VI International Conference on Forest Fire Research D. X. Viegas (Ed.), 2010

WRF-Fire wildfire modeling in the test area of Harmanli,

Nina Dobrinkova Institute of Informationand Communication Technologies – Bulgarian Academy of Sciences, Acad. G. Bonchev str. Bl. 2, 1113, Bulgaria E-mail: [email protected] Georgi Jordanov National Institute for Geophysics, Geodesy and Geography – Bulgarian Academy of Sciences, Acad. G. Bonchev str. Bl. 3, Sofia 1113, Bulgaria E-mail: [email protected]

Abstract

The Weather Research and Forecasting (WRF) Model is a next-generation mesoscale numerical weather prediction system. It has been designed to serve both operational forecasting and atmospheric research needs. It features of multiple dynamical cores and a software architecture allows computational parallelism and system extensibility. WRF model has broad spectrum of applications across scales ranging from meters to thousands of kilometers. Several other models ranging from simple linear algorithms to complex computational fluid dynamics codes have been also developed to simulate the propagation and behavior associated with wildland fire. All attempted in some way to incorporate the effect of the three environmental factors - fuel, weather, and topography, directly corresponding to the fire behavior. Some semi- empirically based fire spread algorithms such as BEHAVE and FARSITE have led to practical in- the-field tools that required simple point or two-dimensional surface values of meteorological fields such as wind as input. The research-quality tools capable of exploring these complex interactions are not easyly available to the research community. The widely-used, established mesoscale models such as MM5 have recently begun to be applied through regional centers to provide more directed meteorological information to fire operations. However our team from the Bulgarian Academy of Sciences has found another tool for wildland fire simulations. In March 2010 has been released a new module combing WRF with surface fire behavior model, called WRF-Fire, Linux based. The semi- empirical formulas calculate the rate of spread of the fire line, the interface between burning and unignited fuel, based on fuel properties, wind velocities from WRF, and terrain slope. This tool is a joint research project of the National Center for Atmospheric Research in US. In this paper we are applying the WRF-Fire module to the collected data from a Bulgarian area near Harmanli, where willand fire from August 2009 is under our scope. The difficulty in our work is collecting and adapting the available data to the WRF-Fire module. This tool is still very much under developing and any area outside USA, needs additional implementation of the domains, fuel burning, land use, slope and terrain maps. In our work we have managed to run successfully 30 min. simulation with limited set of data given to us for our research. The results have shown that WRF-Fire has really the ability to simulate wildland fires even with not full set of data available, which make it appropriate for open area fires simulations.

Keywords: WRF, WRF-Fire model, wildland fires, Environmental modelling

VI International Conference on Forest Fire Research D. X. Viegas (Ed.), 2010

1. Forest fire prevention in Bulgaria until present

1.1 Statistics for Southern and Bulgaria

Forest fires are problem in most of the south European countries, because of the dry climate during the summer and the year-round high temperatures. Statistics have been done among the south European EU member states and the results are presented in Figure 1 below.

90000 80000 70000 60000 50000 40000 30000 20000 Number of Fires 10000 0 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008

Year

Figure 1. Number of forest fires in the southern EU member states for 1971 – 2008

In this statistics Bulgaria and Romania haven’t been included up to 2006, because both countries became member states in 2007. Bulgaria has done for its territory statistics for forest fires for the period 1994-2006. There is also data collected about the number of forest fires for every year between 1971 – 2006 (Dobrinkova, Nedelchev, 2008). This information is published by the National Fire Safety and Civil Protection Service, part of the Bulgarian Ministry of Interior.

1600

1400

1200

1000

800

пожари

600

Брой 400

200

0 1970 1975 1980 1985 1990 1995 2000 2005 Година

(a) (b) Figure 2. (a) Bulgarian forest fires in the period 1994-2006, separated by regions; (b) total number of the forest fires per year in Bulgaria between 1971 and 2006

2

VI International Conference on Forest Fire Research D. X. Viegas (Ed.), 2010

It is clearly seen from Fig. 2(b) that there is a considerable increase of the number of forest fires in Bulgaria after 1990 and especially in 1993 and 2000 when fire activity peaked and more than 1000 wildland fires devastated huge areas of forests in the lowlands as well as in the mountains. The number of fires and the average size of the fires have increased about 6 times in the recent years, however the total burned area and the percentage of vegetation burned during a year has increased dramatically - more than 30 times (Dobrinkova, Nedelchev, 2008). Even though the number of wildfires is increasing and the consequences are not only of environmental, but also of economical and social significance, proper solution haven’t been found. At present most of the countries suffering from wildfires are dealing with the disaster in the moment of its occurrence. It would be much easier to avoid the long and very expensive period of handling with the fire if a working tool such as WRF-Fire can be applied to the areas with high risk of fires.

1.2 Aero-Spatial Observation Center (ASOC) in Sofia, Bulgaria

The center was created in Sofia on 7th of July 2007 to provide the disaster control units with adequate, real time information and to ensure better coordination and effectiveness in the prevention of natural hazards. It is the first aerospace center in Bulgaria. It’s aimed to improve and streamline the process of early warning, prediction and monitoring of natural disasters and accidents on national scale. It allows discovering and following the dynamics of wildland and forest fires, floods, to estimate the loss of forest, control the conditions of the vegetation, soil humidity and erosion as well as air pollution. There are four main sources of information received by the center:  Satellite images from the Dartcom system. Information is received on every 40 min with resolution of 1 km and in five frequency ranges.  Images from MODIS system once per day.  Images from the DMC (Disaster Monitoring Constellation) with much higher resolution, namely 32m, received once on every two or three days depending on the satellites trajectory.  System for aerospace monitoring via a radio navigated or human navigated airplane. Data are sent directly to the center by satellite connection. The system operates on national level and is also used by other governmental organizations e.g. Ministry of Economy and Energy, Ministry of Environment and Water Supplies, Ministry of Agriculture and Forestry and others. The center is in 24/7 operation.

However ASOC still lacks a system for automated satellite images recognition and early detection of forest fires, floods and other natural or human-caused disasters. That is why our team from the Bulgarian Academy of Sciences has started its own research project dedicated to the forest fires models and tools for fire simulations. After a literature based analysis, we found that WRF-Fire is free Linux based model, which can simulate open area fires not only in US, but also in Bulgaria, when the needed set of data is carefully implemented into the run process.

3

VI International Conference on Forest Fire Research D. X. Viegas (Ed.), 2010

2. General description of the model

According to (Mandel, Coen, 2009) the model postulates are the fire propagation speed normal to the fireline as a function of wind and terrain slope, and an exponential decay of fuel from the time of ignition. Consider the fire area Ω = Ω(t) with the boundary Г = Г(t), called the fireline. The fireline evolves with a given spread rate S = S(x, y, t) in the normal direction = (x, y, t). The spread rate S is a function of the components of the wind and the terrain gradient given by the modified formula by (Rothermel, 1972):

(1)

where R0 is the spread rate in the absence of wind, is the wind correction, is the terrain correction, a, b and d are constants, and B0 is the backing rate, that is, the minimal fire spread rate even against the wind. A small backing rate of spread must be specified, since fires are known to creep upwind on their upwind edge due to radiation. The fuel state is maintained as the ignition time ti. In the burning area, the fuel fraction decreases exponentially from the ignition time and is given by the BURNUP formula (Albini, 1994):

(2)

where W (x,y) is the 1/e time constant of the fuel. The heat flux from the fire to the atmosphere is determined from the amount of fuel burned by

(3)

The coefficients R0, Smax, a, b, d, W, and A, which characterize the fuel, are encoded in a table of 13 fuel categories (Anderson, 1982). The fire model input data consists of the fuel category array, which is integrated in the WRF input data and can be alternatively set from the namelist for testing.

2.1 Coupling with WRF

The fire model is in the physics layer. In every time step, it takes as input the horizontal wind velocity , and it outputs the heat flux H, given by (3). Since the fire mesh is generally finer than the atmospheric mesh, the wind is interpolated to the nodes of the fire

4

VI International Conference on Forest Fire Research D. X. Viegas (Ed.), 2010

mesh, and the heat flux is aggregated over the cells of the fire mesh that make up one cell of the atmospheric mesh. At the beginning of an atmospheric time step, the wind is interpolated from the atmospheric mesh to the nodes of the fire mesh. The fire model is then advanced one or more internal time steps to the end of the atmospheric time step. The maximum time step in the fire model is limited by the stability restriction of the numerical scheme. However, the time step for the atmospheric model has been so far short enough for the fire model, and thus only one time step of the fire model is performed. After advancing the fire model, the total heat flux H generated over the atmospheric time step is inserted in the atmospheric model. The heat flux is split into sensible heat flux (a transfer of heat between the surface and air due to the difference in temperature between them) and latent heat flux (the transfer of heat due to the phase change of water between liquid and gas) in the proportion given by the fuel type and its moisture. The heat fluxes are inserted by modifying the temperature and water vapor concentration over a given number cells, with exponential decay away from the boundary. This decay mimics the distribution of temperature and water vapor fields arising from the vertical flux divergence, which is supported by infrared observations of the dynamics of crown fires. This is according to (http://www-math.cudenver.edu/~jmandel/fires/wrf-fire- doc.pdf and http://www.mmm.ucar.edu/wrf/users/workshops/WS2010/abstracts/1-1.pdf)

3. DOMAINS

To have more detailed meteorological background nesting capabilities of WRF were used. The first domain is covering area of 48 km2 with resolution 300m ( 160x160). This domain is producing boundary and initial meteorological conditions for the inner domain and in this domain there are no fire simulations.

Figure 3. Map of domain 1 – blue, domain 2 – green, ignition – red line

5

VI International Conference on Forest Fire Research D. X. Viegas (Ed.), 2010

The inner domain as seen from the map on Fig.3 is located in the middle of the coarse domain. The resolution in D2 is 60m and the area covered is 9.6 km2 (161x161). Meteorological conditions can have high impact on the fire propagation and the use of nested domain gives us much better meteorological background field for the studied area. D2 is centred on the fire ignition line and it is covering the areas of villages Ivanovo, Leshnikovo and Cherna Mogila. This area is located in South-East Bulgaria close to the Bulgarian-Greece border. This particular area is chosen because on 14th of August 2009 there was a wild land fire spreading in wide area and burning everything in the area. The approximate location of the fire ignition is located on the plot map of the area Fig.4, and this is also the place where the simulated fire is ignited.

Figure 4. Plot of the ignition line in WRF-Fire, west from Leshnikovo village

4. Initialization of WRF-Fire

To reproduce and analyze the behavior of the fire propagation in real case we need to input in the model more detailed data for elevation and landcover. To obtain the data various sources were studied and because of their public availability and coverage Shuttle Radar Topography Mission (SRTM) 3 arc sec topography data from NASA/JPL (www2.jpl.nasa.gov/srtm,http://dds.cr.usgs.gov/srtm/version2_1/Documentation/SRTM_To po.pdf) and CORINE landcover data (http://www.eea.europa.eu/publications/COR0- landcover) were chosen. The SRTM3 data has nearly global coverage and is available free. Elevation data is very important for proper simulation of the fire propagation and this is the reason to incorporate

6

VI International Conference on Forest Fire Research D. X. Viegas (Ed.), 2010

more detailed and with higher resolution data. The standard data available for WPS – WRF preprocessor is with 1km resolution and the SRTM3 data is with 100m resolution. The improvement is visible in Fig.5.

Elevation from standard WPS (1km) Elevation from SRTM 3arc sec (100m)

Figure 5. Terrain Height from different sources

To create a database for “fuel category” data, Corine 2006 lancover dataset was used. Based on the landcover types the area of the domain was divided in 13 fuel categories. This data is also very important for the proper simulation of the wild fire in a complex terrain. Because for the domain of interest there is no fuel data available, this data has to be created only on the basis of landcover data. This approach is not very accurate but without having observations and high quality and updated satellite images this is the only possible solution for the moment. In the future simulations accurate fuel data will be acquired either for governmental institutions or from field observations and high quality satellite observations. The fire ignition is set to 14.08.2009 – 12:01, which is 60sec after the simulation start. The ignition line is 1.3km long and 200m wide (Fig.4) with the following coordinates: WGS84: N 41.8425 ; E 25.9362

5. Results

From the results from the simulations several point issues are worth to mention. - WRF-Fire has the ability to simulate real wild fires. No verification of the results is made at this stage.

- Using this model one can forecast the burned area from a wild fire (Sufficient information is required for proper initialization of the model – elevation data, fuel data, land cover)

7

VI International Conference on Forest Fire Research D. X. Viegas (Ed.), 2010

- The importance of coupling fire and meteorological model is illustrated on Fig.6. One can see how big is the influence of the fire on the wind field and how the fireline propagation is influenced from the wind field.

- The importance of better and more accurate elevation, land cover and fuel data.

Temperature and wind before the Temperature and wind in the moment ignition of ignition

Temperature and wind 3min after the Temperature and wind 8min after the ignition ignition

Figure.6 Temperature on 2m (ºC) and wind vectors

On Fig.6 Illustrated the fire line propagation for 8min. We can see the change in the temperature and wind speed and direction caused by the fire. Should be noted that the coupling of meteorological model with fire model (WRF-FIRE) can be used for further research on the impact of the fire (high temperatures and updraft fluxes) on the meteorological conditions. As seen from the results, the fire is causing significant change in the wind direction and speed. On the next figure (Fig.7) is plotted the change in time of heat flux from ground fire. The firs plot is at the moment of ignition of the fire (12:01). On the next plots one can follow the propagation of the fire mainly westerly, because of the east wind. Only 9 minutes after the ignition the fire line has growth and started to spread west, south and north direction.

8

VI International Conference on Forest Fire Research D. X. Viegas (Ed.), 2010

12:01 UTC – time of ignition 12:03 UTC

12:05 UTC 12:10 UTC

Figure.7 Surface heat flux from ground fire (W/m2) and wind vectors

6. Conclusion

In this paper we have managed to fulfil 10 minutes of simulation, which is not the full time of the forest fire duration in the area we observe, but we run successfully the WRF-Fire model outside US and it was having data from different sources collected by our team. However our work has to continue, the high velocity of the wind around the fire line caused numerical instability and future simulations will be made with even smaller time step. One other thing which occurred was the fast wind along the fire line. These problems will be solved in the future and we will go for full three days simulation of the fire duration.

7. Acknowledgement

This work is partially supported by the European Social Fund and Bulgarian Ministry of Education, Youth and Science under Operative Program “Human Resources Development”,

9

VI International Conference on Forest Fire Research D. X. Viegas (Ed.), 2010

Grant BG051PO001-3.3.04/40 and the project of the Bulgarian National Science Fund, Grant No DMU 02/14 with title: Processing of Data Concerning Wild land Fires, Occurred On The Bulgarian Territory In The Recent Years By Using WEATHER RESEARCH AND FORECASTING MODEL-FIRE (WRF-FIRE)”

References

Albini, F. A., 1994: PROGRAM BURNUP: A simulation model of the burning of large woody natural fuels, National Report on Research Grant INT-92754-GR by U.S.F.S. to Montana State Univ., Mechanical Engineering Dept. Anderson, H. E., 1982: Aids to determining fuel models for estimating fire behavior,USDA Forest Service, Intermountain Forest and Range Experiment Station, Research Report INT-122. Dobrinkova N. , L. Nedelchev “ PERUN a system for early warning and simulation of forest fires and other natural or human-caused disasters “ . International Conference “Education and Science” Belovo, Russia, ISBN 5-8353-0822-1, pp 99 – 105, 2008, Mandel, J., Beezley, J.D., Coen, J.L., Kim, M.: Data assimilation for wildland fires: Ensemble Kalman filters in coupled atmosphere-surface models. IEEE Control Systems Magazine 29, 47–65 (June 2009) Rothermel, R. C., 1972: A mathematical model for predicting fire spread in wildland fires, USDA Forest Service Research Paper INT-115.

10