Evaluating the Ability of FARSITE to Simulate Wildfires Influenced by Extreme, Downslope Winds in Santa Barbara, California

Round 1

Reviewer 1 Report

This papers presents fire spread simulations for two cases, using the software FARSITE and FlamMap. This work compares results of simulations with observations, illustrating the difficulties of both models reproducing fire area, perimeter and shape. The paper is well organized and written. There are some points needing clarification and some of the figures can be improved.

- There is an extensive use of acronyms. A small acronyms dictionary would be useful for helping the reader.

- The sentence is lines 81-84 may be a bit confusing, because the authors refer that fire models describe the coupling and then they state that the models are uncoupled. I suggest rephrasing for better clarity. Also replace “flow” by “wind flow”

- In figure 3, please specify at which height from the ground are the wind values given

- Figures showing fire simulations (eg. Fig. 10) should include topography information as colour contours. Fire contours could be just lines. I think this would make the interpretation easier.

- If possible, I would suggest changing the colour code for elevation, ranging from cold to warm colours. The adopted colour code is not very intuitive for interpretation.

- The sentence in lines 209 and 210 is misleading as the simulations underestimate wind speed

- The measured wind speed is largely underestimated by the simulations. The authors should provide some explanation for this. Is it due to limitations in the simulation models? Is it due to convective effects not taken into account?

- Wind gust effects are localized effects that do not prevail over time. I don’t see the point of artificially increasing the wind speed for all simulations for taking into account gust effects.  

- Figure 5 is illustrative that increasing the wind speed is not solution for the problem. There is a downslope effect that is not taken into account and this may be due to the wind shifting direction due to terrain effects. Or maybe the basic rate of spread associated with the fuel models is not the correct one. Anyway, there is so much disagreement between the simulations and the observations that I think the authors should include in the paper some more reflection on this.

- Referring to the previous point, spotting may be the reason. The authors should provide some more information regarding the spotting model that they used. For example, I don’t understand how the spotting model is connected with the slope (lines 349, 350). Burning embers are transported by the wind, so the spotting direction should be directly related with the wind field

- Results in Fig 9 are very strange. Why spotting (launch and landing) almost never occur in the downslope side? The authors should provide some explanation on this.

- The differences between both models (FARSITE and FlamMap) are attributed to the raster versus vectorial character of the models, besides the fact that FlamMap considers constant properties and constant wind over time. Certainly there are other studies comparing the two approaches (raster versus vectorial) that could be mentioned. Also, I feel the differences should not be that big. It is not clear for me if, for comparison between these models, the wind field was taken as constant.

Author Response

We thank the reviewer for their attention and feedback to improve this manuscript. Our detailed reply to the reviewer’s comments are provided below. The original reviewer comments are in blue and our response is in black.

 

Reviewer #1:          

This papers presents fire spread simulations for two cases, using the software FARSITE and FlamMap. This work compares results of simulations with observations, illustrating the difficulties of both models reproducing fire area, perimeter and shape. The paper is well organized and written. There are some points needing clarification and some of the figures can be improved.

We appreciate the careful reading and helpful feedback to help improve our manuscript.

 

- There is an extensive use of acronyms. A small acronyms dictionary would be useful for helping the reader.

We thank the reviewer for this suggestion and agree that the addition or a table including all acronyms could be helpful to the reader. This was also suggested by another reviewer. The table has been added to the appendix and is first cited in the manuscript on line 62.

 

- The sentence is lines 81-84 may be a bit confusing, because the authors refer that fire models describe the coupling and then they state that the models are uncoupled. I suggest rephrasing for better clarity. Also replace “flow” by “wind flow”

We agree with the reviewer. We have re-worded the sentence describing uncoupled and coupled wildfire models, and have replaced “flow” with “wind flow” to provide further clarification. These changes are now shown in lines 94 to 96.

 

- In figure 3, please specify at which height from the ground are the wind values given

We thank the reviewer for suggesting this addition and have added the appropriate changes to the figure between lines 263 and 267. We have also added some discussion on the implications for height differences between lines 274 and 289.

 

- Figures showing fire simulations (eg. Fig. 10) should include topography information as colour contours. Fire contours could be just lines. I think this would make the interpretation easier.

We appreciate the suggestion; however, we believe the essential items for this figure are the final observed perimeter, reference roads, and perimeters (FARSITE) and arrival times (FlamMap). Thus, we combined these variables in Figure 10. Unfortunately, the addition of elevation contours made the figure more difficult to interpret, as did converting fire contours to lines.

 

- If possible, I would suggest changing the colour code for elevation, ranging from cold to warm colours. The adopted colour code is not very intuitive for interpretation.

We thank the reviewer for this suggestion and have stretched the original color bar. Lower elevations are represented by a larger range in green, rather than white and blue, and higher elevations are represented by more shades of red and brown, rather than gray. We changed this in Figures 2 and 8.

 

- The sentence in lines 209 and 210 is misleading as the simulations underestimate wind speed

The authors agree that this statement was misleading and thank the reviewer for the careful reading. This sentence was improved and now reads, “During the Painted Cave fire, simulated winds were typically underestimated by less than 2 m/s at KSBA (Fig. 4f).”

 

- The measured wind speed is largely underestimated by the simulations. The authors should provide some explanation for this. Is it due to limitations in the simulation models? Is it due to convective effects not taken into account?

The wind speed is underestimated in WRF and WN during our wildfire and Sundowner case studies. Comparisons were performed between a remote automatic weather station (RAWS, in the case of the Sherpa Fire) and a National Weather Service station (NWS, in the case of the Painted Cave fire). RAWS and NWS have different protocols for instrumentation. Besides, RAWS are usually placed by the Forest Service in locations that are critical for strong winds, such as canyons and passes. These topographic features are not properly represented in a 1 km grid resolution model. Moreover, parameterization choices have been based on sensitivity tests performed with a few case studies. We have shown that winds in coastal areas are sensitive to these parameterizations more than in the slopes of the mountains (Duine et al. 2019). We have shown in previous works (Duine et al. 2019, Carvalho et al. 2020) that WRF overestimates winds in some locations and underestimate winds in others, and we have attributed these discrepancies to local topographic features that are not accurately represented in the model. This study was not designed to discuss parameterization in WRF. Following the reviewer’s comments, we improved the explanation about WRF representation of winds in lines 274 to 301.

 

- Wind gust effects are localized effects that do not prevail over time. I don’t see the point of artificially increasing the wind speed for all simulations for taking into account gust effects. 

We agree that wind gusts are spatially and temporally variable. Perimeters were underestimated in our simulations using the gridded wind speed without a gust factor applied, (see Figure 5a and 7a). Gusts are usually the result of enhanced turbulence and can accelerate fire spreading rates. WRF does not simulate gusts. Therefore, the underlying idea of using gust factor is to incorporate the effect of gusts in FARSITE and obtain more realistic fire perimeters. Moreover, gust factor has been used to estimate gusts during other extreme wind events in southern California, such as Santa Anas (Mitchell 2013, Fovell and Cao 2017, Cao and Fovell 2018). While our approach is not perfect, the application of a gust factor to WN did increase agreement between the observed and simulated fire perimeters in both case studies, and particularly with the Sherpa Fire. This was further discussed between lines 317 to 321.

Future work could focus on determining gust factors during multiple Sundowners, and applying the gust factor to gridded wind data to more accurately capture the known spatiotemporal variability of wind gusts. This was added to the discussion on future work in lines 601 to 606.

 

- Figure 5 is illustrative that increasing the wind speed is not solution for the problem. There is a downslope effect that is not taken into account and this may be due to the wind shifting direction due to terrain effects. Or maybe the basic rate of spread associated with the fuel models is not the correct one. Anyway, there is so much disagreement between the simulations and the observations that I think the authors should include in the paper some more reflection on this.

The authors somewhat disagree that the wind speed is not the solution for the problem. We believe there are two problems with the initial simulation (without a gust factor applied). The first problem is that the fire spread was too slow; the perimeter one hour after ignition aligns well, but all following perimeters are vastly underestimated in FARSITE. Applying a gust factor significantly helped the spread and produced perimeters closer in size to the observations.

The second problem is the fire spread direction. We agree that the fire did not spread in the same direction as was observed between three and four hours after ignition, although there was good agreement in the preceding hours. In the manuscript, we state that firefighting efforts partially limited the eastward spread (this now has a citation to the local fire chief who provided this information). In the revised manuscript, we have added a statement that local wind shifts or terrain effects not evident in observations from RHWC1, WRF, or WN may have contributed to the westward spread. This is in lines 408 to 410.

 

- Referring to the previous point, spotting may be the reason. The authors should provide some more information regarding the spotting model that they used. For example, I don’t understand how the spotting model is connected with the slope (lines 349, 350). Burning embers are transported by the wind, so the spotting direction should be directly related with the wind field

We thank and agree with the reviewer for the suggestion to add more information on the spotting model and how it relates to our findings. In the manuscript, we added that FARSITE uses Albini’s (1979) equations for spotting and calculated the maximum distance an ember can travel based on wind speed, topography, and ember properties. Furthermore, wind speed is only considered horizontally in FARSITE and is assumed to increase logarithmically with height about the 6.1 m input winds. In nature, spotting is created by turbulence, that, to be properly described, requires a tridimensional wind field (vertical velocity is also important). Embers are blown in the direction of the wind and are thus directly related to the wind field; however, the slope is related to the amount of spotting. Embers can burn out before reaching terrain when that terrain is located at lower elevations relative to the initial position. This is further explained between lines 164 to 167 and 496 to 503.

 

- Results in Fig 9 are very strange. Why spotting (launch and landing) almost never occur in the downslope side? The authors should provide some explanation on this.

The authors agree that an explanation should be provided for the spotting (or lack thereof) phenomenon shown in Figure 9. While we could not determine the exact equation or term that restricted spotting during downslope fire spread, we propose that embers may be launched during this time, but may extinguish mid-air as density and volume is lost (see the above reply). Thus, new fires would not begin and this would not count as spotting in FARSITE or FlamMap as the launched embers never landed. This is explained in the manuscript from line 496 to 503. We also added investigation of the spotting model as future work in the conclusions.

 

- The differences between both models (FARSITE and FlamMap) are attributed to the raster versus vectorial character of the models, besides the fact that FlamMap considers constant properties and constant wind over time. Certainly there are other studies comparing the two approaches (raster versus vectorial) that could be mentioned. Also, I feel the differences should not be that big. It is not clear for me if, for comparison between these models, the wind field was taken as constant.

We thank the reviewer for this comment and agree that the explanation of the differences between the simulated perimeters and arrival times should be expanded. We revised lines 551 to 557 to focus on the differences between the wildfire models; FARSITE propagates a single wildfire using Huygen’s principle whereas FlamMap MTT calculates fire behavior at each grid cell individually. Furthermore, each raster model wildfire simulators use different methods and equations, and thus we cannot make a general statement comparing the raster versus vectorial character of the models.

 

Author Response File: Author Response.docx

Reviewer 2 Report

Understanding wildfire risk is especially important in the wildland-urban interface, where advances in evacuation planning and emergency management preparedness will increase resilience to these natural hazards. Two case studies in coastal Santa Barbara County were selected to simulate wildfires significantly influenced by extreme fire weather conditions associated with downslope winds known as Sundowners. FARSITE and FlamMap were used and the results were compared, which shown that FARSITE has the potential to provide reliable perimeters for simulating wildfires in Santa Barbara. However, both FARSITE and FlamMap were unable to simulate realistic fire perimeters. The paper is fine and it can be accepted for publication.

1. The wildland fire is influenced by many parameters, such as wind, slope, fuel. How to link the physical model (CFD) to the empirical model (FARSITE)?
2. Both FARSITE and FlamMap were unable to simulate realistic fire perimeters How to solve this problem in future work?

 

 

Author Response

We thank the reviewer for their attention and feedback to improve this manuscript. Our detailed reply to the reviewer’s comments are provided below. The original reviewer comments are in blue and our response is in black.

 

Reviewer #2:          

Understanding wildfire risk is especially important in the wildland-urban interface, where advances in evacuation planning and emergency management preparedness will increase resilience to these natural hazards. Two case studies in coastal Santa Barbara County were selected to simulate wildfires significantly influenced by extreme fire weather conditions associated with downslope winds known as Sundowners. FARSITE and FlamMap were used and the results were compared, which shown that FARSITE has the potential to provide reliable perimeters for simulating wildfires in Santa Barbara. However, both FARSITE and FlamMap were unable to simulate realistic fire perimeters. The paper is fine and it can be accepted for publication.

We appreciate the careful reading and feedback to improve our manuscript.

 

1. The wildland fire is influenced by many parameters, such as wind, slope, fuel. How to link the physical model (CFD) to the empirical model (FARSITE)?

We agree that the distinctions and connection between the physical and empirical models needs to be explained in greater depth in the manuscript. In the original manuscript, we state that FARSITE and FlamMap are uncoupled, semi-empirical wildfire spread models. In the revised manuscript, we expand on the differences between uncoupled and coupled models and the limitations for both types. We state that the choice to use empirical models such as FARSITE and FlamMap was due to the simplicity and quick simulation run time that would be beneficial for operational purposes. While physical models may produce fire perimeters more similar to those observed, they have greater computational expense and are rarely used in an operational setting. The changes regarding this suggestion are reflected between lines 94 and 105.

 

2. Both FARSITE and FlamMap were unable to simulate realistic fire perimeters How to solve this problem in future work?

We thank the reviewer for the suggestion to explain further implications for future work regarding one of the main issues in FARSITE identified in our manuscript. These have been added in lines 601 to 607.

 

Author Response File: Author Response.docx

Reviewer 3 Report

This manuscript focused on simulation of wildfire spread with FARSITE numerical model. Two wildfire cases were studied and wildfire perimeters were analysed for different scenarios. The ability of FARSITE model in simulating fire spread was also studied.

The topic and results of this manuscript are important and worth publishing. However, it requires some modifications before being considered for publication. Here is my comments and suggestions to improve the quality of the manuscript:

1-The Authors are suggested to include nomenclature and define all the abbreviations used in the text so that they can be easily accessed by the readers.

2-During wildfire-wind interaction, wind is also influenced by fire which consequently affects fire spread. This increase of wind by fire is reported as "fire-wind enhancement" phenomenon in the literature. Can the effects of this phenomenon be observed by the applied numerical model (FARSITE and FlamMap)? The authors are suggested to discuss this phenomenon in the introduction section and state the strength or limitation of the applied model in capturing the influence of fire on the wind.

3-Does FARSITE consider the effects of buoyancy force of the fire on the simulations? if so, how it is modelled in the software? Components of buoyancy force are also affected by the terrain slope condition. Is this effect considered by the software in the simulations? It is also recommended that more details of fire modelling are included in the modelling section.

 

Author Response

We thank the reviewer for their attention and feedback to improve this manuscript. Our detailed reply to the reviewer’s comments are provided below. The original reviewer comments are in blue and our response is in black.

 

Reviewer #3:

This manuscript focused on simulation of wildfire spread with FARSITE numerical model. Two wildfire cases were studied and wildfire perimeters were analysed for different scenarios. The ability of FARSITE model in simulating fire spread was also studied.

The topic and results of this manuscript are important and worth publishing. However, it requires some modifications before being considered for publication. Here is my comments and suggestions to improve the quality of the manuscript:

We appreciate the careful reading and feedback to improve our manuscript.

 

1-The Authors are suggested to include nomenclature and define all the abbreviations used in the text so that they can be easily accessed by the readers.

We thank the reviewer for this suggestion and agree that the addition of a table including all acronyms could be helpful to the reader. This was also suggested by another reviewer. The table has been added to the appendix and is referenced in the manuscript on line 62.

 

2-During wildfire-wind interaction, wind is also influenced by fire which consequently affects fire spread. This increase of wind by fire is reported as "fire-wind enhancement" phenomenon in the literature. Can the effects of this phenomenon be observed by the applied numerical model (FARSITE and FlamMap)? The authors are suggested to discuss this phenomenon in the introduction section and state the strength or limitation of the applied model in capturing the influence of fire on the wind.

While the effects of the wildfire-wind enhancement can be important for wildfire spread, uncoupled wildfire models such as FARSITE and FlamMap are unable of simulating this phenomenon. We chose to acknowledge this limitation and not utilize coupled wildfire models (which would have this ability) to provide fast assessment of wildfire spread, potentially for operational purposes. We agree that these limitations should be expanded in the paper. This limitation was briefly mentioned in the original manuscript, and the revised version expands on the explanation in lines 94 to 105.

 

3-Does FARSITE consider the effects of buoyancy force of the fire on the simulations? if so, how it is modelled in the software? Components of buoyancy force are also affected by the terrain slope condition. Is this effect considered by the software in the simulations? It is also recommended that more details of fire modelling are included in the modelling section

FARSITE only considers wind speed horizontally, where winds increase logarithmically with height above the 6.1 m (20 ft) wind. Buoyancy is only considered within the spotting equations (Albini 1979) and use ember size, vertical wind speed profile, and upwind surface topography to calculate the duration and travel distance of embers (Finney 1998).

We agree that the details of the model should be further discussed in the model. We have added material between 164 and 168 to further explain the spotting equations behind FARSITE. Additionally, we added a larger discussion regarding the observed spotting patterns and explanation of these limitations in the Spotting Limitations Section (3.3), from line 496 to 503.

Author Response File: Author Response.docx