from 2 to 125 km2 in four hours, owing to the presence of Santa Ana winds. As development on coastal mountain regions expands and man-made fire ignitions continue to occur, fire weather loss and damage will rise.
 It is widely thought that large-scale mechanisms dominate the Santa Ana and Diablo conditions, which are modulated by local effects of the sea breeze and topography. Coastal California offshore flow is governed by several mechanisms, including the local diurnal land-sea temperature differential, upwelling, and large-scale atmosphere and ocean dynamics associated with pressure systems. Santa Ana conditions occur when a north-south pressure gradient is present and may be enhanced by a southward moving trough, or a high-pressure ridge moving into central California [Ryan, 1991].
 The northerly Santa Ana winds have been identified by Castro et al.  using Quicksat satellite observations. Hu and Liu  also used remote sensed Quicksat data to observe oceanic thermal and biologic responses to Santa Ana winds. Raphael  catalogued a 33-year data set of Santa Ana surface meteorological observations and identified days when a surface high-pressure system existed over the Great Basin simultaneously with a surface low-pressure system offshore of southern California, suggesting a relationship between SAOs and these large-scale patterns. Analyses of large-scale weather variables have been diagnosed, along with other variables, to understand the onset and duration of fire weather in the western U.S. [Westerling et al., 2003, 2004]. Recently, Conil and Hall , using a high-resolution regional climate model forced by reanalysis data, performed a cluster analysis to determine three wind regimes in southern California: Santa Ana wind flow, onshore flow, and a common northwesterly flow, with an average of 2 to 5 Santa Ana wind days per month between October and March.
 In section 2 a method for identifying SAOs from large-scale variables is described. In section 3, an analysis of the sensitivity of SAOs to climate change is presented, and concluding remarks are made in section 4.
 The conditions necessary for the establishment of an SAO are defined here as the existence of a stationary high pressure system situated over the Great Basin and/or extending further east, and an offshore Pacific low southwest of Los Angeles with at least a 20 hPa pressure difference between these centers [Raphael, 2003], and the surface-level humidity at the local fire risk area below 40% [Sergius and Huntoon, 1956]. Figure 1 shows the western U.S. and an SAO search domain (boxed area) with coordinates 33°N125°W × 39°N113°W, along with the European Center for Medium Range Weather Forecast (ECMWF) - ReAnalysis 40 years (ERA40, http://www.ecmwf.int/products/data/archive/descriptions/e4/) daily 6 pm surface pressure fields for 19–22 December 1999, when strong Santa Ana winds caused a wildfire to spread, resulting in extensive property loss.
 The above search domain is sufficient for this analysis. A larger search domain will yield additional information on teleconnections that influence the local phenomena. However, this study has taken on an intentional regional analysis to only identify SAO signals and determine their distribution for a range of projected scenarios.
 In this study, evaluation of Santa Ana conditions is based on observations and atmosphere-ocean general circulation model (AOGCM) output for the historical period 1965–1994. The 20 hPa pressure difference criterion was applied to AOGCM historical and projected output to determine the percent change in occurrence of projected to historical simulated SAOs per month. Daily or monthly time series of AOGCM output cannot be directly compared with observations, as AOGCM projections are not predictions. Hence, historical 30-year mean-monthly climatologies for 1965–1994 were defined using pressure fields, and were compared to SAO climatologies based on the observed wind direction (northerly) and humidity (40 percent or less). The observations include the National Climatic Data Center's Surface Airways Observations (electronic resource, 1992) for California, Arizona, and the Western United States; the ERA40 850 hPa geopotential height, surface pressure and specific humidity fields; and weather station temperature, wind, and humidity data [Raphael, 2003] for 1965 to 1994. Data from two AOGCMs, the NOAA Geophysical Fluid Dynamics Laboratory version 2 (GFDLv2 [Delworth et al., 2006]) and the DOE/NCAR Parallel Climate Model (PCM [Washington et al., 2000]) were used here, because they are the two models selected for the California Governor's Climate Change Initiative [Schwarzenegger, 2005; Cayan et al., 2005], and this study is a contribution [Miller and Schlegel, 2005]. The GFDL output included daily pressure and humidity fields, but the PCM output included only daily pressure fields.
 Our earlier climate sensitivity study [Hayhoe et al., 2004] indicated that low- and middle-temperature sensitivity AOGCMs, with high and low Intergovernmental Panel on Climate Change (IPCC) emission scenarios, would result in a four-member ensemble of future outcomes that provide a somewhat conservative envelope for likely future outcomes of three 30-year climatologies. In a similar fashion, the present study uses the low- and mid-range temperature sensitivity PCM and GFDL models forced by the IPCC Special Report on Emission Scenarios (SRES) high (A2) and low (B1) emissions [Intergovernmental Panel on Climate Change, 2000] for three 30-year mean-monthly climatologies, 2005–2034, 2035–2064, and 2070–2099. These scenarios represent the range of IPCC non-intervention emissions futures, with atmospheric CO2 concentrations reaching approximately 550 ppm (B1) to 830 ppm (A2) by 2100.
 In this study we constructed the 1965–1994 mean-monthly SAO climatology - using weather station wind, temperature, and humidity data [Raphael, 2003] - and compared this to the 1965–1994 SAO climatology using the GFDL pressure and humidity data and the PCM pressure-only data. This resulted in a good fit between GFDL-derived SAO days, using pressure and humidity, and the observed number of high offshore-wind, low-humidity SAO days. Figure 2 indicates that the climatological September, October, November, December, January, and February GFDL-derived pressure- and humidity-defined SAO days were 1.2, 3.5, 6.5, 7.1, 8.7, and 5.8, respectively, representing 112%, 76%, 93%, 81%, 70%, and 69% of observations, respectively. The 850 Hpa geopotential height fields were also tested in place of pressure, with nearly identical results.
 Since daily humidity data were available in the GFDL historical and A2 output, but not for the GFDL B1 projection, or any of the PCM output (historical, A2, and B1), all of which were required as part of the Governor's Climate Change Study [Cayan et al., 2005], an approximation based exclusively on pressure was used. Using this approximation, the trends match well, with September having the lowest number of SAO days, an increasing trend toward the maximum in December, and a decreasing trend for the period during January and February (Figure 2). The GFDL pressure gradients with and without humidity for 1965–1994 indicate that omitting humidity increases the number of September-to-February SAOs identified to about twice those constrained with pressure and humidity. This over-estimation by the pressure-only approach is acceptable, because the SAO climate change sensitivity is a relative change (ratio) of the percent of days with pressure-driven high offshore winds and, as a first approximation, is assumed near-linear. This represents a consistent over-estimation for both the historical and projected climatologies, and is summarized in the next section.
 SAO sensitivity (projected/historic) results for the three 30-year climatologies (2005–2034, 2035–2064, 2070–2099) are presented in Figures 3 and 4. Figure 3 was calculated with surface pressure and humidity from the GFDL A2 projections and the GFDL baseline historical simulation, and Figure 4 shows all four projected outcomes (GFDL A2, B1 and PCM A2, B1) based on pressure-only. Note that the mean-seasonal ratio between the number of SAO days with pressure-only compared to those with pressure and humidity is approximately 1.9 for all cases, verifying that for projected periods, the pressure-only case remains about twice the pressure and humidity case.
Figure 3 indicates a shift in GFDL A2-derived SAOs, with the September and October number decreasing and the December number increasing. By the end of the century, the simulated December number increases to 117 percent of the simulated historical. The GFDL and PCM pressure-only SAOs for each month are shown in Figures 4a–4d. GFDL A2 and B1 outcomes indicate that the warmest fall month, September, has an increase in SAO days in the early part of this century, followed by a decrease at the end of the century. The PCM A2 and B1 outcomes suggest decreases in SAO during September throughout this century. More importantly, the strongest historical SAO month, December, decreases for all PCM cases during 2005–2034, but significantly increases for all cases during 2070–2099.
 These initial findings indicate striking differences between early century and late century, and high-emission and lower-emission SAO sensitivities. The most significant and consistent results shown in Figures 4a–4d are shifts in SAOs from earlier to later in the year that are greater under A2 relative to B1 and GFDL relative to PCM, with no clear changes in the January to February SAOs.
 In terms of fire weather threats in the future, this study suggests that there may be SAO increases during critical dry periods, especially late in the season, leading to more extensive wildfire. The climate change summary chapter of the California Governor's Climate Change Report [Cayan et al., 2005] states there is no evidence from the IPCC projections to suggest that the Mediterranean seasonal precipitation regime in California will change. The Report does indicate that for some projections, summer precipitation changes incrementally and in some cases decreases. These findings strengthen the likelihood for increased fire weather risk resulting from increased vegetation drying in future summers, combined with shifts in the frequency of Santa Anas.
4. Summary and Conclusion
 In this paper, we describe an initial analysis of pressure-gradient-derived Santa Ana Occurrences. The results suggest an end-of-century shift in SAOs from September and October to November and December that are greater under a high-emission scenario (A2) and a mid-temperature sensitivity model (GFDL), relative to a low-emission scenario (B1) and a low-temperature sensitivity model (PCM). This initial study begins to investigate an important climate impact on society, ecology, and economy. The impacts of wildfires spreading during SAO conditions will be further enhanced as population and housing development encroaches on coastal mountain wilderness regions. Future research will establish the sensitivity of teleconnected patterns and the significance of low frequency modes of variability.
 Support for this study provided by the California Environmental Protection Agency as a contribution to the Climate Change Science Report to the Governor. We thank Katharine Hayhoe of the Department of Geosciences, Texas Technical University and Mary Tyree of the Climate Research Division, Scripps Institution of Oceanography for AOGCM data support.