SOME DETAILS OF FOREST FIRE STATISTICS IN BRAZIL

The objective of this research was to answer some q uestions about forest fire occurrence in Brazil. A total of 19,377 fires were analyzed in the period 1 998 – 2002. Tabulated data, in percentage, were transformed through 2 / 1 y + and then submitted to the analysis of variance. The means were compared through the SNK test at 5% significance le v l. Results showed significant differences both, in number of fires and burned areas along the month s of the year. August was the month with higher number of fires, followed by July, October and Sept ember; August, September and November presented the highest burned areas. Significant dif ferences were also detected among the group causes and along the daily hours. Incendiary was the leadi ng cause, statistically different from the other groups, both in number of fires and burned area. Ab out 69% of the fires, corresponding to 66% of the burned area occurred between 12 noon to 6:00 PM. Nu mber of fires starting hourly, from 1:00 PM to 4:00 PM, didn’t differ statistically among them, bu t were significantly different from the other daily hours. As for the burned areas, fires starting at 5 :00 PM should be included in the previous group. No statistical difference was detected among the days of the week both, for the number of fires and the burned areas, showing that, on the contrary of the popular belief, fires don’t preferentially occur during the weekends.


INTRODUCTION
Fire control efficiency can be improved if the forest manager has knowledge of some characteristics of fire occurrence in the forest district he supervises.Knowing the fire season extension allows the activation of the suppression forces when the danger is high and to reduce the preparedness during the months of low risk.Determination of the main causes will help to direct the prevention actions to reduce them.The knowledge of the higher risk hours of the day will permit the establishment of the lookout towers surveillance schedule.Another information that also can be useful is the distribution of fire occurrences through the weekdays.There is a belief among fire fighters that most fires occur during the weekend, but only a significant record of fire occurrences could prove if this is true or not.
According to Crosby (1954) the success in forest fire suppression depends mainly upon the speed and power of initial attack.The maintenance of an efficient system of fire control is quite expensive and therefore economically unacceptable to be active all year round.So, it is important to know the extension of the fire season and other aspects related to the fire occurrence.
The first paper dealing with fire statistics in Brazil was published in 1974 (SOARES e CORDEIRO 1974), based on recorded fires in the central region of Paraná State, from June1965 to May 1974.Only in 1984 the first paper showing fire statistics nationwide was published (SOARES, 1984), based on data collected in 1983.After that, several papers were published (SOARES 1988;SOARES 1989;SOARES;SANTOS 2002b;SANTOS 2004), always improving the data basis.
The objective of this paper was to carry out statistical analysis on some aspects of fire occurrence, such as monthly, hourly, and weekday distribution, and group causes, in order to establish the fire season extension, the time of the day and the day of week with higher occurrence percentages, and the main group cause.

METHODS
The database used to carry out this research was the information contained in the questionnaires, received from forest companies and other forest institutions that operate fire control systems and record the fire occurrences, in the period 1998-2002.Therefore, it is necessary to clarify that the data used in the research refers to protected areas from 9 (Amapá, Bahia, Espírito Santo, Minas Gerais, Pará, Paraná, Rio Grande do Sul, Santa Catarina, and São Paulo) of the 26 Brazilian States.
The questionnaires, accompanied of letters explaining the objectives and the importance of the project, were sent to forest companies, forest associations, national and state parks, regional IBAMA (Brazilian Forest Service) offices, state forest services, and similar institutions, in order to cover the broader possible universe of information generator organizations.
The information extracted from the questionnaires for this research were: i) date of occurrence; ii) burned area; iii) time of fire detection; and iv) probable fire cause.

Monthly distribution
The fire occurrences as well as the burned areas were distributed along the months of the year to establish the country's fire season.

Hourly distribution
The fire occurrences and the respective burned areas were also distributed along the hours of the day to establish the peak of fire starting.

Distribution through the days of the week
In order to answer to the question whether fires occur preferentially during the weekends, an analysis of the occurrences distribution through the days of the week was carried out.

Probable causes
The fires that had the probable causes identified were classified according to the following groups, as suggested by FAO (1954): i) Lightning; ii) Debris burnings; iii) Incendiary; iv) Smoking; v) Campfires; vi) Forest operations; vii) Railroad; viii) Miscellaneous.Fires that didn't have the cause identified were not included in the statistics.

Statistical analysis
To proceed the analysis of variance the months of the year, the daily hours, the days of the week, and the group causes were considered the treatments, and the years (5) of data collecting, the replications.Because the data utilized were expressed in percentages, with some low values and even zeros, the transformation by 2 / 1 y + was used (STEEL; TORRIE 1960) and normal distribution tests were applied.Mean values were compared by the SNK procedure at 5% significance level.

RESULTS AND DISCUSSION
The statistical analysis showed significant differences both, in the number of fire occurrences and in the burned areas, along the months of the year (Table 1).August was the month with higher number of fires in the studied period, followed by July, October and September.As for the burned areas, August, September and October didn't statistically differ among them, but were significantly different from all other months.The results suggest that the fire season in Brazil, in the period 1998-2002, extended from July to October, when almost 60% of the fires occurred, corresponding to nearly 90% of the burned area.Soares;Cordeiro (1974) analyzing data from the State of Paraná, concluded that most fires and respective burned areas were recorded from July to December.Soares;Santos (2002a), based on data from Brazilian protected lands, in the period 1983-1997, suggested that the fire season in the country extended from July to November.
Data from table 2 shows that significant differences were observed in fire occurrence and burned areas during the hours of the day.Fires were recorded mainly from 1:00 to 4:00 PM (almost 45% of the occurrences), whereas most of the areas were burned from 1:00 to 5:00 PM (more than 50% of the affected areas).On the other hand, fires seldom occurred between midnight and 7:00 AM.Unpublished data from the Monte Alegre forest district, in the State of Paraná, showed that in 1982, a severe fire year, more than 79% of the fires were recorded between 11:00 AM and 6:00 PM, and more than 96% between 9:00 AM and 10:00 PM, period of the company lookout tower's operation.
Data presented in table 3 shows that although the number of fire occurrences on Sundays was a little higher than the other days, no statistical difference was detected.Regarding the burning areas, due to the great data variability, there were no significant differences among the days of the week.These results contradict a popular belief that says that fires occur mainly during the weekends.
Incendiaries were the main group of fire causes (Figure 1), significantly different from all others, and followed by Debris burning (Figure 2) and Miscellaneous.Lightning (Figure 3), the only natural cause was responsible for less than 2% of the reported fires.As for the burning areas, incendiaries were also the main group, followed by Debris burning.In the previous surveys on forest fires causes (SOARES e CORDEIRO 1974;SOARES 1984;SOARES 1989) the main cause was Debris burning.However, from 1994 on, Incendiaries became the leading cause of forest fires in Brazil (SOARES e SANTOS 2002b;SANTOS 2004).Unfortunately, this is a tendency observed in most parts of the world (SOARES 1997).

CONCLUSIONS
The results obtained in this research allow the following conclusions: • The fire season in Brazil, in the period 1998-2002, extended from July to October, when almost 60% of the fires occurred and nearly 90% of the area was burned.• Most fires occurred between 1:00 and 4:00 PM and the majority of the area burned was a consequence of the fires that started from 1:00 to 5:00 PM. • No significant differences were observed in the number of occurrences or in the burned areas, among the days of the week.• Incendiaries were the leading group of fire occurrences and burned areas, seconded by Debris burning and Miscellaneous in the number of fires and Debris burning in the affected areas.

Figure 3 .
Figure 3. Fire caused by a lightning that stroke a snag in a broadleaf forest.Figura 3. Incêndio causado por um raio que atingiu um toco de peroba.
* Means followed by the same letter are not statistically different by the SNK test at 5% level.

Table 2 .
Hourly distribution of fire occurrences and burned areas in the period 1998-2002.Tabela 2. Distribuição das ocorrências de incêndios e áreas queimadas através das horas do dia no período 1998-2002.Means followed by the same letter are not statistically different by the SNK test at 5% level. *

Table 3 .
Weekly days distribution of fire occurrences and burned areas, from 1998 to 2002.Tabela 3. Distribuição das ocorrências de incêndios e áreas queimadas através dos dias da semana no período 1998-2002.Means followed by the same letter are not statistically different by the SNK test at 5% level. *

Table 4 .
Fire occurrences and burned areas by cause group, from 1998 to 2002.Tabela 4. Ocorrências de incêndios e áreas queimadas por grupo de causa no período 1998-2002.*Means followed by the same letter are not statistically different by the SNK test at 5% level.