Autonomous Early Warning System for Forest Fires
Tested in Brandenburg (Germany)

(IFFN No. 22 April 2000, p. 84-90)


Abstract

Forest fires cause significant economic damage and hazard to environment all over the world. Apart from preventive measures, early warning and fast extinction of fires are the only chance to avoid major casualties and damage to nature, especially in regions with dense population. As a common method, trained staff observes the endangered areas. In Germany alone, several hundred observation towers were erected in the forests. The staff works up to 12 hours per day and usually under difficult circumstances (extreme temperatures, isolation, continuous concentration).

To date, all attempts to develop a technical system for this task have failed to outlast the test stage. In most cases the chosen components do not work reliable enough. The Autonomous Early Warning System For Forest Fires AWFS described herein is based on new concepts of hard- and software. It is adapted to the specific conditions in densely wooded regions and detects fire by the trail of smoke.

AWFS consists of a rotating digital camera with a special filter and an innovative electronic system. Thus, an utmost high level of reliability is achieved. The noise is extremely low and allows high radiometric resolution (14bit). Digital data are transmitted from the camera to the computer via optical fibers and get evaluated. The necessary software forms the central component of the system. It recognizes smoke almost in real time by analyzing its typical dynamic and stochastic features. This became possible by modifying know-how gained in space projects. However, only recent development of fast CPUs and high capacity storage media allowed to finally solve complex problems of real-time picture processing at low cost. Warnings are autonomously passed over to a central unit, where an operator will evaluate them. For this purpose, comprehensive and user-optimized software was developed. It visualizes all information necessary for taking further steps and assists in decision-making.

AWFS was installed and tested on three observation towers in Brandenburg, Germany, during the 1999 forest fire season. It became apparent that the main requirement for absolute reliable smoke detection was met. The false alarm rate due to weather and harvest activities commonly remained below 1%, which is well acceptable. Other improvements will be effective soon. The testing forest authority confirmed that the system is mature for service and easy to use. The number of systems will be increased therefore. Moreover, other German states and some European countries are very interested in this technology.

Introduction

The probability of fires in forests and fields is steadily increasing due to climate changes and human activities. In Europe, up to 10,000 km² of vegetation are destroyed by fire every year, and even 20,000 to 90,000 km² in North America. Prognoses presume from the assumption that forest fires including fire clearing in tropical rain forests will halve the world’s forest stand by 2030. Vegetation fires result in high human death toll, speed up the extinction of species and worsen the greenhouse effect. Approximately 20 % of the CO2 emissions into the atmosphere are released by forest fires, estimated the Enquiry commission "Prevention To Save The Earth Atmosphere" of the German Bundestag in 1990. Germany, too, will be affected by the impacts of global climate change. In Brandenburg and some other German states enormous economic damage is caused by forest fires. For instance, in Brandenburg alone more than 1,000 forest fires are registered every year. 90 % of them are caused by human activities. Large fires cause a total damage of up to 70,000 DM/ha. The annual financial loss caused by forest fires in Germany amounts to a two-digit number of millions. However, preventive and fire extinction measures cost even several times this sum.

In order to minimize damage, forest fires must be recognized as soon as possible (within a few minutes). Therefore, great efforts are made in all respective regions to achieve early recognition. Although numerous technical methods were tested, reliability was in no case sufficient to develop a product suitable for the German market. As an example, infrared sensor systems tested in Spain can only detect the fire itself. However, smoke is the feature relevant for early recognition of fires in densely wooded areas. Optical systems as AWIS in the Netherlands (Breejen et al. 1998) and Firehawk in South Africa are also in a test phase. But they often imply a high rate of false alarm caused by clouds, light reflection, agricultural activities and industrial plants.

In Canada and Russia an early warning system based on aircraft patrolling is used, which means late recognition of forest fires though and is expensive to operate. Evaluating satellite data is also not very successful, as spatial and time resolution are not sufficient to allow local prevention. Moreover, clouds obstruct the view very often. . However, with the German project BIRD a new generation of imaging infrared sensors for Earth remote sensing objectives including is developed (Briess et al. 1999). A major intention of the BIRD is to demonstrate the scientific and technological value and the technical and programmatic feasibility of fire detection of under low-budget constraints. The start of this small satellite mission is planned for the end of year 2000. Within the European project FUEGO an operational mini satellite constellation is studied (Gonzalo 1996.). It will become operational in 2004 to provide early fire outbreak detection and high resolution fire-line monitoring. Hence to date experienced fire-watchers are employed everywhere in the world to observe endangered forests. In Germany several hundred observation towers are manned during main forest fire season. The fire-watchers observe the forests up to 12 hours per day under utmost difficult circumstances (extreme temperatures, awkward hygiene conditions, isolation, only short breaks from concentration) and report about any smoke formation. Apart from that, authorities usually have to spent large sums on the construction of observation towers, as these edifices need to be built, maintained and operated in accordance with relevant legislation and regulation. As an example, approximately DM 350,000 are required to build one observation tower in Brandenburg.

The pilot project "Autonomous Early Warning System For Forest Fires" (AWFS) was ordered and supported by the forest authority of Peitz, Brandenburg, and promoted by the European Union. It comprised installation and testing of a system for the following tasks:

A solution for this complex undertaking was found by further developing know-how from unmanned space missions and consistently adapting it to the problem of forest fire recognition.

Technical description of AWFS

According to the forest authority’s specification, an autonomous early warning system for forest fires must meet the following technical requirements:

System concept

Principally, there are various methods suitable to recognize vegetation fires, e.g. analyzing picture information provided by digital cameras or by infrared imagers, or detecting emission lines of conflagration gases, or active measurements with Lidar evaluating the laser signal backscattered from smoke particles. To find out which method is suited best, several preliminary tests were performed observing controlled fires with various sensors (CCD-camera, IR-radiometer, IR-spectrometer). We had to take into account our own results as well as experience gained internationally on the field of early forest fire recognition, also given facts on site in Germany, the technical requirements mentioned above and the required user-friendliness and economic efficiency. Bearing all this in mind, we chose a sensor type based on a digital CCD-camera with high resolution. Smoke detection within the visible spectral region is especially important in densely wooded forests, as open flames (which IR-sensors respond to) give alarm too late. Furthermore, cameras provide the operator in the control center with expressive images and hence make it easier for him to evaluate the situation. It was one of the project’s main objectives to allow human contribution in a suitable way during the process of evaluating the alarm and selecting the appropriate fire fighting method. For such purpose, the control center is equipped with a number of computer-assisted supports.

(18 KB)

Fig.1. Digital CCD-camera

The digital high resolution Frame Transfer CCD-camera with special filter (see Figure 1) scans the forests from the top of the observation tower. AWFS can also be mounted to braced poles of mobile phone providers, high buildings or other suitable locations. The images are resolved with 14 bits and transmitted via optical fibers to the computer unit which is located in the tower too. There they get analyzed by means of specially developed software. If there seems to be a smoke formation, compressed pictures and further details (time, position) are reported via ISDN to the control center, where they are processed in a PC displayed on the screen. The operator receives all information he needs to make decisions. Currently, one control center can support up to 7 towers. In each tower up to 8000 digital images with a data volume of 16 GByte are produced and evaluated every day.

Hardware components installed in the observation tower

The basic components of AWFS are shown in Fig.2. The most distinctive feature of the CCD-camera is its innovative electronic concept of four functional groups: CCD-head, clock-driver, analogue signal chain and controller. The camera is mounted on the very top of the tower by means of a pan and tilt unit (PTU). It takes the camera approximately 10 minutes to come full circle. The controller generates or manages all digital control signals for the CCD transport cycles, analog signal processing and PC-interface. Incoming commands are interpreted and carried out. The video signal is pre-processed in the signal chain on analog basis, then submitted to correlated double sampling, before it runs through further conditioning and multi-level filter. After the signal is digitalized in a 14bit analog-to-digital converter the image data are serialized and transmitted to the controlling PC via optical fibers.

(12 KB)

Fig.2.Basic components of AWFS

The electronic components are utmost resistant against environmental conditions, stand out for their low energy consumption and extremely low noise. Due to the high radiometric resolution (~16,000 different grey scale values) the camera covers a wide range. Even very structures can be resolved under all sorts of lighting conditions. The 70 mm objective with 10° field of view allows 2 m geometric resolution in 10 km distance. Tests confirmed that the red-free filter increases the contrast between vegetation and smoke, as red light is hardly reflected at all by chlorophyll.

The pan and tilt unit can be positioned with a relative precision of up to 0.2° and with an absolute precision of 1° after being oriented in the landscape by means of GPS-defined land marks. During scanning stage three single images are taken in 1-second intervals for every camera position. Then, full image information is transmitted to a controlling PC at the tower bottom, where the data are evaluated, stored and passed on to an image processing computer. Both computers work with the operating system MS Windows NT.

Moreover, the controlling PC covers the following functions:

Data transmission to the PC in the control center is currently achieved by a wired ISDN-connection. However, it is also possible do use radio transmission or other specific networks.

Image processing software

The image processing computer uses complex algorithms to identify smoke in real time. Simultaneously, it calculates the optimum exposure time and sends it to the controlling PC. The image processing software is the heart of the AWFS. It evaluates in only a few seconds typical features like dynamic and stochastic behaviour. In every camera position several images are taken. First of all, exact matching must be achieved for the images taken from the same camera position, because the towers tend to swing considerably in the wind. Next, the horizon line is determined in the image for orientation purposes. The smoke is identified then by means of dynamic and structural features and by its grey scale value. It is prerequisite to reliable recognition that several features are taken into consideration.

In a first step, typical features of smoke are looked for by analyzing a standard difference image. Wind and thermal convection of hot smoke gases change the grey scale value of smoke areas in the subsequent images. However, other environmental phenomena (e.g. clouds, wind, dust formations, reflections, bird flights, cars) can cause similar effects for short periods of time (in comparison with the time scale of the dynamic behaviour of smoke). Moving objects (cars, planes, birds) can be eliminated by an additional evaluation of the third image, because smoke is quite stationary within the time between first and last image (several seconds). Irrelevant space frequencies are eliminated by band-pass filtering the standard difference image. As the special red filter reduces the green colour of the forest significantly, smoke of wood fire always stands out against the surrounding forest. This fact supports additional suppression of interfering signals (as an example, smoke is easy to distinct from cloud shadow therefore. Finally, adaptable threshold values prompt the decision, whether fire alarm is to be released or not.

In a second step, the texture is evaluated. The respective method is based on the structural analysis of the texture of smoke, which can be clearly discerned from the surrounding structures. It works even without any comparison image and hence does not react to changes in illumination nor to moving objects like vehicles. From mathematical view, the structures are described as stochastic effects superimposed to the average grey scale value. Therefore, it is first necessary to calculate the estimated average grey scale value, so that the stochastic arises from the difference between the original and the estimated image. The mathematical basis is explained by Hetzheim (1999). Typical smoke structures are separated then by means of various procedures. A second image, which is prerequisite to the first step described above, are reused to verify the results.

In accordance with the observed features both methods proceed classing the identified possible smoke areas with probabilities. These are condensed to one total probability then. As an example, the total probability is low, if the identified areas do not overlap and clearly differ in size.

Each of the two methods described above works sufficiently enough to detect smoke on its own, but their simultaneous employment increases the reliability of smoke detection considerably.

Control Center

The control center, too, is equipped with a PC and appropriate software. It deals with the following tasks:

click to enlage (94 KB)

Fig.3. Fire detection from Kathlow tower on 8 August 1999

By means of the software developed for the control center, even operators who might not be familiar with modern PC-technology can easily make themselves acquainted with their scope of duties within a few days. Their knowledge about local conditions as well as the information the system provides them with enable them to soon make qualified decisions on initiation of fire fighting activities. Annex 2 shows a possible display variant on the monitor.

Results and discussion

Only practical operation can demonstrate the performance of an autonomous early-warning system for forest fires. Therefore, a pilot test was started during the forest fire season 1999, after several tests with controlled fire were made. Supplementary to traditional fire watching methods, AWFS were installed on three observation towers (Kathlow, Reuthen, Jerischke) of the Spree-Neiße district in southern Brandenburg, which is a region with very high forest fire risk. The numerous open pits and power plants in this region with their dust and smoke emissions make the task of fire watching especially difficult. The control center is located at the premises of the local forest authority in Peitz. The test results were monthly evaluated and reported in cooperation with the responsible officers.

In the test region 16 forest and field fires happened. All these fires were detected and indicated by the AWFS within the set time limit. Despite the approximately 10 min time of revolution, fire indication sometimes (especially during late afternoon) was given even earlier than by the experienced observation tower staff, who obviously suffered from symptoms of tiredness. Figure 3 is an example of automatic smoke detection.

False alarms mean a specific problem. One differentiates between alarm due to irrelevant smoke sources, e.g. chimneys, and proper false alarm. Irrelevant smoke sources are usually stationary objects. Therefore, a facility was created, that allows the operator to permanently mask these sources. The complex image analysis during which various features are evaluated in a multi-step process has been described above and proven to efficiently avoid potential false alarms.

Figure 4 presents the statistical evaluation of the false alarm rates at all three pilot project locations during the period from 16/8/1999 to 18/9/1999 (end of forest fire season). For most days the rate of false alarm is clearly less than 1%. About 230 decisions about smoke formations are to be made hourly by the software on each tower. A rate of 1 % means approximately 2 false alarms per hour, which the operator can easily cope with. However, under certain weather conditions the number of false alarms increases due to light reflection, ascending water vapor (after short but heavy rain) or low clouds. Dust formation as a result of harvest activities can be taken for smoke, too. Even experienced staff has often considerable problems to differentiate properly though. The region around Kathlow has higher rates, which is due to Cottbus city and the Jänschwalde open pit and power plant being in the range of view of the camera. The option to mask such smoke sources was not entirely used yet in the pilot project.

(4 KB)

Fig.4. Number of days with different rates of false alarm at the different AWFS locations from 16 August to 18 September 1999

Any problems occurring were continuously evaluated during the test stage and the software improved accordingly. As a result, the user was satisfied with the AWFS.

All in all, AWFS offers the following advantages:

Further development

The test-stage turned out to be successful. The experience gained will be considered and evaluated during the next few months. A new generation of AWFS will be developed and tested.

Here are the focal points of future development:

Due to its universal basic structure, AWFS can be used in other areas as well. The concept of the system (digital camera with high spatial and radiometric resolution, wide range in brightness, real-time image processing, autonomous alarm signal transmission to a control center) and the experience gained so far in detecting complex structures in natural environment are suitable for various observation tasks, e.g. environmental monitoring or security duties. The system can not only observe sensitive areas, but also autonomously give alarm, transfer data to any other place and selectively store images.

References

Breejen, E. den, Breuers, M., Cremer, F., Kemp, R.A.W., Roos, M., Schutte, K., and Vries, J.S. 1998. Autonomous Forest Fire Detection. In: Viegas, D.X. (Ed.), Proceedings of the III International Conference on Forest Fire Research and 14th Conference on Fire and Forest Meteorology held in Luso, Coimbra, Portugal on 16-20 November 1998, pp. 2003-2012.

Briess, K., Bärwald, W., Gerlich, T., Jahn, H., Lura, F., and Studemund, H. 1999. The DLR Small Satellite Mission BIRD. Proc. of 2nd IAA Symposium on small satellites for Earth observation, Berlin, April 12 - 16, p. 45-48

Gonzalo, J. 1996. FUEGO programme. Proc. IAA Symp. on Small Sat. for E. O., Berlin 1996, IAA-B-902.

Hetzheim, H. 1999. Analysis of Hidden Stochastic Properties in Images or Curves by Fuzzy Measures and Functions and their Fusion By Fuzzy-or Choqut Integrals. Proc. , 5th International Conference on Information Systems Analysis and Synthesis, Orlando, Florida, pp. 501-508.

 

 

E. Kührt, T. Behnke, H. Jahn, H. Hetzheim,
J. Knollenberg, V. Mertens, G. Schlotzhauer

Deutsches Zentrum für Luft- und Raumfahrt
Rutherfordstr. 2
D-12489 Berlin
GERMANY

 and

B. Götze
INO control GmbH
Schulstr. 6
D-01728 Possendorf
GERMANY


 Top
Country Notes
IFFN No. 22