Modeling Wildland Fire with DIRSIG Zhen Wang, Anthony Vodacek, Robert L. Kremens, Ambrose Ononye Center for Imaging Science, Rochester Institute of Technology 54 Lomb Memorial Drive, Rochester, NY 14623
ABSTRACT The purpose of this paper is to describe a physics based fire model in DIRSIG. The main objectives are to utilize research on radiative emissions from fire to create a 3D rendering of a scene to generate a synthetic multispectral or hyperspectral image of wildfires. These synthetic images can be used to evaluate detection algorithms and sensor platforms. To produce realistic flame structures and realistic spectral emission across the visible and infrared spectrum, we first need to produce 3D time-dependent data describing the fire evolution and its interaction with the environment. Here we utilize the Clark-Hall model to represent the finescale dynamics of convective processes in a wildland fire for visualization. Then the grid-based output from the model can be imported into DIRSIG along with the spectral emission of a wildland fire, in order for DIRSIG to run the ray-tracing model to create the synthetic scene. The technical approach is based on a solid understanding of user requirements for format and distribution of the information provided by a high spatial resolution remote sensing system. Keywords: DIRSIG, wildland fire, blackbody radiance, thermal radiance
1. INTRODUCTION Wildland fire is of great importance for global climate change research and as a natural hazard, is an important problem in disaster management. Thermal remote sensing systems are being used to observe fire activity for Further author information: (Send correspondence to Zhen Wang) Zhen Wang: E-mail:
[email protected], Anthony Vodacek: E-mail:
[email protected]
wildland fire management. These systems usually have high spatial resolution in the MWIR and LWIR so that very small hot spots can be identified. DIRSIG, a 3D Image Generation Model, can meet Remote Sensing requirement. DIRSIG can simulate both high spatial and high spectral resolution scene, which is useful for algorithm testing, sensor platform evaluating, and better understanding of fire propagation as viewed from a remote sensing platform. Many methods of modeling fire and flame have been proposed. Lee et al.3 presented a fire as an evolving front and a system of particle-based flames. They deposited emitters which could emit flame particles according to their internal state and simulate fire motion. Wei et al.4 have proposed the use of textured splats as display primitives for an open surface fire model. They started with fire particles, updated the velocity, and then projected the light onto surroundings. Perry and Piscard5 developed a technique for rendering flames and modeling their spread over polygonally defined objects. They also render flames as modified particles system. Nguyen et al.8 used a computational fluid dynamics approach to visualizing flame. Some of these approaches might be inappropriate to visualizing flame as it is viewed from a remote platform because of the short integration time and spatial resolution. Our goal is not only realistically visualize the flame, but also incorporate the fire spectrum across the visible, near IR and thermal IR spetrum into the simulated scene. Currently we don’t know the flame extinction, so we suppose the flame is thick enough to be opaque. Our implementation of the thick fire model is as follow. First, the Clark-Hall Coupled Atmosphere-Fire Model7 is run to give the dynamic buoyancy which affects the rise of smoke and soot, and created 3D facetized objects on the basis of buoyance infomation. Then blackbody radiation model was used to estimate the temperatures of the fire color for visualization in DIRSIG. Finally, these 3D buoyancy objects and blackbody radiation were incorporated into DIRSIG to run ray tracing model, and compared the result to that rendered with real flame spectrum measured with a field spectroradiometer.
2. BLACKBODY RADIATION AND COLOR The Plank or blackbody radiation equation for the spectral radiant exitance from a surface is:
−1 hc wm−2 µm−1 Mλ = 2πhc2 λ−5 e λkT − 1
(1)
We plotted blackbody curve at 1100K along with fire spectral radiance measured with a field spectroradiometer, as shown in Figure 1.
Figure 1. ASD data compared to Blackbody radiation
The dashed curve is blackbody radiation at 1100K, scaled to generally match the radiometer spectrum. The two lines fit pretty well from visible region to NIR. They differ a lot in the wavelength longer than 1.8 µm due to the sensor saturation. We can tell that the fire is pretty ideal blackbody radiator. Next problem is how the blackbody radiator looks in DIRSIG rendered scene. With Wien’s Displacement Law, i.e.,
λmax =
2898 [µmK] T
(2)
we can predict the spectrum where the peak radiance from a blackbody at certain temperature will occur, vice versa. Plot (a) in Figure 2 shows the radiance of selected color temperature. Scene (b) is the downlooking natural color rendition of panels of different radiator temperature with a grass covered land surface. The panels on the left column are standard red, green, blue, black, white, and gray panel. The upper three panels on the right column are an extended sodium emission source with different intensity, which appears orange as expected. Rest of the panels are for blackbody radiators with temperatures of 1500K, 1400K, 1300K, 1200K, and 1100K
corresponding to fire temperatures, as shown in Figure 2.
(a) BB radiation
(b) panels with low BB temperature
Figure 2. BB radiation at low temperature.
All the temperatures have peak radiation outside the visible region. Most of the radiated energy is in longer wavelengths. By properly scaling the scene, we can see that the blackbody radiator at 1500K and 1400K looks yellow, 1300K looks orange, and 1200K looks red. Figure 3 is a downlooking color rendition of several spheres of bright blackbody emission at temperature 1500K, 1300K, 1200K, and 1100K, from top to bottom. The spheres are at ground level, so there is red illuminated area around the spheres.
3. MODELING FIRE AS EMITTING PARTICLES Wildland fires are very complex and its behavior depends on the reactants, topography, humidity, wind, and etc.. We use Clark-Hall Model to calculate fire buoyancy and volume information within computational region. DIRSIG is a 3D based image generation model. The objects in the scene should have certain facets to allow ray tracing. But, fire and smoke do not have smooth and well-defined surfaces. They are not rigid objects and their motion can not simply be described by transformation.6 To meet these requirments, we simulate fire as a group of many facetized spheres, such that both facets of objects and volume of fire are defined. Fire is an emitting source. In DIRSIG, the sources are treated spectrally and therefore allow the user to
Figure 3. spheres with Blacbody radiation
observe illumination effects from them. In DIRSIG scene, we assign the facets of sphere blackbody radiance property at 1500K, and simulate fire as extended area source by placing the spheres in a pattern defined by the Clark-Hall model.
(a) Simulate fire as source particles
(b) Simulate fire as source spheres
Figure 4. Simulated flame with different primitive particle size.
Figure 4 compares the natural color rendition between different size of sphere. In both scene there are two logs at the bottom and flame on the top. In scene (a) the spheres are small enough to be considered as particles
space,yet keeps the facets of the sphere. Each particle is displayed as a point light source and appears yellow. Most of the background is dark while there is still red illuminated area at the lower part of the fire. In scene (b) the spheres have much larger radii and they are connected to each other. Along with source properties, the partly-overlapped spherers look like a group and appear bright yellow. A sphere behind another sphere is not obscured but adds more intensity to the pixels covered. There is red, gradual, more intense illuminated area around the bright fire. Figure 5 compares the scene with different fire sizes. Both of them are taken from the same time step of C-H Model.
(a) smaller fire
(b) larger fire
Figure 5. Flames of different size, from the same time step of dynamics model
4. SIMULATE HYPERSPECTRAL IMAGE OF WILDLAND FIRE Knowing the spectrum signature of all the material, we could generate a multi spectral real world scene which is useful for further illustration. The scene used to incorporate flat panel-shaped sources was an aerial image of forest fire, as shown in Figure 6. We first run classification algorithm on the aerial image, associate a digital count in the image to a DIRSIG material ID to creat a material map, which is helpful to indicate the presence of healthy vegetation and burned area on the terrain. Then, use hit maps to find the coordinates of fire pixels in the aerial image, and put panel-shaped sources in the corresponding positions in simulated image. Based on buoyancy from C-H Model, we simulated smoke as 3D facetized spheres. The source spectral radiance used was that obtained using a field
spectrometer.
(a) Aerial image of forest fire
(b) Simulated image with ASD data
Figure 6. True color top view of forest fire scene.
The fire pixels with the field spectrometer spectral data look yellow, which agrees with the true color of flame when measured on the ground. The input spectral used to simulate the fire spectrum are affected by atmospheric absorption in the final rendering. Since the field instrument was very close to the burning fire, say, several meters, when the measurement was taken, there was no atmospheric influence (Figure 7a). When we put the sensor at 2500m altitude and run the ray tracing, atmospheric attenuation becomes apparent (Figure 7b).
(a) Input ASD data
(b) Simulated fire spectrum
Figure 7. Comparison of fire spectrum before and after rendering.
5. CONCLUSIONS In this paper we described the method to simulate fire as source sphere in DIRSIG to make it both spatially and spectrally realistic. To simulate the radiosity, we first utilized blackbody radiation model to find the temperature which could give right color in DIRSIG. Source panels with temperature corresponding to fire do produce realistic visualization. We presented overlapping 3D facetized spheres as extended area sources for DIRSIG. The most important aspect of the sphere system is that it is a surface-based representation and allows ray tracing on each sampled point source on the surface. The Clark-Hall based fire shape combined with the emitting spheres yields reasonable looking flame structure. The spheres in this paper are opaque. In reality, there are thin fires as well and that are transparent. The next step of our research is to incoorporate fire extinction properties and simulate flames of thin fire.
ACKNOWLEDGMENTS The authors would like to thank Emmett Ientilucci for advice and helpful comments through the work.
REFERENCES 1. Brown,S.D, DIRSIG User Documentation, Digital Imaging and Remote Sensing Laboratory, March 26, 2003 2. Ientilucci,E.J., Brown,S.D., Schott,J.R., Low-Level Simulations and Extended Area Source Modeling, RIT/DIRS Report 00/01-68-159, Rochester, NY, Januraty,(2001). 3. Haeyoung Lee, Laehyun Kim, Mark Meyer, Mathieu Desbrun, Meshes on Fire 4. Xiaoming Wei, Wei Li, Klaus Mueller, Arie Kaufman, Simulating Fire with Texture Splats 5. C.H.Perry, R.W.Piscard, Synthesizing Flames and their Spreading, Proc. of the Fifth Eurographics Workshop on Anim. and Sim., Oslo, Norway, Sept 1994 6. William T. Reeves (1983), Particle Systems - Technique for Modeling a Class of Fuzzy Objects, Computer Graphics, 17:359-376.
7. Terry L. Clark, Janice Coen and Don Latham (2004), Description of a coupled atmosphere-fire model. Intl. J. Wildland Fire, In press 8. Duc Quang Nguyen, Ronald Fedkiw, Henrik Wann Jensen: Physically based modeling and animation of fire. SIGGRAPH 2002: 721-728
