The Need for Improved Forest Fire Detection

Transcription

1 The Need for Improved Forest Fire Detection by Peter Kourtzl Introduction In Canada, for the years 1980 to 1984, the average seasonal number of forest fires was 9 270; these fires burned 2.7 million hectares of forest and incurred an average annual fire control cost of $238 million. The annual damage caused by these fires was estimated to be 31 0 million dollars (Can. For. Serv. 1986). The traditional response to severe fire losses has been to strengthen the fire suppression capability, especially in aircraft and helicopter operations. A significant portion of these expenditures have been aimed at strengthening the agencies' ability to fight large fires. The detection component of the operation has received little benefit from the increased expenditures, although few statistics have been published on the subject. However, the proportion of the total budget spent on detection has probably been decreasing. reason, fire detection was high on the priorities of the early forest protection agencies. Beginning in the 1920s and lasting until the mid-1960s, large networks of lookouts formed the backbone of the detection effort. By 1960, in eastern Canada alone, there were nearly lookouts in operation. Emergence of Aerial Fire Detection Although air patrols were used since shortly after the First World War, they did not come into general use until the late 1960s. Then the rising costs of manned lookouts and their limited effectiveness made air patrols an economically attractive alternative. Today, aerial detection has replaced many lookouts and is carried out over about two-thirds of Canada's protected forests. Detection in the remaining area depends upon lookouts, usually combined with some form of supplementary air patrol. The principle of early detection combined with quick The fire management organization of a typical province initial attack while the fire is small has long been recognized as consists of a provincial headquarters, a half dozen regional the most effective approach to controlling forest fires. For this support centres and, within each region, three to six districts each with their own dispatch centre. Currently most decisions on daily fire control operations are made at the district level. 'Petawawa Nat~onal Forestry Inst~tute. Canad~an Forestry Service, Chalk River. However, coordination of aerial detection activity usually is Ontarlo 2Prof D.L Martell def~nes a f~re arrival as one that has become vlsuaily at the level based On detection requirements detectable solicited daily from each district. Single or light twin-engine 272 aoirt 1987 The Forestry Chronicle

2 leased aircraft are used to patrol the region on predetermined patrol routes. The particular patrol plan in use on any given day depends on recent thunderstorm activity, visibility, and level of fire weather indices. The arguments that supported the change from lookouts to aerial detection centred mainly on the dollar savings associated with air patrolling and the proven lack of effectiveness of the lookouts. Difficulties with the lookout system included a shortage of motivated, reliable observers, unionization of the observers and the associated pressuresfor more standard wages and working conditions, high maintenance and access costs, and generally poor performance in detecting fires. The main advantage of the air patrol system was perceived to be the ability to customize daily patrol routes in response to anticipated fire situations. On days not expected to have fires no flights were to be carried out, thus saving money. On severe fire days intensive air patrols were to be carried out in those parts of the region expected to have fires. Although detection costs on such days were high, the intensive coverage was aimed at earlier detection of potentially dangerous fires. Unfortunately much of anticipated benefits have yet to materialize. One reason for this has been the nature of aircraft contracting procedures and regulations. Until now, at the beginning of each season, regions contracted from the private sector for the regional services of needed aircraft. The contracts were usually for the dedicated use of aircraft and pilots for a guaranteed minimum number of hours operating from specific airports. The more hours guaranteed, the lower was the cost per hour for the aircraft. Much of the motivation for flexible daily patrol planning was lost with this type of contracting arrangement. No savings were possible during inactive fire seasons because of the high numbered minimum hours guaranteed. During active fire seasons, once contract hours were used, the additional higher cost of extra patrol time discouraged adding extra patrols on severe fire days. More flexible contracts permitting province-wide movements of aircraft, when not required locally, would make better use of surplus contract hours. Fortunately, the deregulation of the aircraft industry provides new opportunities to restructure detection contract procedures and basing strategies to respond to variable fire detection requirements. Another reason for the lack of effectiveness of aerial detection has been the difficulty in preparing and executing daily patrol plans. The daily planning process is complex, involving integration of a location-specific prediction of regional fire occurrence, the day's expected fire behaviour, knowledge of the state of the initial attack plan for the day, and, for regional detection policy and budget limitations. Until now, the major component in the planning process, daily fire prediction, has been missing. Only recently have lightning location systems become operational; a few are currently coupled to computerized daily fire prediction systems. At present, most agencies perceive the daily detection planning process to be a relativley routine task. Typically, assignments of aircraft patrol routes and patrol frequencies are made using predefined guidelines based on the day's weather index and visibility conditions. Fire Behaviour and Its Relation to Detection Effective daily detection planning requires an understanding of how a forest fire behaves over time. Three phases in the life of a typical Canadian forest fire can be identified. First is the ignition phase where either man or lightning starts the fire. This process occurs in a very short time interval ranging in length from a few milliseconds with lightning to a few minutes with ignition sources such as cigarettes. Having an ignition source in forest fuel in no way guarantees that a detectable fire will result. Moisture and fuel density conditions may be such that sustained combustion is impossible. For example, one should not patrol areas just because a lightning location system indicates that lightning has occurred. Most lightning strikes do not result in fires. Also, a large but unknown proportion of fires self-extinguish shortly after ignition. This becomes particularly evident during patrols behind thunderstorms during periods of high fuel moisture. The rate of lightning fire detection increases sharply compared with later patrols. One other potential source of ignition is the natural spontaneous combustion of certain forest fuels. It definitely occurs in sawdust piles, probably occurs in drying muskeg and other heavy organic buildups, and may occur in the decaying parts of standing dead trees. Estimates on the proportion of fires owing to this cause vary somewhere between 5% to lo%, but these estimates are impossible to confirm. No Canadian fire agency recognizes spontaneous combustion as an ignition source. Only eyewitness accounts, laboratory experiments, and theory point to this as a cause. Of all lightning fires reported, about 20% have no physical evidence of lightning (Kourtz 1967). They are assumed to be lightning-caused because of their remote locations. So little is known about the nature of spontaneous combustion fires that it is impossible at present to consider them in daily detection planning. Smoldering is the second phase in the life of a typical fire. This phase consists of a relatively dormant, low energy smoldering state. In most forest regions of Canada, a fire ignited in the latter part of the day must smolder overnight and into the next burning period if it is to become a detectable fire. This phase will not be present when the fine surface fuels are extremely dry. Such conditions occur frequently during midsummer in the intermountain region of western Canada and occasionally during periods of extreme drought in the northern and eastern forests. Visual detection of a fire by an air patrol during the smoldering phase is very uncertain but not impossible. Whether or not puffs of smoke are visible depends on the fire's energy, fuel moisture, detection visibility, and air stability. In eastern Canada smoldering will occur throughout the late evening, night, and early morning. Visual detection flights can be carried out during the early morning but the chance of detecting a smoldering fire is quite low. Most agencies do not fly patrols much before noon on typical fire days. This is not necessarily the best strategy, especially when multiple lightning fires may be present. Even with low detection probability, early morning flights may be well worthwhile, as exolained later. The third phase in the life of a fire is characterized by sufficient energy buildup to sustain flame spread. It is during this phase that most detections are made. Fire report statistics indicate that a large proportion of Canadian fires are detected while burning in low to moderate energy states. With few exceptions, smoke emissions are sufficient to enable easy visual detection if the observer is sufficiently near, Provided that low to moderate fire weather conditions persist, these low energy fires result in near constant area burned and stable control costs, not matter when they are detected or attacked. August 1987 The Forestry Chronicle 273

3 Thus, detection and initial attack efforts will appear to be successful. Indeed, during periods of low and moderate fire weather conditions, few if any air patrols may be required. Because most fires are detected while they are in a low to moderate energy state there is a tendency by officials to deemphasize the importance of a good fire detection system. However, a very small proportion of the total number of fires in every region absolutely require early detection. These "critical" fires have fuel and weather conditions that permit rapid growth with high energy release rates. The probability that man's suppression attempts will stop these fires once they have developed to their full energy potential is relatively low. Often such escaped fires only stop spreading when they run out of suitable fuel or when the weather modifies the fuel so that it no longer supports rapid combustion. On the other hand, if the "critical" fire is detected and attacked while it is smoldering, or just entering a rapid spread state, it is easily extinguished with few resources. Fire Occurrence and Detection The variable daily and yearly occurrence rates of forest fires has a strong influence on the design of a fire detection system. The annual number of Canadian forest fires recorded in the past 30 years varied between in 1954 to in 1981 with an average of (Ramsey amd Higgins 1982). Lightning started about 20% of these fires and man was responsible for most of the remainder. The actual number of fires varied considerably from one locale to another. Much of this variation is caused by irregular occurrence of lightning fires. An efficient detection system should be able to respond to these varying rates of occurrence either through flexible local aircraft contracts or provincial relocations of aircraft. Important differences exist between lighting- and humancaused fires that must be considered in planning daily detection activities. In general, lightning fires can occur anywhere over a large forest region while human-caused fires are located near areas with road or water access. Humancaused fires usually are ignited in small numbers and these are well distributed throughout the day. They become detectable as burning conditions develop according to a welldefined diurnal pattern. Lightning fires are most often ignited in multiples, clustered within fairly small geographic areas, and often within short time intervals. It is not uncommon for a boreal forest region of km2 to have 20 or 30 lightning fires each day during dry periods. Extreme situations have been experienced in British Columbia where more than 500 fires have occurred in a single 24-hour period. Daily human-caused fire occurrence, on the other hand, presents few surprises in terms of numbers and location. The number of such fires that will occur the next day is usually about the same as occurred during the current day provided the weather has remained constant. The failure of effective fire control operations is most often associated with the multiple lightning fire situation. Under normal fire load conditions there are insufficient resources for detection and initial attack to handle the multiple lightning fire situation. Should multiple lightning fires occur unexpectedly, local resources can be quickly exhausted, leaving newly arriving fires free to develop rapidly. Under these conditions human-caused fires also become important. Although they are relatively few in number they too can easily escape if the initial attack effort is delayed or weak. One or more large fires, regardless of the cause, can drain resources to a level where further detection action is pointless. Initial attack resources are just not available for newly detected fires. Daily fire occurrence prediction is a major key in improving fire detection effectiveness. Fire prediction provides advanced warning of the multiple fire situation so that additional resources can be transferred to the region in a timely fashion. Likewise, regional detection and initial attack resources can be transferred to other regions when not required. Historical records have shown that the number of human-caused fires that will occur in a local area on a given fuel moisture class day is adequately predicted by the Poisson distribution (Martell 1972). An estimate of the Poisson parameter or mean number of fires expected can be determined from local historical fire and weather records. Experience has shown that this relationship, combined with forecast fuel moisture, permits reasonably accurate predictions of daily human-caused fires. Likewise, the occurrence of lightning-caused fires appears to be quite predictable given estimates of the number of cloud-to-ground lightning flashes, as supplied by electronic monitoring systems and the moisture contents of the appropriate fuels (Kourtz 1974). Ignition, smoldering, and arrival2 models are currently being developed at the Petawawa National Forestry Institute to improve lightning fire predictions. Relying on the Public for Detection Statistics indicate that the public detects the majority of fires even in regions with intensive air patrol activity. In general, air patrols are credited with detecting only between 20% to 40% of the fires. This percentage varies inversely with population density. In some regions the public can be relied upon to report fires and only the occasional supplementary patrol is required. In areas with a high number of remote lightning fires, air patrols have a much larger percentage of first discoveries. In such areas the lack of communications facilities decreases the effectiveness of the public. All significant fires will be reported by the public if left to burn long enough. Each region must address the question of how long a delay between fire arrival and report time can be tolerated. Where fuels, weather conditions, and population density are such that large fires are very unlikely, potential damage is light, and fires are reported promptly by a concerned population there is justification for heavy reliance on the public. For most regions, public reporting can be used during low and moderate fire weather conditions. As mentioned earlier, it doesn't matter when these fires are detected and attacked: fire damage will be low. However, during high and extreme burning conditions an aggressive air patrol policy must be carried out. 4" Aerial fire detecti n is relatively inexpensive compared with the cost of a mo rn aerial initial attack operation. If a provincial fire manage ent agency is currently spending millions of dollars to support an initial attack program and is willing to spend large amounts on suppressing a fire in a region of interest, it is foolish to scrimp on detection expenditures. To rely on the public to report fires under these conditions does not make economic sense. Arguments based on the fact that an area has had few fires or that the public have reported all the fires in the past are weak. It only takes one extreme-day detection failure and its associated costly fire to nullify years of false detection savings. 274 aoot 1987 The Forestry Chronicle

4 Poor Visibility: the Archilles' Heel of Modern Fire Control The multiple fire situation is not the only cause for failure of a fire management system to meet its suppression objectives. Poor detection visibility can set off a vicious chain of events leading to disaster. In every region of Canada, there are days each season when detection visibility is less than 5 km. On these days air patrols cover such a small percentage of the region that patrols are of little value and often not flown. Large fires can easily develop unnoticed if poor visibility coincides with high fire occurrence and severe burning conditions. By the time these fires are reported they can be hopelessly out of control. The best fire organization, having correctly predicted the fire load and prepositioned suppression resources, is nearly helpless when poor visibility causes a failure of the detection system. There is no easy solution to the visibility problem. Infrared vidicons that operate in the 1- to 2-micron portion of the spectrum (still in the region of reflected sunlight) can give television-like images of the forest through thick smoke and haze. But this same haze penetration feature also causes them to not see the wisps of smoke we wish to detect. Also, vidicon devices cannot compare with the human eye in spatial resolution and consequently are inadequate for long-range detection. A satellite infrared imaging system that has the necessary resolution to detect a small fire must be in low, fast orbit, resulting in coverage of a specific ground point once every one or two weeks. Perhaps a dozen satellites in low orbit would be required to give two-hour repeat coverage of all Canada's forested area. Also, cloud masks small infrared sources from satellite view, making such expensive systems extremely unreliable. Forest fires emit considerable amounts of microwave radiation that does penetrate cloud. So far, however, it has been impossible to construct a detection system with sufficient sensitivity and a small enough antenna to be practical. High altitude, airborne infrared scanners operating in wavelengths above 3 microns can easily detect exposed thermal emissions from small, smoldering fires (Hirsch 1971 ). This approach requires a high altitude aircraft, sophisticated navigation equipment, and an infrared sensing system that is specifically designed for fire detection. Such a system costs in excess of several million dollars. Attempts to use this system during the late 1960s and early 1970s were generally unsuccessful because of the inability to route the system over enough fires. At that time daily fire prediction systems were not functioning. In the near future, province-wide fire prediction will be available that could support a limited airborne infrared detection operation aimed at supplementing the visual system in times of multiple lightning fire occurrence and poor visibility. In fact, the poor visibility problem can only be alleviated through this approach. Quality of Detection Statistics Many of our present ideas concerning fire detection are based on historical fire and weather statistics. One must be cautious in drawing conclusions based on these data. We have little idea of the number of fires that are ignited or the times of these ignitions. Our records show only those fires on which some action was taken. For example, holdover durations of lightning fires before and after the introduction of lightning sensors are very different. Before the sensors, many storms occurred unknown to observers. Therefore, reported holdover times were long. Conversely, with human-caused fires, the present notion that these fires arrive shortly after ignition is probably incorrect. Many probably smolder through one or more burning periods. Past detection decisions have a huge influence on the reliability of historical data. If a detection planner did not believe that detection flights on a specific day would be useful, no flights would be flown and, therefore, no fires would be detected. These human-biased observations create major problems in using historical detection data. For instance, a specific lightning fire may have a three-day reported holdover time. This may be because no detection flights were flown on the first two days after the storm. Had a flight been flown on the dav, after the storm, the fire miaht have been detected. A similar situation ex.& in verifying fire predictions. If the detection planner does not choose to patrol on a day predicted to have fires, these fires, if present, will probably not be reported until at least the following day. Therefore, one might conclude that, because no fires were reported, the prediction scheme is faulty. Relation of Detection and Initial Attack The process of forest fire management can be broken down into the component activities of prevention, fire occurrence prediction, detection, initial attack, and large-fire suppression. The level of effort put into each of these components influences the effectiveness of the total fire control operation. Detection is only one component of the fire management system and, as such, must not be viewed in isolation. Assuming the goal of a fire management agency is to reduce the negative impact of forest fires as effectively as possible under a budget constraint, the best policy with today's knowledge and technology probably is to suppress all fires before they begin rapid growth. This means that efforts must be concentrated on prevention, prediction, detection, and initial attack rather than on large-fire suppression. In planning daily detection patrols consideration must be given, one way or another, to the impact of detection decisions on the performance and cost of the total system. Each component of the fire management system can be viewed as a factor of production in an economic model. Theory states that under a budget constraint with the goal of maximizing total productivity, we should spend money on each factor such that its marginal contribution to total production equals that of all other factors (Henderson and Quandt 1958). Otherwise, spending money on some other factor will contribute more to total productivity. This theory presents an interesting guide to the proportion of money to spend on detection and initial attack. It would be foolish to spend most of the agency's budget on detection and not be able to attack the fires effectively after they are detected. This situation existed in the early days of fire control with its large lookout networks supported by horse, wagon, and canoe for transport of initial attack crews. At the other extreme, and tending toward our present situation, a good initial attack force is ineffective if poor detection exists resulting in many initial attacks on fires already hopelessly out of control. The most productive combination of the two components occurs when each component's marginal contribution to the fire management objective are equally balanced. Although it is difficult to prove, it now appears that most agencies are spending too little on detection. The last August 1987 The Forestry Chronicle 275

5 - Lines of equal effectiveness --- Lines of equal total detection/ini tial attack budget ooo=*- Line of rational combinc?tion of detection and initial attack effort 10 escaped fires per season 0 Y n $ spent on initial attack Figure 1. Determining the Detection Budget. For a desired level of effectiveness of 7, the total budget (n) IS required with $x spent on detection and $y on initial attack. dollar spent on detection contributes more to the fire management objective than the last dollar spent on initial attack. Reallocations of expenditures from initial attack to detection are required to bring the system into balance. Figure 1 presents a graphic solution to the problem of determining the proper balance between detection and initial attack expenditures. The vertical axis represents the annual dollars spent on detection and the horizontal axis the annual dollars spent on initial attack. The straight diagonal lines represent total dollars spent on detection and initial attack for the fire season. For a given budget there is a corresponding diagonal cost line representing all possible allocations to detection and initial attack. The solid curved lines represent points of equal combined effectiveness measured by some suitable criterion such as the expected number of escaped fires during the fire season. Given the goal of operating at a specific level of effectiveness, an agency would be indifferent as to where on the corresponding effectiveness curve it operated if costs were irrelevant. The dotted line represents the optimum path on which a rational agency should operate. This is the locus of points representing the most effective combination of detection and initial attack expenditures for a given total budget. Agency managers could use their own version of this diagram to assist in determining the proper detection-initial attack mix. Subjective approximations of the effectiveness curves can be made using a panel of detection and initial attack specialists. Related to this exercise, and part of the reason for the lack of serious interest in the detection business, is the difficulty in measuring the worth of a detection system. The measure of value of a detection system is far more complex than simply the number of fires detected per unit of patrol effort. The criterion must consider the impact of detection decisions on the goal of the organization. Clearly no direct measure of effectivenes is possible. We must content ourselves with a measure that, hopefully, reflects the importance of detection to the total fire management system. For example, we might operate a detection system in a manner that will result in the expected number of fires that will escape initial attack below a specific threshold level. In actual practice, such an objective might reduce to a rule such as: patrol an area if the chance of detecting a fire is above a specific threshold. Under this rule, all fires are treated in a similar manner, independent of their potential energy states. As fires arrive, they are 'cleared' from the system as quickly as economically possible. This is perhaps not a bad strategy given our present state of knowledge of the fuel moisture and weather forecasting. Detection Patrols for Another Reason There is another important but less obvious relationship between detection and initial attack. Detection patrols are 276 aoot 1987 The Forestry Chronicle

6 carried out not only to quickly detect fires but also to gain information on the state of the fire occurrence situation. Early morning patrols in dry weather after thunderstorms are an example. Such patrols detect only a few of the fires that will eventually arrive that day. However, if such a patrol detects several fires, there is an excellent chance that the region is facing a multiple fire situation. Precious time is gained to build up detection and initial attack strength for that day. Detection patrolling to gather information will become increasingly important as sophisticated decision support system are developed. Night infrared patrols will have as their main goal the collection of information aimed at supporting the decision for provincial shifts of resources. The fact that such patrols will locate about one quarter of the existing fires that must be attacked the next day will be a secondary benefit. Summary Although the figures have not been published, it appears that the proportion of the presuppression budget spent on fire detection has been steadily decreasing for the past two decades. This decline in detection activity can be attributed, mainly to the detection system's perceived poor cost/ performance record along with ever increasing dollar demands by the initial attack component of the fire management system. Some of the main problem areas that exist with Canadian detection systems include: - The present form of commercial aircraft contracts results in lack of flexible use of aircraft in repsonse to daily regional and provincial needs. New contract arrangements or ownership options should be explored. - General lack of daily lightning-and-human-caused fire predictions for the detection planning regions. Reliable fire occurrence forecasts should significantly improve the efficiency of the detection system by aiding in the scheduling and routing of patrol aircraft to those areas with the potential for critical fires. Lightning fire predictions are especially important and, in their present stage, go a long way toward alerting the agency to the troublesome multiple lightning fire event. - Delegation of detection planning activities to less experienced personnel. Detection planning is a complex process integrating all that is known about fire weather, fire behaviour, fuels, predicted fire occurrence, and agency policy and budget restrictions. Training at a more advanced level would be appropriate. Detection planning models to assist in patrol scheduling and aircraft routing have been developed by the Petawawa National Forestry Institute and are now being tested in the field. These models should prove useful in dealing with the complexity of daily detection planning. - Lack of appreciation of the relationship between detection and initial attack and their combined importance in achieving the agency's fire management goals. The contributions of these components must be balanced using a common measure of effectiveness related to the goals of the agency. The last dollar spent on detection should contribute as much as the last dollar spent on initial attack toward the agencies fire management goal. Otherwise, money is being wasted. - One of the weakest points of any modern fire management organization is its inability to carry out effective detection under poor visibility conditions. Often the occurrence of poor visibility conditions coincides with periods extreme fire weather. The results, too often, are fires hopelessly out of control by the time they are reported and attacked. The one solution to this problem lies with the use of airborne infrared detection systems. Research and development work is required to build a modern, solid state sensor system. Such a system would be a most valuable asset to any large provincial fire agency. - A policy of heavy reliance on the public for detection extending into periods of high and extreme fire weather. Under prolonged periods of low and moderate fire weather conditions public reporting of fires is usually adequate. However, intensive detection efforts are required in these same areas during extreme conditions. The money saved by cutting detection efforts will be spent many times over by the initial attack component having to fight larger, more active fires. In conclusion the performance of the fire detection component is critical to the successful meeting of Canadian fire management goals. Most problems with the present aerial detection system can be corrected through agency policy reformulation, advanced training programs, and the introduction of computer technology. A large research effort is required to develop an updated technical solution to the poorvisibility problem. References Canadian Forestry Service Selected forestry statistics, Canada Can. For. Serv. Inf. Rept. No. E-X-37 Martell, D.L An application of statistical decision theory to forest fire control planning. M.Sc. Thesls, Dep In Eng, Univ. Toronto. Henderson, J.M. and R.E. Quenolt Microeconornlc Theory: A Mathematical Approach. McGraw-Hill, New York. Hirsch, S.N., R.F. Kruckeburg and F.H. Madden The bispectral forest fire detection system. 7th Symp. Remote Sensing Environ. Proc., May 1971, Ann Arbor, Mich. Kourtz, P.H Lightning behaviour and lightning fires in Canadian forests. Can. Dept. For. Rural Dev., For. Branch, Forest Fire Research Institute, Ottawa. Dept. Pub. No Kourtz, P.H A system to predict the occurrence of lightning caused forest fires. Can. For. Serv., Inf. Rept. No. FF-X-47. Ramsey, G.S. and D.G. Higgins Canadian Forest Fire August 1987 The Forestry Chronicle 277