WRF simulation over complex terrain during a southern California wildfire event



[1] In October 2007, the largest wildfire-related evacuation in California's history occurred as severe wildfires broke out across southern California. Smoke from these wildfires contributed to elevated pollutant concentrations in the atmosphere, affecting air quality in a vast region of the western United States. High-resolution numerical simulations were performed using the Weather Research and Forecast (WRF) model to understand the atmospheric conditions during the wildfire episode and how the complex circulation patterns might affect smoke transport and dispersion. The simulated meteorological fields were validated using surface and upper air observations in California and Nevada. To distinguish the performance of the WRF in different geographic regions, the surface stations were grouped into coastal sites, valley and basin sites, and mountain sites, and the results for the three categories were analyzed and intercompared. For temperature and moisture, the mountain category has the best agreement with the observations, while the coastal category was the worst. For wind, the model performance for the three categories was very similar. The flow patterns over complex terrain were also analyzed under different synoptic conditions and the possible impact of the terrain on smoke and pollutant pathways is analyzed by employing a Lagrangian Particle Dispersion Model. When high mountains prevent the smoke from moving inland, the mountain passes act as active pathways for smoke transport; meanwhile, chimney effect helps inject the pollutants to higher levels, where they are transported regionally. The results highlight the role of complex topography in the assessment of the possible smoke transport patterns in the region.

1. Introduction

[2] Wildland fires frequently threaten life and property in the fire-adapted ecosystems of California (CA) in the United States (U.S.). Wildfires are associated with substantial increases in the concentration of gaseous pollutants, such as carbon monoxide, nitrogen oxides, ozone, volatile organic compounds, and particulate matter [Phuleria et al., 2005]. Aerosols that result from smoke emitted by wildland fires can be lifted into the free atmosphere and undergo long-range transport, even playing a role in the Earth's climate system [Haywood and Boucher, 2000; Wang et al., 2006].

[3] The fire season in CA usually starts around the middle of May and ends in the end of October when cooler temperatures and larger rainfall amounts become more common. However, most of the largest and most damaging of fires occur in the autumn when Santa Ana winds frequently occur [Keeley et al., 2009]. Between 20 and 23 October 2007 over two dozen wildfires broke out across southern CA, ranging from Santa Barbara County to the U.S.-Mexico border. Many of the wildfires spread quickly, aided by dry, westerly Santa Ana winds preceded by severe drought conditions. Over the following 16 days these wildfires destroyed nearly 1500 homes, burned over 500,000 acres of land, caused 9 deaths and 85 injuries, and forced over 900,000 people to be evacuated (the largest evacuation in the history of CA) [Karter, 2008; Keeley et al., 2009; Sutton et al., 2008]. Smoke from these wildfires degraded air quality in southern CA and long-range transport of smoke over Central CA and Nevada was observed during the period.

[4] In order to accurately simulate the transport and dispersion of smoke from wildfires, an adequate representation of the meteorological conditions in the area of interest is vital. When predicting smoke transport in southern CA, meteorological simulations must resolve the impacts of the Pacific Ocean and some of the sharpest terrain gradients in the continental U.S., both of which are challenging for any meteorological model. Both observational [Niccum et al., 1995; Tanrikulu et al., 2000; Zhong et al., 2004] and numerical modeling [Burk and Thompson, 1996; Lin and Jao, 1995; Seaman et al., 1995; Stauffer et al., 2000] studies have made efforts to represent the impact of complex flow patterns in this region on air pollution. These studies highlighted the importance of local-scale circulations associated with land-ocean contrasts and orographic forcing. Other investigations have revealed how these local-scale circulations can interact with synoptic-scale flow to influence the regional distribution of air pollutants [Bao et al., 2008; Beaver and Palazoglu, 2009; Kim and Stockwell, 2008; Lu and Turco, 1996]. It is widely acknowledged that synoptic-scale subsidence and the associated temperature inversion are key factors in episodes of poor summertime air quality in this region. In some areas of CA, such as the Central Valley and the L.A. Basin, topography acts to trap pollutants, which exacerbates the local air quality problems [Lu and Turco, 1996; Niccum et al., 1995; Tanrikulu et al., 2000], while pollutants that are transported above the elevated inversion to the free atmosphere can become sources for long-range transport, affecting air quality in areas of the Intermountain West [Blanchard et al., 1999; Langford et al., 2010; Poulos and Pielke, 1994].

[5] An accurate simulation of the transport and dispersion of smoke from wildfires in southern CA requires a meteorological model that can resolve complex interactions between synoptic flows, thermally driven circulations, and interactions between elevated inversions and local boundary layer structure. We hypothesize that a numerical simulation of meteorological conditions during wildfire events that is capable of resolving these multiscale flow patterns will, when coupled with a particle dispersion model, provide insight into impact of the complex southern CA terrain on the transport and dispersion of smoke. The October 2007 wildfire event provides a unique opportunity to test this hypothesis. This paper simulates the meteorological conditions using the Weather Research and Forecast (WRF) [Skamarock et al., 2005] model and examines the performance of the model by comparing the simulation results with meteorological observations. Several previous studies have used data from CA to validate model simulations, but these verification studies were confined to small area coverage (e.g., Central Valley or L.A. Basin), a short time period with typical summertime synoptic condition, or a small amount of observational data [Bao et al., 2008; Kim and Stockwell, 2008; Michelson, 2008]. The October 2007 southern CA wildfire event lasted for more than 2 weeks, covered a large region, and encompassed multiple synoptic flow regimes and regional flows such as Santa Ana winds, allowing a comprehensive testing of WRF model's ability in simulating the interactions of synoptic flows with regional and local circulations in regions of highly complex topography.

[6] The model setup and a description of the study area are presented in section 2. The simulated meteorological conditions are validated against surface stations and rawinsonde observations in section 3. Analysis of near-surface wind patterns over complex terrain is presented in section 4. Finally, section 5 presents a summary and conclusions.

2. Model Setup and Case Environment

2.1. Model Setup

[7] The WRF model, developed by the National Center for Atmospheric Research (NCAR), is a state-of-science, nonhydrostatic mesoscale meteorological model that has been used for weather forecasting and weather related research around the world. WRF has options for physical parameterizations suitable for a broad spectrum of applications.

[8] For this study, the WRF model version 3.0 was used. It was configured with three one-way nested grids, at horizontal grid spacings of 36, 12, and 4 km and 49 vertically stretched sigma levels. The innermost 4 km grid covered central and southern CA and southern Nevada, while the outermost 36 km domain encompassed the western U.S. and the eastern edge of the Pacific Ocean (Figure 1). The model was initiated at 0000 UTC on 15 October and run continuously through 0000 UTC 31 October. Multiple shifts in the synoptic pattern occurred during this time period, and the wildfire outbreak occurred between 20 and 23 October, making it an ideal time to investigate the variability of near-surface flow patterns and their potential impact of these flow patterns on smoke transport and dispersion. Figure 2 displays satellite imagery captured on 24 October 2007, showing the widespread fire activity and the extent of smoke in southern CA and over the Pacific Ocean at this time.

Figure 1.

Modeling domains used in this study. The grid spacing for the outer (d01), middle (d02), and inner (d03) domains are 36 km, 12 km, and 4 km, respectively.

Figure 2.

Image captured by NASA's Terra satellite showing southern California during the fire episode on 24 October 2007. Red pixels indicate fire activity.

[9] We defined a study period starting on 15 October and ending on 31 October in order to investigate the antecedent and concurrent meteorological conditions for the wildfire outbreak on 20–23 October. The 2 week long simulation period encompasses a variety of synoptic patterns and enables a broader evaluation of the WRF model performance for this time period; it also avoids introducing discontinuities into the meteorological simulations to facilitate future smoke dispersion analysis that is planned to follow this work. Additionally, the areal coverage of the modeling domain is considerably larger than the area directly affected by the wildfire, which permits observational data both from the region of interest and from nearby locations to be used for model validation.

[10] The parameterizations used for this study are summarized in Table 1. The initial and boundary conditions for the WRF simulations were prescribed using the 3 hourly 32 km North American Regional Analysis (NARR) [Mesinger et al., 2006]. The sea surface temperature was specified from the National Centers for Environmental Prediction (NCEP) Eta analysis and was kept constant throughout the simulation period. Damping was employed at the model top (100 hPa) to prevent reflection from the upper boundary.

Table 1. WRF Physics Schemes Used in the Present Study
 Physics Scheme
MicrophysicsLin et al. scheme [Lin et al., 1983]
CumulusKain-Fritsch (none for domain 3) [Kain and Fritsch, 1993]
Planetary boundary layer (PBL)MYJ [Janjic, 1996, 2002; Mellor and Yamada, 1982]
Surface layerMYJ [Janjic, 1996, 2002; Mellor and Yamada, 1982]
Land surfaceNoah [Chen and Dudhia, 2001]
RadiationRRTM for longwave radiation [Mlawer et al., 1997]; Dudhia for shortwave radiation [Dudhia, 1989]

2.2. Case Environment

[11] Southern CA is characterized by sharply varying topographical features such as hills, mountains, plains and mesas (see Figure 3). The Sierra Nevada mountain range, extending from northwest to southwest, lies in the east of the state and to the west of the Great Basin. Mt Whitney (elevation of 4,421 m), lying within Sierra Nevada, is the highest peak in the contiguous U.S. and represents the sharpest topography gradient of any location in North America. The Central Valley is located between the Sierra Nevada and the coastal mountain ranges. Transverse ranges, the only west-east oriented ranges in this region, separate the Central valley and the Mojave Desert. The region is bordered on the west by the Pacific Ocean, which promotes variations in climate regimes and vegetation between coastal and inland areas in southern CA [Durrenberger, 1968].

Figure 3.

Map with the meteorological stations over the innermost domain. The lines AA', BB', and CC' are the horizontal positions of vertical cross section discussed in the text.

[12] For 7 days prior to the wildfire outbreak, the west coast of the U.S. was under the influence of a strong upper level ridge. Beginning on 15 October, a deep trough off the southwest coast of Canada and an upper level ridge over Mexico contributed to moderate (12–15 m s−1) 500 hPa west-southwesterly winds over the U.S. west coast. As the trough over Canada propagated to the east, the upper level ridge over Mexico extended to the north, and by 18 October a strong 500 h Pa geopotential height gradient was in place between the two regions (Figure 4a). Winds during this time were generally zonal with strong (35–40 m s−1) westerly 500 hPa winds prevailing over most of the western U.S. By 21 October, the ridge had moved off to the east, and the west coast was dominated by a strong upper level trough, bringing 500 hPa northwesterly winds over most of CA and Nevada. By 23 October, an upper level ridge arrived off the coast of southern CA and helped develop moderate (5–15 m s−1), north-northeasterly 500 hPa winds over CA. This upper level ridge dominated the west coast for the next 3 days, generating 500 hPa easterly flow over southern CA (Figure 4b). During the morning of 26 October another upper level ridge began to form over the Baja Peninsula while a trough began to form over the Pacific Northwest. These two systems contributed to 500 hPa southwesterly winds with wind speeds ranging from 10 m s−1 near central and southern CA to 25 m s−1 in northern CA. By 27 October a cutoff low over the Pacific Northwest developed and moved south to the coast of central CA (Figure 4c), and light to moderate (5–10 m s−1) southwesterly winds dominated southern CA.

Figure 4.

The (left) 500 hPa height and (right) wind speed fields at 00Z (1600 PST) on 3 days (18, 23, and 27 October) from the NARR data. The shading indicates wind speed (m s−1), and the contours are isoheights (m).

3. Simulation Results and Validation

[13] Because the accuracy of smoke transport and diffusion simulations produced by a dispersion model is dependent upon the accuracy of the input meteorological information, a detailed comparison between simulated and observed meteorological fields is both necessary and useful. For local-scale smoke transport and dispersion modeling applications, it is important that the simulations reproduce the near-surface meteorological conditions that are more directly impacted by local-scale thermal and/or orographic forcing. On a regional scale, it is crucial that the simulation reproduces the structure and evolution of the observed mixed-layer evolution and winds within and above the mixed layer [Banta et al., 2004; Pielke and Segal, 1986].

3.1. Model Evaluation: Surface Meteorological Fields

[14] Simulated surface meteorological parameters from the 4 km grid spacing innermost domain were validated against hourly observational data. The data from 71 surface weather stations in southern and central CA and Nevada were obtained from the National Climatic Data Center (NCDC), which combines surface meteorological data from the National Weather Service (NWS) first order, ASOS and COOP networks (see Figure 3). Scatterplots of simulated and observed temperature, water vapor mixing ratio, and wind speed at the surface for the entire study period are shown in Figure 5. The simulated surface temperature exhibits no clear overall bias, but there is a tendency for the model to produce slightly higher temperatures when observed temperatures are low and slightly lower temperatures when observed temperatures are high. The difference between observed and simulated water vapor mixing ratio is minimum for low values; dry departure is obvious for medium and high mixing ratio values. A scatterplot of the observed and simulated wind speed reveals a large spread in the data with no clear tendency for the model to produce lower or higher wind speeds. The large spread evident in this analysis of the surface winds was also found in a 3 km grid spacing simulation near Mexico City during the MILAGRO campaign [Zhang et al., 2009]. Many of the differences between model and observations can be attributed to the inability of the 4 km model grid to adequately resolve the fine-scale processes that affect the observed winds in complex terrain.

Figure 5.

Simulated and observed (top) hourly temperature (°C), (middle) mixing ratio (g kg−1), and (bottom) wind speed (m s−1) at all surface meteorological stations.

[15] Because smoke transport and dispersion in southern CA strongly depends upon meteorological conditions at different elevations, the model performance in regions delineated by different elevations is investigated. This was achieved by grouping the 71 surface stations in Figure 3 into three categories based on their location and base elevations: coastal sites, valley and basin sites, and mountain sites. For each category, the mean and standard deviation (STD) of the simulated and observed values and bias and correlation coefficients (R) were computed and the results are summarized in Table 2.

Table 2. Statistics for Observed and Simulated Surface Temperature, Mixing Ratio, and Wind Speed
  • a

    N is the number of observed data points.

Wind Speed (m s−1)
Temperature (°C)
Mixing Ratio (g kg−1)

[16] The STD of observed surface temperature was larger for the mountain category and smaller for the coastal sites. In contrast to temperature, the STD of observed surface water vapor mixing ratio was the lowest in the mountainous category and highest for the coastal sites. Generally speaking, these results indicate that mountain sites exhibited colder conditions with more variable temperatures and lower moisture during the study period. The coastal sites were warmer with more homogeneous temperatures and more moisture variability. The observed temperature and water vapor mixing ratio for the valley/basin category fell between these two categories. For observed surface wind speed, the largest STD was found in the mountain category and the smallest at the valley/basin sites.

[17] There were warm (positive) temperature biases for the coastal category (0.39 K) and the valley/basin category (0.35 K), and a cold (negative) bias (−0.22 K) for the mountain category. These biases are comparable to observation errors. For water vapor mixing ratio, there was a wet (positive) bias of 0.5 g kg−1 for the mountain category, and dry (negative) biases of −0.52 g kg−1 and −0.99 g kg−1 for the valley/basin and coastal categories, respectively. Wind speed biases were all negative and the values were generally small except perhaps for the valley/basin category with a bias of −0.5 m s−1. For all three categories, the correlation coefficients were higher for temperature and water vapor mixing ratio than for wind speed, which is in agreement with the scatterplot presented in Figure 3. For temperature and water vapor mixing ratio, the mountain category had the highest correlation coefficients between the simulated and the observed values, which was followed by the valley/basin category, with the coefficient for the coastal category somewhat lower. For wind, the correlation coefficients for the three categories were very similar. And the value (0.614–0.631) is comparable to that (0.59) from the previous model study over Mexico City [Zhang et al., 2009].

3.2. Model Evaluation: Upper Level Meteorological Fields

[18] Standard twice daily rawinsonde observations at 00Z (16 PST) and 12Z (0400 PST) from Oakland (OAK), San Diego (NKX) and Mercury (DRA) (Figure 3) were employed to evaluate the simulated boundary layer and free atmospheric conditions. The data were obtained from the University of Wyoming Weather Web (http://weather.uwyo.edu/upperair/sounding.html). The first two stations, one in the northern part of the model domain (OAK) and one in the southern part (NKX), are located in coastal areas according to the categorization in section 3.1, although OAK is farther away from the Pacific Ocean than NKX and therefore would expected less frequent influence by marine air. The DRA station (elevation ∼1.1 km above mean sea level or MSL) is more representative of inland mountainous areas.

[19] The time variation of the upper air meteorological conditions during the simulation period was investigated using time-height cross sections at the three sites (Figures 6, 7, and 8). At NKX, there was subsidence warming associated with the building of an upper level ridge on 20 (the day before a series of wildfires broke out), 24 and 29 October, with cooling on other days (Figure 6). The subsidence warming contributed to the development of a strong elevated inversion that is known to correlate with pollution events [Boucouvala and Bornstein, 2003]. While the WRF simulation reproduced the observed subsidence warming, there were differences in the base elevation of the subsidence warming. OAK and DRA profiles showed similar patterns except for the timing of the transition, which is attributable to their different geographical locations.

Figure 6.

Time-height cross section of the (left) observed and (right) simulated potential temperature (°C) at (top) Oakland, (middle) San Diego, and (bottom) Mercury. White space indicates missing data.

Figure 7.

Time-height cross section of the (left) observed and (right) simulated specific humidity (g kg−1) at (top) Oakland, (middle) San Diego, and (bottom) Mercury. White space indicates missing data.

Figure 8.

Time-height cross section of the (left) observed and (right) simulated vector wind at (top) Oakland, (middle) San Diego, and (bottom) Mercury. White space indicates missing data.

[20] Figure 8 shows that a shift from initially westerly winds to northeasterly winds occurred on 22 October both in the observation and in the simulation. For 22–23 October, the stronger simulated low-level winds at NKX indicate that the model produced a stronger-than-observed Santa Ana wind. In the horizontal wind field (not shown), Santa Ana wind maximum in the simulation appears more northward compared with observations, so the low-level wind in the model wind profiles show a stronger wind that does not appear in the observed soundings. Westerly onshore flow coincides with increased water vapor mixing ratios (from beginning to around 21 October) while dry warm inland offshore winds decreased the water vapor mixing ratios for later dates, which is evident from the soundings at OAK and NKX (Figure 7). The inland site DRA exhibited less variation in water vapor mixing ratio. The simulated water vapor mixing ratio at DRA, however, still exhibited a diurnal variation with higher values during the day and lower values at night (Figure 7). Overall, the simulated water vapor mixing ratio was slightly higher than what was observed, which was the opposite of the bias indicated by the scatterplot of surface water vapor mixing ratio presented in Figure 3.

[21] The mean and STD of the differences between the simulation and observations as a function of height are shown in Figure 9. At OAK, the simulated potential temperature was slightly overestimated below 700 m, with the maximum differences appearing near 400 m, and underestimated between 1000 m to 4000 m. Simulated specific humidities were lower that the observations below 1000 m and higher above 2000 m. Simulated wind speeds were, in general, overpredicted by about 1 m s−1 in the lowest 1000 m. Simulated wind direction was more easterly than observed by about 20° and the departure gradually decreased with height to became very small above 3000 m. The simulated temperatures and specific humidity were cooler and higher, respectively, from the bottom to the top of the profile at NKX, with the largest errors occurring near the surface. Simulated wind speed was overestimated in the lowest 1000 m and underestimated above this level, which was similar to OAK. Also similar to OAK, simulated wind direction at NKX was more easterly below 2000 m while the differences became smaller aloft. The deviation varied substantially with height at the two coastal sites, but was relatively uniform with height at DRA. At DRA, the profiles of simulated temperature and specific humidity were cooler and slightly moist at all levels compared to the observations. Simulated wind speed was overestimated by nearly 1 m s−1 in the lowest 2000 m, and underestimated at higher levels. Wind direction differences were generally small at all levels, although it is noteworthy that the STD was larger below 2500 m, which was not seen for the two coastal sites.

Figure 9.

The mean and standard deviation of the differences between simulated and observed vertical profiles of (a) temperature, (b) specific humidity, (c) wind speed, and (d) wind direction at three upper air locations: (left) Oakland, (middle) San Diego, and (right) Mercury.

[22] Mixed-layer height (MH), which plays a primary role in determining the height to which surface atmospheric constituents will rise, is an important input parameter for air pollution modeling. In Figure 10, the model-derived MHs are compared with those from radiosonde measurements at 0000 UTC. The MH was determined using the gradient method, which identifies the level at which the variation of potential temperature with height changes from nearly constant with height to consistently increasing with height. The simulated MH varied between 900 and 2000 m at OAK and NKX, and between 2000 and 2500 m at DRA. The simulated MHs were within 20% of the observed values and the correlation coefficient between the observed and the simulated MHs across all sites and over the entire simulation period was 0.955. There was a slight underestimation of MH at DRA.

Figure 10.

The observed and simulated mixed-layer height during the simulation period. The observed values are computed from 25 0000 UTC radiosonde soundings.

[23] The above comparison helps to establish the qualitative and quantitative differences between the WRF simulation and observations for this event. This analysis can be used as a means of assessing smoke transport and dispersion simulations that employ the simulated meteorological fields as input, and to establish whether differences between observed and simulated smoke distributions can be attributed to errors in the meteorological model or to the representation of smoke within the transport and diffusion model.

4. Wind Flows in Complex Terrain and Their Potential Influence on Smoke Transport

[24] The validation exercise completed in section 3 provides us with some insight into the ability of the WRF simulation to reproduce the observed meteorological conditions in southern CA at and before the time of the southern CA wildfires in October 2007. In order to address how the simulated flow fields might impact the transport and diffusion of smoke in different synoptic conditions, we will now analyze simulated wind flows at the surface and aloft for three different synoptic patterns that occurred during the 16 day simulation. Figure 11 depicts the 1600 UTC surface and 1.5 km MSL level simulated winds under contrasting synoptic conditions which will be hereafter referred to as: onshore flow (17 October), strong Santa Ana flow (23 October), and weak offshore flow (25 October).

Figure 11.

The simulated horizontal wind vector (a, c, e) surface and (b, d, f) 1.5 km MSL on 17 October (Figures 11a and 11b), 23 October (Figures 11c and 11d), and 25 October (Figures 11e and 11f).

[25] On 17 October, under the influence of a north-south gradient in geopotential height (see similar pattern in Figure 4a), prevailing simulated winds were from the northwest aloft (Figure 4b), nearly parallel to the coastline. The southwesterly winds were somewhat enhanced crossing over the southern Sierra Nevada, the Traverse Range and the Coastal Range, due possibly to downslope flows associated with mountain waves [Durran, 1990]. Compare to winds aloft that was relatively uniform, the surface flow field was much more complex with sharp changes in wind speed and direction and areas of convergence and divergence. There was weak onshore flow penetrating inland, until it encountered the steeply sloping terrain (Figure 11a). Two convergence zones were evident in the simulation at this time: one north and west of Los Angeles near the Newfall pass (the location of Newfall, Cajon, and Banning passes are noted in Figure 3); the other located in the east L.A. Basin. Weak onshore flow to the west, westerly winds coming from the Sierra Nevadas, and the terrain blocking all appear to contribute to the formation of these convergence zones. Weak winds occurred inside the Central Valley and the L.A. Basin, a result of terrain blocking. Winds were stronger on the eastern slopes of the Sierra Nevada and southern slopes of the Traverse Range. A number of factors may have contributed to the development of these strong downslope winds including downward water vapor mixing of higher momentum air aloft and thermally driven flow by regional-scale pressure gradients [Zhong et al., 2008]. The relatively strong downslope flows were oriented perpendicular to the Newfall and Cajon passes; at the Banning pass, however, a component of the flow is oriented parallel to the pass, and the horizontal map suggests that flow through the pass was occurring (Figure 11a). Vertical cross sections through each pass (Figure 12) support the assertion. In this situation, we would not expect there to be substantial transport of smoke through the two northern passes, either from the coast to inland or inland to the coast. However, at the Banning pass, the cross section indicates strong flow through the pass at the surface and aloft, which suggests that smoke transport through the pass could have occurred on this date, had there been a fire present at that time.

Figure 12.

Potential temperature and wind vector on the vertical profile at the transect of (a) Newhall pass, (b) Cajon Pass, and (c) Banning Pass on 1200 17 October.

[26] On the 23rd, a high-pressure system was in place over the Great Basin (see Figure 4 over 500 mb) and the prevailing winds at 1.5 km MSL level were from the northeast. Strong Santa Anna winds were apparent through the three mountain passes and with generally weaker winds to the west (Figures 11c and 11d). However, at the northern Newfall pass, very strong surface winds were apparent that extended from the coastline over the ocean. An area of nearly calm surface winds formed in the west L.A. Basin. In this synoptic situation, we would expect smoke to the east and northeast of the mountain passes to be transported into the coastal area, and smoke near the coast to be transported offshore.

[27] On the 25th, weak synoptic forcing contributed to light winds from the southeast over the land. The winds are slightly stronger and more organized at 1.5 km (Figure 11f), and light and highly variable at the surface (Figure 11e). Offshore winds were found through all three passes and the winds were somewhat stronger at the passes than at lower elevations, but the weak overall flow and generally weak flow through the passes would not be expected to contribute to efficient ventilation of smoke in the region.

[28] These analyses highlight the impact of steeply sloping, complex terrain in southern CA on the potential transport of smoke and other surface pollutants under different synoptic patterns. The sensitivity to both wind speed and wind direction is highlighted in this analysis, and the potential impact of these characteristics on the transport and diffusion of smoke and other near-surface pollutants is evident. This sensitivity is particularly evident for the “onshore flow” pattern (Figures 11a, 11b and 12). The mountain barrier and undulating land would tend to limit the potential for smoke and pollutant transport inland, with the three mountain passes being the most likely avenues for particles to escape. This analysis agrees with Poulos and Pielke's [1994] findings in a study of L.A. pollutants traveling to Arizona under southwesterly flow. The convergence zone they observed near the L.A. Basin and identified as a potential location for the accumulation of pollutant, was also noticeable in our study.

[29] In order to understand how the various circulation patterns might affect smoke transport and dispersion, we have used the WRF flow fields to drive a Lagrangian particle dispersion model (LPDM). The LPDM used is FLEXPART, which is successfully coupled with WRF to simulate the transport and dispersion of particles on local and regional scales [Fast and Easter, 2006]. Three representative flow patterns discussed above were examined: onshore flow (17 October), Santa Anna wind (23 October) and weak offshore flow (25 October). For each case, particles were released from the surface to 700 m above ground continually over a 24 h period. The release position is in agreement with the geographic location of Santiago Fire (see Figure 3), which is the longest-lasting fire during the October 2007 Southern California Wildfire Outbreak. Particles are then tracked in space and time as they were transported and dispersed in the atmosphere.

[30] Figures 13, 14, and 15 show the time evolution of the horizontal and vertical distributions of the particle concentration. As is discussed earlier, when high mountains prevent the smoke from moving inland, the mountain passes act as active pathways for smoke to advect and disperse. Under the influence of onshore flow (Figure 13), the flow moves southeastward mainly through the Banning pass and the low-elevation area southward (0400 UTC). When the wind direction is favorable, some of the particles also move through the Cajon Pass (1200 UTC). Except for the fire origin area, enhanced concentration is also found stemming from the Cajon Pass and along the pathway toward the U.S.-Mexico border. Besides moving through the mountain passes, the growth of PBL will aid the plume to escape. As is shown from the vertical structure of the plume (Figure 13f), by 20 00 UTC (1300 local time), the boundary layer grows deep and mixing the plume vertically to nearly 3 km, reducing surface concentration.

Figure 13.

(a–c) Horizontal and (d–f) vertical distributions of particle concentration under the influence of onshore flow pattern that occurred on 17 October 2007. The line in Figures 13a–13c indicates the position of the vertical cross section in Figures 13d–13f. The red lines in Figures 13d–13f are potential temperature contour at 2° intervals.

Figure 14.

Similar to Figure 13 but for Santa Ana flow pattern that occurred on 23 October 2007. The two lines in Figures 14a–14c indicate the positions of the two vertical cross sections.

Figure 15.

Similar to Figure 13 but for weak offshore flow pattern that occurred on 27 October 2007.

[31] Compared to the onshore flow patterns that transport the particles to the interior, the Santa Ana wind transports the plume quickly offshore (Figures 14a14c). Besides the difference in transport direction, the plume spread out more rapidly from the source with much larger horizontal extension in the Santa Ana wind scenario compared to that of the onshore flow. Four hours after the release, the leading edge of the plume has reached over 200 km away from the source area. The speed of the transport and dispersion of the plume is relatively steady and 8 h later it has traveled another 200 km. The higher concentration region progresses southeastward and mainly along the lower part of the plume. As we examine the cross section along the enhanced concentration we can find that the high concentration is packed inside the lowest levels, within 0.5 km from the sea surface (Figures 14e and 14f). The transport mechanism for lower and upper level plume differs. Upper part of the plume travels west-northwestward, induced by the upper level prevailing wind, rising to a higher elevation (Figures 14g14i). Unlike the onshore flow pattern, the daytime PBL growth only influences the vertical extension of the small remaining particles inland, while over the ocean, the plume is limited to 0.5 km which remains relatively steady during the day (Figure 14f).

[32] For the weak offshore wind scenario, smoke generally transports northwestward, following the upper level wind; a fraction extends further northwestward to the ocean, aided by the above mentioned gap flow through the Newfall pass (Figures 15a15c). Due to the blocking effect of the high mountains, the smoke is constrained to the western side of the coastal range. Meanwhile, the week wind degrades the ventilation. The enhanced concentration is found for a relatively long time to blanket the L.A. Basin. Before the day-time development of PBL, the plume height is around 2.5 km as seen from the cross section (Figures 15d and 15e). As it is not high enough to overcome the mountains south of the Central Valley, the plume finds its way to progress at lower elevation (1200 UTC). Aided by the growth of PBL in the afternoon, the plume was ejected to about 4 km by the chimney effect, approaching the Central Valley from aloft (2000 UTC).

5. Conclusions

[33] WRF model simulations were carried out for regions of Southern and Central California from 15 to 30 October 2007 to assess the ability of WRF in simulating the complex meteorological conditions over the region and investigate the potential influence of the regional and local circulations on the transport and dispersion of smoke from the October 2007 Southern CA wildfires. The simulated meteorological fields were compared to observations from a network of surface and upper air stations in CA and Nevada. To distinguish WRF performance in different geographic regions, the surface stations were grouped into coastal sites, valley and basin sites, and mountain sites, and the results for the three categories were analyzed independently. The differences in the observations among the three groups were reproduced by the WRF simulation. The standard deviation of the observations and simulation results were similar, except for temperature where the simulated STD was smaller than observed. A comparison between the simulated and observed surface variables in the three categories revealed a negative temperature bias for the mountain category and positive biases elsewhere, a positive water vapor mixing ratio bias for the mountain category and negative biases elsewhere, and negative wind speed biases for all three categories. Correlation coefficients were higher for temperature and water vapor mixing ratio than for wind speed, and the mountain category generally exhibited higher-correlation coefficients than the coastal category.

[34] Comparisons with vertical profiles of observed and simulated potential temperature, water vapor mixing ratio, and wind speed revealed that the WRF successfully reproduced the overall temporal and vertical distribution of these variables. A cold, wet bias was evident at most of the sounding locations, and aboveground wind speeds were generally higher in the model than in observations. Simulated wind directions were, on average, about 20° too far to the east, compared to the observations. Overall, the model simulated water vapor mixing ratio was slightly higher than what was observed. Simulated mixed-layer heights were slightly underestimated, but differences were within 20% of the observed values. Correlation coefficient between the observed and the simulated mixed layer heights across all sites and over the entire simulation period was 0.955. A comparison between the WRF simulations and observations provides a means of assessing whether errors in smoke transport and dispersion simulations that employ these simulated meteorological fields as input can be attributed to errors in the meteorological model or to the representation of smoke within the transport and diffusion model.

[35] Taking advantage of the 2 week long simulation time, wind flow over complex terrain under different synoptic conditions was analyzed. Finally, WRF-FLEXPART simulations was used to illustrate the possible transport and dispersion patterns of fire smoke over the region under the influence of three major circulation patterns. Potential smoke transport and dispersion from the mountain passes into eastern CA and Nevada and to the west over the ocean were identified; meanwhile, chimney effect (or mountain venting) helps inject the pollutants to higher level to escape from terrain blocking, where advection can transport and disperse the pollutants downwind, affecting regional air quality. The results highlighted the important role of the region's complex topography in the determination of potential smoke and pollutant transport patterns. This analysis of flow patterns can assist in the development of abatement strategies and provide policy guidance for reducing the adverse health impacts of wildfires on the public.


[36] The authors would like to thank Jerome Fast for providing the FLEXPART code and for his help on running the code. This research is supported by the multiagency Joint Fire Science Plan Project 9-1-04-1 and by the USDA Forest Service Northern Research Station under agreements 07-JV-11242300-049 and 09-JV-11242306-089.