1. Introduction
From the fall through early spring, offshore winds, or what are commonly referred to as Santa Ana winds, occur over Southern California from the coastal mountains westward and from Ventura County southward to the Mexican border. These synoptically driven wind events vary in frequency, intensity, and spatial coverage from month to month and from year to year, thus making them difficult to categorize. Most of these wind events are associated with mild to warm ambient surface temperatures ≥ 18°C and low surface relative humidity ≤ 20%. However, during the late fall and winter months, these events tend to be associated with lower surface temperatures as a result of the air mass over the Great Basin originating from higher latitudes and other seasonal effects. There are a variety of ways to define a Santa Ana event through the analysis of local and synoptic-scale surface pressure and thermal distributions across Southern California (Raphael 2003). We view these offshore winds from a wildfire potential perspective, taking into consideration both the fuel characteristics and weather. As we have found, the index discussed herein provides a robust descriptor of both Santa Ana winds and the potential for wildfire activity. Used in conjunction with a mean sea level pressure (MSLP) map type, this is a powerful method for separating Santa Ana wind events from the more typical nocturnal offshore flows that occur throughout the coastal and valley areas (i.e., land breeze) during the year.
From 21 through 23 October 2007, Santa Ana winds generated multiple large catastrophic fires across Southern California (Moritz et al. 2010). Most notable was the Witch Creek fire in San Diego County, where wind gusts of 26 m s−1 were observed at the Julian weather station along with relative humidity values of ≈5%. However, high-resolution model simulations at 667 m showed that wind velocities were much higher in unsampled areas (Cao and Fovell 2016). This event became the catalyst for the development of a comprehensive Santa Ana wildfire potential index to better inform fire agencies, first responders, private industry, and the general public about the severity of an approaching event. This index could also help augment fire weather watches and red flag warnings from the National Weather Service by providing value-added information about an impending event.
The Predictive Services Unit, functioning out of the Geographic Area Coordination Center (GACC) in Riverside, California, is composed of several meteorologists employed by the USDA Forest Service. In 2009, Predictive Services began working on an index to categorize Santa Ana wind events according to the potential for a large fire to occur (Rolinski et al. 2011). This unique approach addresses the main impact Santa Ana winds can have on the population of Southern California beyond the causal effects of windy, dry weather. Following on, and improving upon this work, the Forest Service (through Predictive Services) collaborated during a three-and-a-half-year period with the San Diego Gas and Electric utility (SDG&E) and the University of California, Los Angeles (UCLA), to develop the Santa Ana wildfire threat index (SAWTI). This index employs a gridded 3-km model to not only assess meteorological conditions, but also incorporates an estimation of fuel moisture to determine the likelihood of rapid fire growth during Santa Ana winds.
The SAWTI domain covers the coastal, valley, and mountain areas of Southern California from Point Conception southward to the Mexican border. This area has been divided into four zones based in part on the different offshore flow characteristics that occur across the region (Fig. 1). Zone 4, which covers Santa Barbara County and was the last zone to be included into the index (thus the reason for the discontinuity within the sequential order of zones going from north to south), does not typically experience Santa Ana winds in the classic sense. Strong northwest-to-north winds in this zone can either precede a Santa Ana wind event or can occur independently (typically in the summer), which in the latter case are more commonly known as “sundowners” (Blier 1998). In both cases, these downsloping winds are common to the south slopes of the Santa Ynez Mountains, an east–west coastal range that runs parallel to, and a few miles inland from, the shoreline. Although not frequent, significant fire activity associated with these winds in this zone has occurred in the past, which is why this geographic area is now represented in the index. Santa Ana winds across zones 1 and 2 are primarily a result of offshore surface pressure gradients (locally and/or synoptically) interacting with the local terrain to produce gap winds through the Soledad Canyon, the Cajon Pass, and the Banning Pass (Hughes and Hall 2010; Cao and Fovell 2016). These winds also tend to precede the Santa Ana winds that occur across San Diego County by 12–24 h. Across zone 3, offshore winds take on a more “downslope windstorm” characteristic driven largely by the tropospheric stability (Cao and Fovell 2016). Other factors that led to the division of the zones were changes in terrain, National Weather Service Forecast Office boundaries, and local news media market areas. The SAWTI is more than a tool for meteorologists and fire agency managers to assess the severity of Santa Ana winds; it is also a tool for the general public to help better prepare for impending events that could lead to catastrophic fires. Therefore, the idea of displaying the product via zones keeps the index simple and easy to understand for all user groups. The following discussion centers around the assessment of fire potential related to Santa Ana winds, the methodology behind the weather and fuel components of the index, and its operational implementation.
Map of SAWTI zones. Inset shows SAWTI zones in reference to the state of CA. Letters denote locations of NDVI grassland sites with underlying topography shaded. Site names are provided in the lookup table to the right. County boundaries shown in red.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Map of SAWTI zones. Inset shows SAWTI zones in reference to the state of CA. Letters denote locations of NDVI grassland sites with underlying topography shaded. Site names are provided in the lookup table to the right. County boundaries shown in red.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Map of SAWTI zones. Inset shows SAWTI zones in reference to the state of CA. Letters denote locations of NDVI grassland sites with underlying topography shaded. Site names are provided in the lookup table to the right. County boundaries shown in red.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
2. Methodology
a. Large fire potential: Weather component
We define a large fire within the four SAWTI zones to be 100 ha. The potential for an ignition to reach or exceed this value depends on a number of components, for example, various meteorological and fuel conditions, suppression strategy, topography, accessibility, and resource availability. We achieved this threshold by employing a historical fire database that was constructed by Predictive Services. This database was assembled by collecting fire occurrence data (1990–2013) from all state and federal fire agencies within the confines of California. For example, some of the fire agencies include the USDA Forest Service, the Bureau of Land Management, the National Park Service, and the California Department of Forestry and Fire Protection (CALFIRE) to mention a few. This database contains information such as ignition date, acres burned, containment date, etc., and contains 32 683 records. The value of 100 ha was achieved by determining what the largest fire was for each day within the database and then taking the 95th percentile of all daily largest fires. The determination of this semiempirical threshold was also guided by decades of experience guiding coordinated attacks on wildfires throughout Southern California. Moreover, in most cases when this threshold is exceeded, the GACC becomes engaged in resource mobilization to assist in fire suppression. Current methods for evaluating fire potential include various indices from the National Fire Danger Rating System (NFDRS; Bradshaw et al. 1983) and from the Canadian Forest Fire Danger Rating System (CFFDRS; Preisler et al. 2008). The Fosberg fire weather index (FFWI) is one such metric and is a function of wind speed, humidity, and temperature with output values ranging from 0 to 100 (Fosberg 1978). While the FFWI may show elevated output values for a Santa Ana wind event, it can also show elevated values for any day therefore making it too generic for our purposes.
Relationship of large fire (≥100 ha) occurrence and relative size with respect to average wind speed and dewpoint depression across zone 1 between 1 Jun and 20 Sep from 1992 to 2012. Bubble size represents relative fire size.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Relationship of large fire (≥100 ha) occurrence and relative size with respect to average wind speed and dewpoint depression across zone 1 between 1 Jun and 20 Sep from 1992 to 2012. Bubble size represents relative fire size.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Relationship of large fire (≥100 ha) occurrence and relative size with respect to average wind speed and dewpoint depression across zone 1 between 1 Jun and 20 Sep from 1992 to 2012. Bubble size represents relative fire size.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
As in Fig. 2, but between 21 Sep and 31 Dec.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
As in Fig. 2, but between 21 Sep and 31 Dec.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
As in Fig. 2, but between 21 Sep and 31 Dec.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
b. Large fire potential: Fuel moisture component
1) Dryness level
The DL is a function of ERC and 
2) Live fuel moisture
The observed LFM is the moisture content of live fuels (e.g., grasses, shrubs, and trees) expressed as a ratio of the weight of water in the fuel sample to the oven dry weight of the fuel sample (Pollet and Brown 2007). Soil moisture as well as soil and air temperature govern the physiological activity, which results in changes in fuel moisture (Pollet and Brown 2007). LFM is a difficult parameter to evaluate because of the irregularities associated with observed values. For instance, samples of different species of native shrubs are normally taken twice a month by various fire agencies across Southern California. However, the sample times often differ between agencies and the equipment used to dry and weigh the samples may vary from place to place. In addition, sample site locations are irregular in their distribution and observations from these sites may be taken sporadically. This presents a problem when we attempt to assess LFM over the region shown in Fig. 1.
Apart from taking fuel samples, there are several ways of estimating LFM using meteorological variables, soil water reserves, solar radiation, etc. (Castro et al. 2003). In particular, we developed an approach to modeling the LFM of chamise or greasewood (Adenostoma fasciculatum), a common shrub that grows within the chaparral biome in Southern California and is particularly flammable because of its fine, needlelike leaves and other characteristics (Countryman and Philpot 1970; Fovell et al. 2016, manuscript submitted to Int. J. Wildland Fire). This strategy makes use of historically observed LFM data from 10 sampling sites across Southern California and soil moisture from the 40–100-cm layer (
3) Annual grasses Gag
Following the onset of significant wetting rains, new grasses will begin to emerge in a process called green-up. While the timing and duration of this process fluctuate from year to year, some degree of green-up usually occurs by December across Southern California. During the green-up phase, grasses will begin to act as a heat sink, thereby preventing new ignitions and/or significantly reducing the rate of spread among new fires. By late spring these grasses begin to cure with the curing phase normally completed by mid-June. In (2), 
We give 
Probability of fires ≥ 0.04 ha predicted by NDVI-derived 
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Probability of fires ≥ 0.04 ha predicted by NDVI-derived
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Probability of fires ≥ 0.04 ha predicted by NDVI-derived
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Relationship between NDVI and greenness.
Selected NDVI regressors.
We applied this model to the 21 sites in the four zones shown in Fig. 1. It is recognized that at some stations and times, the NDVI predictions are somewhat out of phase (i.e., the up and down ramps are too early or too late) with the observations, and the peaks are over- or underpredicted at different locations and times. The marked drought year of 2007 is clearly a problem at some locations, especially in zone 2. However, considering the fact that this is a simple universal model with only five regressors applied across Southern California, we believe it has shown adequate skill overall (Cao 2015).
c. Large fire potential: Weather and fuels
Comparison of 
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Comparison of
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Comparison of
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Average (left) 
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Average (left)
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Average (left)
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Average 
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Average
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Average
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
3. Operational SAWTI
a. Model configuration
The data ingested to compute the four-zone, 6-day LFP operational forecasts come from multiple sources at different temporal and horizontal resolutions ranging from hourly to daily, and from 3 to 12.5 km, respectively (Fig. 9). To reduce the exposure to error in fields with long accumulation periods, we sourced input variables for LFM and NDVI from the NLDAS-2 data (constructed using a land surface model in conjunction with assimilated observations and atmospheric model output). In contrast, hourly DFM and ERC values are predicted using offline models (Nelson 2000; Carlson et al. 2007; NFDRS) forced by WRF weather output.
Flowchart depicting operational LFP input models and datasets, derived variables, and the final LFP equation.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Flowchart depicting operational LFP input models and datasets, derived variables, and the final LFP equation.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Flowchart depicting operational LFP input models and datasets, derived variables, and the final LFP equation.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
DFM and ERC are calculated from meteorological variables predicted using WRF version 3.5 (Skamarock et al. 2008), run at 3- and 6-km horizontal resolution. We selected a WRF configuration that minimized errors with respect to near-surface temperature, winds, and dewpoint during Santa Ana wind events (Cao 2015; Cao and Fovell 2016). This configuration includes the simple WRF single-moment 3-class microphysics scheme (Hong et al. 2004), the GCM version of the Rapid Radiative Transfer Model (RRTMG) shortwave and longwave radiation schemes (Iacono et al. 2008), the MM5 Monin–Obukhov surface layer scheme, and the Asymmetrical Convective Model version 2 boundary layer scheme (Pleim 2007). The Noah land surface model (Tewari et al. 2004) with four soil layers was used in conjunction with the MODIS land-use dataset. Each operational WRF forecast dynamically downscales the 12-km-resolution 0000 and 1200 UTC North American Mesoscale Forecast System (NAM) 1–3.5-day forecasts to 3-km resolution. We use a two-way-nested WRF domain configuration consisting of a 3-km-resolution innermost domain nested within a 9-km-resolution outermost domain with 51 vertical levels. To extend the forecast out to 6 days, the 0.25°-resolution Global Forecast System (GFS) is downscaled using WRF to 6-km resolution. We use a two-way-nested WRF domain configuration consisting of a 6-km-resolution innermost domain nested within an 18-km outer domain and a 54-km outermost domain with 46 vertical levels. To help determine bounds and behavior of the SAWTI equations and place forecasts into some historical perspective, we dynamically downscaled the 32-km-resolution North American Regional Reanalysis (NARR; Mesinger et al. 2006) dataset to 3-km resolution using WRF over the historical period spanning January 1984–December 2013. We used a two-way-nested WRF domain configuration consisting of a 27-km-resolution outer domain, 9-km-resolution inner domain, and 3-km-resolution innermost domain with 51 vertical levels. WRF was integrated across 3.5-day periods with the first 12 h from each period discarded as spinup time.
b. Calculating SAWTI
1) Weather
Time periods over which 
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Time periods over which
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Time periods over which
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
2) Fuels
Recall that DL relates ERC and DFM to historical fire activity. To provide a DL forecast, DFM and ERC are computed across the spunup WRF forecast period. To avoid the potentially long spinup times required by DFM, the DFM must be initialized at each grid point across the WRF domain. Since a publicly available gridded observed DFM product does not exist, DFM is initialized using the previous day’s DFM forecast valid at the fourth hour of the current WRF forecast. The first 4 h of each WRF forecast are removed to allow for model spinup and to avoid contamination of DFM and ERC as a result of relatively unrealistic atmospheric inputs. Because of the need for these continuously spunup DFM time series, WRF forecasts must be uninterrupted. However, if any WRF forecasts are missed, DFM forecasts could be initialized using output from earlier WRF/DFM forecasts, which are archived for at least a month.
Quasi-observational data (NLDAS-2) are available for estimating LFM and NDVI using (3) and (5), respectively. The 22-day lagged soil moisture required for LFM is provided from the Noah land surface model output of the NLDAS-2 dataset. For NDVI, the latest NLDAS-2 output is used (typically a 5-day lag), which provides vegetation fraction, 2-m relative humidity, and soil moisture. Archived NLDAS-2 data are needed going back to the previous 1 September for cumulative precipitation. Both LFM and NDVI are regridded from the NLDAS-2 data at 12.5 km to the 3-km horizontal resolution, matching the WRF domain using bilinear interpolation, and are held constant across the 6-day forecast period. In contrast to weather that is calculated hourly, fuel conditions are calculated only at 1300 LST, representing fuel conditions for the entire day.
c. Public dissemination
Social science was incorporated during the early stages of the developmental process of SAWTI (Wall et al. 2014). The Desert Research Institute provided a social scientist to conduct an in-depth survey of five communities across Southern California. Much of the survey centered on questions regarding how the public obtains weather and fire information and their associated responses to that information. The results of the survey were used to help determine the type of information that would be presented in the product. In conjunction with the social science, historical weather and fuels data were correlated to historical fire occurrence records to develop index threat level categories. For example, for each SAWTI zone we compared daily FMC values along with daily 
Using historical fire occurrence data between 1992 and 2011, the relationship between binned FMC, 
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Using historical fire occurrence data between 1992 and 2011, the relationship between binned FMC,
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Using historical fire occurrence data between 1992 and 2011, the relationship between binned FMC,
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
The SAWTI has four threat levels that range from “marginal” to “extreme.” When Santa Ana winds are either not expected or will not contribute to significant fire activity, then a “no rating” is issued for that day. For example, it could be possible that if a strong Santa Ana wind event were to transpire after appreciable rains occurred or when fuels are wet, the event would be categorized as a no rating. For definitions of other threat levels, see Table 3. Tied to each threat level is a list of recommended actions suggested to the public to better prepare for an impending event. Examples include the following instructions: “Clean debris away from your house, charge your cell phone and make sure you have plenty of gas.” The list of recommended actions expands as the threat levels increase. This aspect of the product is critical, as it serves to link categories of severity with public awareness.
Categories of threat levels and their descriptions.
The product consists of an online web page (http://sawti.fs.fed.us) that displays a 6-day forecast of the above-mentioned categories for each of the four zones across Southern California (Fig. 12). A map of the region stands as the centerpiece of the page and graphically shows the categories that are colorized, ranging from gray (no rating) to purple (extreme). The product is issued once daily but can be updated more frequently as conditions warrant. The web page allows users to obtain more information such as viewing the latest weather observation from select stations when zoomed in on the map. The page will also display active and nonactive fires (via icons) on the map when such activity is present. Selecting one of these icons will provide the user with specific fire information such as acreage burned, percent contained, and links to more data. SAWTI also has a Twitter feed (https://twitter.com/sawti_forecast), where users are notified about changes in threat levels.
Online operational SAWTI product.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Online operational SAWTI product.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Online operational SAWTI product.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
The product was beta tested for a year prior to it becoming a public product in the fall of 2014. During the beta test phase, the index performed well in capturing all events that occurred during the fall of 2013 through the spring of 2014, which featured events that ranged from no rating to high. Several notable events occurred during this period: 16 January 2014 (Colby fire), 29 April–1 May 2014 (Etiwanda fire), and 13–14 May 2014 (the San Diego fires). Fire agencies that were granted access to the index during this time used the product to make critical decisions regarding the allocation and mobilization of shared fire resources prior to when these fires occurred. Specifically, the event that occurred on 13–14 May 2014 was especially notable because of the fact that the winds were unusually strong during this period, and that multiple large fires occurred as a result. Figure 13 shows a map of the fires across San Diego County, while Fig. 14 shows the SAWTI in beta test form for this event. The product was officially released to the public on 17 September 2014 via a press release and at an associated press conference. Since that time, the product has been used by local news media across the San Diego and Los Angeles metropolitan areas, as well as being shown on The Weather Channel.
Map of active fires (icons) on 14 May 2014 across San Diego County.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Map of active fires (icons) on 14 May 2014 across San Diego County.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Map of active fires (icons) on 14 May 2014 across San Diego County.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
SAWTI (in beta test) during 14–15 May 2014.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
SAWTI (in beta test) during 14–15 May 2014.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
SAWTI (in beta test) during 14–15 May 2014.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
d. Validation
Fire potential is very difficult to validate since our model is based on a conditional probability (i.e., getting an ignition). In addition, once an ignition occurs there are a number of human behaviors that cannot be predicted that can influence fire potential. For instance, if the SAWTI indicates a high likelihood of having a large fire for a particular Santa Ana wind event and one does not occur, it does not necessarily mean the model performed poorly. There may not have been an ignition during the event, or there may have been an ignition, but adequate fire-fighting resources were made available to be successful in suppressing the incident before the fire became large. There have been a few times where the index displayed a no rating and a large fire occurred, but this has been very rare.
Modeling fuel conditions accurately presents certain challenges. Regarding DFM, our ability to validate WRF DFM and ERC is limited given the sparse observations across this domain. Various Remote Automated Weather Stations (RAWSs) calculate DFM using measured atmospheric inputs including near-surface temperature, relative humidity, precipitation, and solar radiation. We validate WRF DFM and ERC across two years of the 30-yr historical period at 14 RAWSs (Fig. 15). These stations were selected so that at least three stations represent zones 1–3. Zone 4 has relatively fewer RAWSs reporting DFM and ERC measurements for the time period of interest; thus, only one station represents zone 4. At each RAWS location, the closest WRF grid cell with the smallest elevation difference was selected for validation. We show two example time series plots (Figs. 16 and 17), for the Goose Valley and Claremont RAWSs. At the Goose Valley RAWS site (Fig. 16), the WRF DFM and ERC output agrees well with RAWS measurements for most of the two years examined, with only slightly positive biases of 0.24 and 2.14 for 
RAWSs used to validate WRF DFM and ERC.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
RAWSs used to validate WRF DFM and ERC.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
RAWSs used to validate WRF DFM and ERC.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
RAWS (blue line) and closest WRF grid cell (orange line) time series of (top) 100- and (middle) 1000-h dead fuel moisture, and (bottom) ERC spanning January 2012–December 2013 for Goose Valley. WRF output coincides with RAWS 1300 LST measurements. Each plot is annotated with WRF output bias, RMSE, and the Spearman correlation.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
RAWS (blue line) and closest WRF grid cell (orange line) time series of (top) 100- and (middle) 1000-h dead fuel moisture, and (bottom) ERC spanning January 2012–December 2013 for Goose Valley. WRF output coincides with RAWS 1300 LST measurements. Each plot is annotated with WRF output bias, RMSE, and the Spearman correlation.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
RAWS (blue line) and closest WRF grid cell (orange line) time series of (top) 100- and (middle) 1000-h dead fuel moisture, and (bottom) ERC spanning January 2012–December 2013 for Goose Valley. WRF output coincides with RAWS 1300 LST measurements. Each plot is annotated with WRF output bias, RMSE, and the Spearman correlation.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
As in Fig. 16, but for the Claremont RAWS.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
As in Fig. 16, but for the Claremont RAWS.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
As in Fig. 16, but for the Claremont RAWS.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
WRF error statistics at each RAWS for time spanning January 2012–December 2013.
e. Climatology
The historical dataset described previously provides us with an unprecedented 30-yr climatology of the fuel and weather variables related to wildfires across the four SAWTI zones in Southern California. Having this dataset has allowed us first to create breakpoints within the raw SAWTI output necessary for the development of the four threat levels that are integral to the final public product. To do this, we correlated historical fire occurrence data with historical 
This unique dataset informs us about the historical significance the fuels, weather, and SAWTI events have had during the past 30 years. Having the ability to put past, but perhaps more importantly, forecasted SAWTI events into historical perspective helps inform the public and first responders about the nature and the characteristics of an impending event. For example, we can authoritatively state that the Santa Ana wind event that helped to spawn the Witch Creek fire (and served as the catalyst for the development of this index) was ranked as the highest event in the 30-yr dataset for zones 1 and 2.
As we continue to explore this dataset, we hope to gain a better understanding of the climatology of Santa Ana winds during the past three decades, including detecting and understanding interannual trends and cycles in event frequencies and strength. Figure 18 shows the number of days when Santa Ana winds occurred across zone 3 for the period spanning 1984–2013. This figure reveals a noticeable upward trend in the frequency of Santa Ana wind days during approximately the last 10 years, ending in 2013. Preliminary research shows that this long-term trend in frequency (possibly associated with a longer-term interannual cycle) coincides with a predominately negative phase of the Pacific decadal oscillation (PDO). Further investigation conducted in a future paper will seek to explore the causal mechanisms for this trend in frequency, as well as other trends in Santa Ana wind characteristics.
Number of Santa Ana wind days per rain year (1 Jul–30 Jun) for years spanning 1984–2014 (solid black line). Dashed line is a polynomial fit to the data, which helps to depict the longer time period trends.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Number of Santa Ana wind days per rain year (1 Jul–30 Jun) for years spanning 1984–2014 (solid black line). Dashed line is a polynomial fit to the data, which helps to depict the longer time period trends.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
Number of Santa Ana wind days per rain year (1 Jul–30 Jun) for years spanning 1984–2014 (solid black line). Dashed line is a polynomial fit to the data, which helps to depict the longer time period trends.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-15-0141.1
4. Summary and conclusions
As the wildland–urban interface (WUI) continues to expand across Southern California, the sources of ignition will increase, leading to a greater probability for large and destructive fires during Santa Ana wind events. This puts the public and firefighter safety at risk, thus the increasing need to categorize such events in terms of their effect on the fire environment.
Predictive Services’ initial efforts to categorize Sana Ana winds helped to provide the leadership and guidance necessary for the development of the SAWTI. Through the successful collaboration between the government, academia, and the private sector, high-resolution model data along with satellite-derived variables allowed us to incorporate fuel and weather data into the index on a gridded domain within Southern California. Challenges surrounding the assessment of fuel conditions include the difficulty in determining different fuel moisture parameters, which can sometimes result in a less accurate evaluation of fuel conditions. Further refinement of the model is needed to improve the overall output. However, during the beta testing process, the index performed very well with positive responses from the recipients of the preliminary output. Since its public unveiling, the index has been well received by the media and the fire community.
Our 30-yr dataset is unprecedented. Not only does it provide us with 30 years’ worth of fuel moisture data across Southern California (which is useful in relating fuel conditions with drought), it also gives us quantifiable outputs of average wind velocity, dewpoint depression, and the SAWTI itself. This allows us to put past and future Santa Ana events (magnitude, duration, and spatial coverage) into historical perspective, which is significant. Future studies in the climatology of such events can be conducted, leading to a better understanding of why certain trends exist.
Fire agencies and first responders, private industry, the general public, and the media now have a new operational tool that determines the severity of Santa Ana wind events. Furthermore, they will have a clearer understanding of the severity of an event based on the potential for large fires to occur. Specifically, a more effective media response will result in the general population (particularly those living within the WUI) being more proactive in its response to an impending event.
Acknowledgments
The authors thank Dr. Jim Means for his helpful suggestions and input into this project. We also appreciate the advice from Beth Hall, Tamara Wall, Mark Jackson, and Alex Tardy. The data used in this study were acquired as part of the mission of NASA’s Earth Science Division and archived and distributed by the Goddard Earth Sciences (GES) Data and Information Services Center (DISC). Major funding for this project was provided by SDG&E.
REFERENCES
Blier, W., 1998: The sundowner winds of Santa Barbara, California. Wea. Forecasting, 13, 702–716, doi:10.1175/1520-0434(1998)013<0702:TSWOSB>2.0.CO;2.
Bradshaw, L. S., Burgan R. E. , Cohen J. D. , and Deeming J. E. , 1983: The 1978 National Fire Danger Rating System: Technical documentation. General Tech. Rep. INT-169, Intermountain Forest and Range Experiment Station, U.S. Forest Service, 44 pp. [Available online at http://www.fs.fed.us/rm/pubs_int/int_gtr169.pdf.]
Cao, Y., 2015: The Santa Ana winds of Southern California in the context of fire weather. Ph.D. thesis, University of California, Los Angeles, 172 pp.
Cao, Y., and Fovell R. G. , 2016: Downslope windstorms of San Diego County. Part I: A case study. Mon. Wea. Rev., 144, 529–552, doi:10.1175/MWR-D-15-0147.1.
Carlson, J. D., Bradshaw L. S. , Nelson R. M. Jr., Bensch R. R. , and Jabrzemski R. , 2007: Application of the Nelson model to four timelag fuel classes using Oklahoma field observations: Model evaluation and comparison with National Fire Danger Rating System algorithms. Int. J. Wildland Fire, 16, 204–216, doi:10.1071/WF06073.
Castro, F. X., Tudela A. , and Sebastiá M. , 2003: Modeling moisture content in shrubs to predict fire risk in Catalonia (Spain). Agric. For. Meteor., 116, 49–59, doi:10.1016/S0168-1923(02)00248-4.
Chuvieco, E., Cocero D. , Riaño D. , Martin P. , Martìinez-Vega J. , de la Riva J. , and Pérez F. , 2004: Combining NDVI and surface temperature for the estimation of live fuel moisture content in forest fire danger rating. Remote Sens. Environ., 92, 322–331, doi:10.1016/j.rse.2004.01.019.
Clinton, N. E., Potter C. , Crabtree B. , Genovese V. , Gross P. , and Gong P. , 2010: Remote sensing–based time-series analysis of cheatgrass (bromus tectorum L.) phenology. J. Environ. Qual., 39, 955–963, doi:10.2134/jeq2009.0158.
Countryman, C. M., and Philpot C. W. , 1970: Physical characteristics of chamise as a wildland fuel. USDA Forest Service Research Paper PSW-66, 16 pp. [Available online at http://www.fs.fed.us/psw/publications/documents/psw_rp066/psw_rp066.pdf.]
Elmore, A. J., Mustard J. F. , Manning S. J. , and Lobell D. B. , 2000: Quantifying vegetation change in semiarid environments: Precision and accuracy of spectral mixture analysis and the normalized difference vegetation index. Remote Sens. Environ., 73, 87–102, doi:10.1016/S0034-4257(00)00100-0.
Fosberg, M. A., 1978: Weather in wildland fire management: the fire weather index. Proc. Conf. on Sierra Nevada Meteorology, Lake Tahoe, CA, Amer. Meteor. Soc., 1–4.
Hong, S.-Y., Dudhia J. , and Chen S.-H. , 2004: A revised approach to ice microphysical processes for the bulk parameterization of clouds and precipitation. Mon. Wea. Rev., 132, 103–120, doi:10.1175/1520-0493(2004)132<0103:ARATIM>2.0.CO;2.
Hughes, M., and Hall A. , 2010: Local and synoptic mechanisms causing Southern California’s Santa Ana winds. Climate Dyn., 34, 847–857, doi:10.1007/s00382-009-0650-4.
Iacono, M. J., Delamere J. S. , Mlawer E. J. , Shephard M. W. , Clough S. A. , and Collins W. D. , 2008: Radiative forcing by long-lived greenhouse gases: Calculations with the AER radiative transfer models. J. Geophys. Res., 113, D13103, doi:10.1029/2008JD009944.
Jin, Y., and Coauthors, 2015: Identification of two distinct fire regimes in Southern California: Implications for economic impact and future change. Environ. Res. Lett., 10, 094005, doi:10.1088/1748-9326/10/9/094005.
Mesinger, F., and Coauthors, 2006: North American Regional Reanalysis. Bull. Amer. Meteor. Soc., 87, 343–360, doi:10.1175/BAMS-87-3-343.
Moritz, M. A., Moody T. J. , Krawchuck M. A. , Hughes M. , and Hall A. , 2010: Spatial variation in extreme winds predicts large wildfire location in chaparral ecosystems. Geophys. Res. Lett., 37, L04801, doi:10.1029/2009GL041735.
Nelson, R. M., Jr., 2000: Prediction of diurnal change in 10-h fuel stick moisture content. Can. J. For. Res., 30, 1071–1087, doi:10.1139/x00-032.
Pleim, J. E., 2007: A combined local and nonlocal closure model for the atmospheric boundary layer. Part I: Model description and testing. J. Appl. Meteor., 46, 1383–1395, doi:10.1175/JAM2539.1.
Pollet, J., and Brown A. , 2007: Fuel moisture sampling guide. Utah State Office, Bureau of Land Management, Salt Lake City, UT, 32 pp. [Available online at http://www.wfas.net/nfmd/references/fmg.pdf.]
Preisler, H. K., Chen S. C. , Fujioka F. , Benoit J. W. , and Westerling A. L. , 2008: Wildland fire probabilities estimated from weather model-deduced monthly mean fire danger indices. Int. J. Wildland Fire, 17, 305–316, doi:10.1071/WF06162.
Raphael, M. N., 2003: The Santa Ana winds of California. Earth Interact., 7, doi:10.1175/1087-3562(2003)007<0001:TSAWOC>2.0.CO;2.
Rolinski, T., Eichhorn D. , D’Agostino B. J. , Vanderburg S. , and Means J. D. , 2011: Santa Ana forecasting and classification. 2011 Fall Meeting, San Francisco, CA, Amer. Geophys. Union, Abstract NH33B-1573.
Rothermel, R. C., 1972: A mathematical model for predicting fire spread in wildland fuels. USDA Forest Service Research Paper INT-115, 40 pp. [Available online at http://www.fs.fed.us/rm/pubs_int/int_rp115.pdf.]
Skamarock, W. C., and Coauthors, 2008: A description of the Advanced Research WRF version 3. NCAR Tech. Note NCAR/TN-475+STR, 113 pp., doi:10.5065/D68S4MVH.
Tewari, M., and Coauthors, 2004: Implementation and verification of the unified Noah land surface model in the WRF model. Preprints, Conf. on Weather Analysis and Forecasting/16th Conf. on Numerical Weather Prediction, Seattle, WA, Amer. Meteor. Soc., 14.2a. [Available online at https://ams.confex.com/ams/84Annual/techprogram/paper_69061.htm.]
Wall, T. U., Knox Velez A. L. , and Diaz J. , 2014: Wildfire readiness in the Southern California wildland-urban interface. Desert Research Institute Research Rep., 16 pp. [Available online at http://www.dri.edu/dripublications.]
White, M. A., Thornton P. E. , and Running S. W. , 1997: A continental phenology model for monitoring vegetation responses to inter-annual climatic variability. Global Biogeochem. Cycles, 11, 217–234, doi:10.1029/97GB00330.
