system analysis of weather fire danger in predicting large fires in siberian forests

8
ISSN 00014338, Izvestiya, Atmospheric and Oceanic Physics, 2011, Vol. 47, No. 9, pp. 1049–1056. © Pleiades Publishing, Ltd., 2011. Original Russian Text © A.V. Rubtsov, A.I. Sukhinin, E.A. Vaganov, 2010, published in Issledovaniya Zemli iz Kosmosa, 2010, No. 3, pp. 62–70. 1049 INTRODUCTION A forest fire occurence is caused by the combina tion of at least three necessary conditions: presence of flammable fuel, its inflammability due to low moisture content, and presence of a source of inflammation. The first component is described by the spatial distri bution of vegetation and, hence, types of flammable fuel (FF). The moisture content of FF directly depends on the weather conditions preceding the cur rent day. A source of inflammation is a probable factor of anthropogenic (people) or natural (lightning) ori gin. In this work, we analyze the effects of weather fac tors on large fires with the aim of estimating the pre dictability of fire events. A fire is considered catastrophic if it is highly intense (temperature and flame height), covers a large area of more than 10 000 ha, and results in significant or total damage of vegetation (Yefremov and Shvi denko, 2004). The conception of catastrophic fire with respect to the fire area shifts toward an increase in these areas (Sukhinin et al., 2004) in conditions of glo bal warming (Trenberth et al., 2007). The importance of Russian boreal forests in greenhouse gases emis sions (Yurganov et al., 2005) and the global carbon balance (Kasischke et al., 2005) makes urgent the problem of studying forest fires of this zone. Determination of the weather fire danger (WFD) (Sofronov et al., 2005), also called fire danger due to weather conditions, is an important problem of fire monitoring both in Russia and all over the world. The related analysis is carried out on the basis of existing WFD models with weather data as input parameters. Many countries monitor forest fires, including predic tion of potentially firehazardous weather conditions. The predicting systems are based on empirical equa tions derived from the data of previous years or adopt modifications of available systems. Thus, national WFD systems exist in Canada (Van Wagner, 1987) and the United States (Bradshaw et al., 1984); they are used in adapted form in Spain (Viegas et al., 1999). The Russian system of WFD estimation (Vonskii et al., 1981; Sofronov et al., 2005) is simpler in comparison with the ones mentioned above; it exists in two ver sions (MI1 and MI2) and has regional scales of WFD classes (Sofronov et al., 2005). This work is a preliminary study for development of a complex system for fire prediction in Siberia based on WFD indices. System Analysis of Weather Fire Danger in Predicting Large Fires in Siberian Forests A. V. Rubtsov a, b, *, A. I. Sukhinin a, b , and E. A. Vaganov a a Siberian Federal University, Institute of Space and Information Technology, Krasnoyarsk, Russia b Sukachev Institute of Forest, Siberian Branch, Russian Academy of Sciences, Krasnoyarsk, Russia *email: [email protected] Received May 26, 2009 Abstract—The prediction results of largescale forest fire development are given for Siberia. To evaluate the fire risks, the Canadian Forest Fire Weather Index System (CFFWIS) and the Russian moisture indices (MI1 and MI2) were compared on the basis of the data of a network of meteorological stations as input weather parameters. Parameters of active fires were detected daily from the NOAA satellite data for the period of 1996–2008. To determine the length of the fire danger season, the snow cover fractions from Terra/MODIS data (2001–2008) were used. The features of fire development on territories with different types of flammable fuel are considered. The statistical analysis of the areas and number of fires typical of each vegetation class is made with the use of the GLC2000 vegetation map. A positive correlation (~0.45, p < 0.05) between the cumulative area of local fires and the MI1 and Canadian BUI and DMC indices is revealed. The Canadian ISI and FWI indices describe best the diurnal dynamics of fire areas. The above correlations are higher (~0.62, p < 0.05) when we select the fires larger than 2000–10000 ha in size for the forested areas. Other cases point to the lack of a linear relation between the fire area and the values of all indices, because the fire spread depends on many natural and anthropogenic factors. Keywords: satellite data, AVHRR, MODIS, moisture indices, meteorological data, snow cover fraction, veg etation types, fire prediction, Siberia. DOI: 10.1134/S0001433811090143

Upload: a-v-rubtsov

Post on 30-Sep-2016

215 views

Category:

Documents


2 download

TRANSCRIPT

Page 1: System analysis of weather fire danger in predicting large fires in Siberian forests

ISSN 0001�4338, Izvestiya, Atmospheric and Oceanic Physics, 2011, Vol. 47, No. 9, pp. 1049–1056. © Pleiades Publishing, Ltd., 2011.Original Russian Text © A.V. Rubtsov, A.I. Sukhinin, E.A. Vaganov, 2010, published in Issledovaniya Zemli iz Kosmosa, 2010, No. 3, pp. 62–70.

1049

INTRODUCTION

A forest fire occurence is caused by the combina�tion of at least three necessary conditions: presence offlammable fuel, its inflammability due to low moisturecontent, and presence of a source of inflammation.The first component is described by the spatial distri�bution of vegetation and, hence, types of flammablefuel (FF). The moisture content of FF directlydepends on the weather conditions preceding the cur�rent day. A source of inflammation is a probable factorof anthropogenic (people) or natural (lightning) ori�gin. In this work, we analyze the effects of weather fac�tors on large fires with the aim of estimating the pre�dictability of fire events.

A fire is considered catastrophic if it is highlyintense (temperature and flame height), covers a largearea of more than 10000 ha, and results in significantor total damage of vegetation (Yefremov and Shvi�denko, 2004). The conception of catastrophic fire withrespect to the fire area shifts toward an increase inthese areas (Sukhinin et al., 2004) in conditions of glo�bal warming (Trenberth et al., 2007). The importanceof Russian boreal forests in greenhouse gases emis�sions (Yurganov et al., 2005) and the global carbon

balance (Kasischke et al., 2005) makes urgent theproblem of studying forest fires of this zone.

Determination of the weather fire danger (WFD)(Sofronov et al., 2005), also called fire danger due toweather conditions, is an important problem of firemonitoring both in Russia and all over the world. Therelated analysis is carried out on the basis of existingWFD models with weather data as input parameters.Many countries monitor forest fires, including predic�tion of potentially fire�hazardous weather conditions.The predicting systems are based on empirical equa�tions derived from the data of previous years or adoptmodifications of available systems. Thus, nationalWFD systems exist in Canada (Van Wagner, 1987) andthe United States (Bradshaw et al., 1984); they areused in adapted form in Spain (Viegas et al., 1999).The Russian system of WFD estimation (Vonskii et al.,1981; Sofronov et al., 2005) is simpler in comparisonwith the ones mentioned above; it exists in two ver�sions (MI1 and MI2) and has regional scales of WFDclasses (Sofronov et al., 2005).

This work is a preliminary study for development ofa complex system for fire prediction in Siberia basedon WFD indices.

System Analysis of Weather Fire Danger in Predicting Large Firesin Siberian Forests

A. V. Rubtsova, b, *, A. I. Sukhinina, b, and E. A. Vaganova

a Siberian Federal University, Institute of Space and Information Technology, Krasnoyarsk, Russiab Sukachev Institute of Forest, Siberian Branch, Russian Academy of Sciences, Krasnoyarsk, Russia

*e�mail: [email protected] May 26, 2009

Abstract—The prediction results of large�scale forest fire development are given for Siberia. To evaluate thefire risks, the Canadian Forest Fire Weather Index System (CFFWIS) and the Russian moisture indices (MI1and MI2) were compared on the basis of the data of a network of meteorological stations as input weatherparameters. Parameters of active fires were detected daily from the NOAA satellite data for the period of1996–2008. To determine the length of the fire danger season, the snow cover fractions from Terra/MODISdata (2001–2008) were used. The features of fire development on territories with different types of flammablefuel are considered. The statistical analysis of the areas and number of fires typical of each vegetation class ismade with the use of the GLC2000 vegetation map. A positive correlation (~0.45, p < 0.05) between thecumulative area of local fires and the MI1 and Canadian BUI and DMC indices is revealed. The CanadianISI and FWI indices describe best the diurnal dynamics of fire areas. The above correlations are higher(~0.62, p < 0.05) when we select the fires larger than 2000–10000 ha in size for the forested areas. Other casespoint to the lack of a linear relation between the fire area and the values of all indices, because the fire spreaddepends on many natural and anthropogenic factors.

Keywords: satellite data, AVHRR, MODIS, moisture indices, meteorological data, snow cover fraction, veg�etation types, fire prediction, Siberia.

DOI: 10.1134/S0001433811090143

Page 2: System analysis of weather fire danger in predicting large fires in Siberian forests

1050

IZVESTIYA, ATMOSPHERIC AND OCEANIC PHYSICS Vol. 47 No. 9 2011

RUBTSOV et al.

DATA

Daily Weather Data

The majority of WFD estimation techniquesrequires daily weather data and, in the case of theCanadian system, availability of many variables. Weused the data of the weather station network of theNational Climatic Data Center (NCDC NOAA,USA) (Lott, 2005) along with the Avialesookhranadata. The NCDC provides for hourly weather observa�tions, which we used for composing the series of dailyvalues. However, there are gaps in data of someweather stations, mainly on snow cover depth and pre�cipitation. In the case of long distances between sta�tions, we used data with an acceptable number of lags.The selection criteria were as follows: (1) the totalperiod of lack of data does not exceed 30% of the totalobservation days per year; (2) gaps are no longer than5 consecutive days in a data series. In view of this, thenumber of weather stations taken in WFD indices cal�culations varied from about 250 to about 300 everyyear. The distribution of all weather stations over allthe years is shown by the triangles on the map of fires(Fig. 1).

Active Fires Database

Operational data reception from NOAA satellites(1996–2008) and the Terra/MODIS one (2004–2008) was carried out at the Institute of Forest of theSiberian Branch of the Russian Academy of Sciencesin cooperation with the All�Russian Research Insti�tute on Problems of Civil Defense and Emergencies ofthe Ministry of Emergency Situations of the RussianFederation. Spatiotemporal information on fires wasdetected, filtered, vectorized, and stored in the fire

database (Sukhinin et al., 2004). The databaseincludes area size and spatial contours of active firesfor each day and resulting values of fire parametersthroughout the fire duration. Satellite data are pro�cessed in a semiautomatic mode. Individuallydetected fire pixels are unified into common outlinesdepending on a specified value of the distance betweenfire points. Orienting to coverage of a long time inter�val and a quite high spatial distribution, we usedAVHRR data from NOAA satellites. WFD indiceswork on a day scale; hence, they have been verified onthe basis of daily fire data. The fire statistics have beencalculated on the basis of generalized information.The spatial distribution of fires in 1996–2008 is shownin Fig. 1.

Snow Cover

Calculation of WFD indices begins every year withthe determination of the date of complete snow melt�ing (Vonskii et al., 1981). There is no strict definitionof the conditions of onset of this moment in the liter�ature. Snow can repetitively precipitate at the end ofwinter for a short time with intervals of 10–15 days;therefore, we consider the first episode of completesnow melt as the starting date of calculation of WFDindices. Subsequent snowfalls are considered equiva�lent to liquid precipitation.

In the case of lack of weather station data on thesnow pack depth, we used the Terra/MODIS data ona presence of snow (Riggs et al., 2003). The result ofcalculation of the snow cover fraction (MOD10A1)has a global coverage daily temporal and 0.05 spatialresolution. The algorithm for snow detection is based

72

68

64

60

56

52

48

14013012011010090807060

Fig. 1. Map of fires detected by NOAA AVHRR over the period of 1996–2008. The outline shows the territory under study.Weather stations are marked by triangles.

Page 3: System analysis of weather fire danger in predicting large fires in Siberian forests

IZVESTIYA, ATMOSPHERIC AND OCEANIC PHYSICS Vol. 47 No. 9 2011

SYSTEM ANALYSIS OF WEATHER FIRE DANGER IN PREDICTING LARGE FIRES 1051

on a combination of visible and near�IR measure�ments data (Riggs et al., 2003).

Vegetation Types

The problem of studying the features of fire in dif�ferent types of vegetation is urgent. The Global LandClassification (GLC2000) (Bartholomé et al., 2002),produced at the Joint Research Centre on the basis ofthe SPOT4�Vegetation satellite data, was used as avegetation map; the classification was constructedwith allowance for regional peculiarities. In the analy�sis, we used the regional data with the legend forNorthern Eurasia (Bartalev et al., 2003); it representsmore vegetation classes for the territory of our study incomparison with other open�access products of vege�tation remote sensing (RS) (International Geosphere�Biosphere Program land cover map (Belward et al.,1999), EOS�Terra�MODIS Vegetation ContinuousFields (Hansen et al., 2003), MODIS Land CoverClassification (Friedl et al., 2002), and so on). The ini�tial GLC2000 Classification includes 27 types: six for�est types, two shrubland types, three wetland types,two grassland types, four tundra types, one burnedarea type, three types of croplands in complexes, andfive nonvegetation types (Bartalev et al., 2003). Weregrouped vegetation classes by unifying layers withsimilar properties with respect to the behavior and sus�ceptibility to fires. The resulting classification consistsof 11 types: evergreen needleleaf forest, broadleavedforests, mixed forest, deciduous needleleaf forest,shrublands, steppe/grasslands, wetlands, tundra,croplands and complexes, forests in complex withother vegetation, burned areas.

METHODS

WFD Systems

In this work, we compare the Canadian and Rus�sian WFD systems. The Forest Fire Weather IndexSystem (CFFWIS) is a complex system based on long�term experiments on forest fire study. A detaileddescription of the system is given in (Van Wagner,1987); the program code for index calculation waspublished in (Van Wagner, 1985). The system includesthree FF classes with different rates of drying: finematerials of the upper litter layers (FFMC index) ofabout 0.25 kg/m2 in dry weight with a high rate of dry�ing; dense materials of organic layers of moderatedepth (DMC index) of about 5 kg/m2 in dry weight;and materials of the deep, compact, organic layers ofabout 25 kg/m2 in dry weight (Sofronov and Volok�itina, 1996; Van Wagner, 1987). The system alsoincludes three indices describing the fire behavior:BUI, ISI, and FWI; ISI represents the rate of firespread; BUI is a temporary variable combining DCand DMC for their representation in FWI; FWIreflects the intensity of combustion of the fire front

and is a general index of WFD assessment throughoutforested territories (Sofronov and Volokitina, 1996;Van Wagner, 1987). There are four FWI scales in theCanadian system for adapting this variable to differentWFD classes. We preferred the S scale, which is opti�mal for all Canadian regions.

Input parameters for CFFWIS are air temperature(Tair, °С), relative humidity (RH, %), wind speed(w, m/s), rain (r, mm), and current month. Experi�ments of previous years on the model verification inCanada have shown the correlation between the indi�ces and different fire parameters. Thus, ISI is closelyrelated to the fire area, BUI is closely related to the fireactivity, and FWI is a good general indicator of all fireparameters (Van Wagner, 1987).

Despite the fact that the FFMC index describes themoisture content of the upper layer of fine FF, whereany fire begins, the variation in this index does notallow to reveal a fire onset and to consider it as anexplicit indicator of probable fire development andspread. Values of this index are high (60–99 units) dur�ing the whole season of FWI calculation and fre�quently fluctuate, since FF of this type dries in severalhours. FFMC was taken into account in comparisonof indices with fires only in an auxiliary manner.

The moisture indices MI1 and MI2 are modifica�tion of the Nesterov complex fire danger index (Von�skii et al., 1981). The input parameters for these indi�ces are air temperature (Tair), dew�point temperature(Td), and precipitation (r). Snow absence is the condi�tion of calculation start (the date of complete snowmelting). Each MI has individual calculation equationfor Тair ≥ 0°C and Тair < 0°C.

Determination of Large Fire Areas

Figure 2 shows the annual fire statistics withrespect to the number and total area of fires on the ter�ritory under study. As follows from the statistics, thereis a tendency toward an increase in the fire activity(area and number) and the annual average fire area.The following parameters of individual fires have beenobtained from analysis of the whole fire database: themaximum total area Smax = 2.4 million hectares, thearithmetic mean total area Smean ≅ 800 ha, the maxi�

mum daily area > 200000 ha, and the daily aver�

age area growth ≅ 600 ha.Fires in regions of Siberia with forest vegetation

types were the focus of our study. However, many firesin Southern Siberia are not classified as forest fires.The fire distribution over the main vegetation types isshown in Fig. 3. This estimation is approximate, sincethe vegetation was classified as of 2000 (Bartholoméet al., 2002; Bartalev et al., 2003); and its temporalvariability was not taken into account. The vegetationcomposition within fire contours was often multicom�ponent. Thus, the statistics of vegetation areas passed

DaymaxSDaymeanS

Page 4: System analysis of weather fire danger in predicting large fires in Siberian forests

1052

IZVESTIYA, ATMOSPHERIC AND OCEANIC PHYSICS Vol. 47 No. 9 2011

RUBTSOV et al.

by a fire took into account all fires, and the area of veg�etation of one type was determined by the percentageof this type of all other types within the fire contour.The number of fires typical for the ith vegetation typewas determined from those fires during which the areaof burned vegetation of this type was more than 25% ofthe total fire area. Other cases belong to the small�mosaic composition of FF types.

The statistics of fire areas with respect to vegetation

types is given in Table 1. The mean total area for the majority of types varies from 250 to 420 ha; thisparameter is two times higher (848 ha) for deciduousneedleaf (larch), and the maximum total area is 6

times higher ( = 1250000 ha). The daily mean

area ( ) is less than by about 20–30%; thismeans that the main part of fire area is burned in oneday, or the fire spreads over already burned territory inthe following days.

Totalmean( )S

TotalmaxS

DaymeanS Total

meanS

Verification of the WFD Indices

The capability of WFD indices to describe the firebehavior was verified via calculation of the Pearsoncorrelation for two data time series. Time series ofWFD indices were contrasted with the series of sumsof fire areas within a preset radius around a givenweather station. It should be noted that dry weatherconditions do not always lead to fire occurrence(absence of an ingnition source), and an increase inthe WFD indices is accompanied sometimes by a totalabsence of fire. In view of this peculiarity, we built thedata series using the selection of days when fire areaswere nonzero and included pairs when zero WFD cor�responded to fire absence.

As follows from satellite fire monitoring data anal�ysis, several fire events can be observed from March toNovember (Fig. 4). Within one event, wildfiresdevelop in phases, spreading gradually (single�ignitionfires) or mosaic�like (multiple ignition fires) over a ter�

4 × 104

3 × 104

2 × 104

1 × 104

0

Number of fires

0

3.0 × 107

2.5 × 107

2.0 × 107

1.5 × 107

1.0 × 107

5.0 × 106

Fire area, ha

AreaNumber

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

2008

Fig. 2. Dynamics of the area (left) and the number (right) of fires versus years on the territory under study over the period of 1996–2008.

Table 1. Statistics of fire areas versus vegetation types

AreaEvergreen needleleaf

forest

Broadle�af forest

Mixed forest

Deciduous needleleaf

forest

Shru�blands

Steppe and grasslands Wetlands Tundra

Croplands and com�

plexes

Forests in complexes

, 103 ha 206.8 57.0 60.9 1250.3 46.8 188.6 41.6 159 172.5 23.1

, ha 414 348 322 848 298 367 248 400 354 130

, 103 ha 30.9 19.5 40.7 165.4 9.4 29.5 11.4 11.3 23.5 11.6

, ha 315 297 268 549 248 311 194 292 308 144

SmaxTotal

SmeanTotal

SmaxDay

SmeanDay

Page 5: System analysis of weather fire danger in predicting large fires in Siberian forests

IZVESTIYA, ATMOSPHERIC AND OCEANIC PHYSICS Vol. 47 No. 9 2011

SYSTEM ANALYSIS OF WEATHER FIRE DANGER IN PREDICTING LARGE FIRES 1053

ritory. They can last 35–40 days, during which two tothree maxima of daily sums of burned areas can existwithin a territory on fire. WFD can locally vary, but thegeneral WFD trend remains increasing. To take thisfeature into account, we calculated the daily dynamicsof cumulated sums of fire areas beginning from thestart of fire phase development (Saccum) in addition tosimple daily sums of fire areas. The criterion of divi�sion into peaks (fire scenario differentiation) isabsence of fire during next three days; it was chosenfrom considerations that a short period (one to twodays) of no fires can be caused by clouds (without pre�cipitation) appearing within the satellite’s field of view,though really there can be fires during this time.

To study the descriptive character of long�termtrends and to exclude the effects of minor oscillationsof WFD indices on the correlation factor, time seriesof more dynamic (inert) WFD indices (FWI, MI1)were moving averaged over five previous days andadditionally included in the analysis.

The following time series of data were studied forcorrespondence: (a) series of daily fires (Sday) with allthe daily WFD indices; (b) series of daily fire areaswith averaged WFD indices (FWI, MI1); (c) series ofaccumulated daily fire areas (Saccum) with all the dailyWFD indices; (d) series of accumulated daily fire areas(Saccum) with averaged WFD indices (FWI, MI1).

Source of Errors

The main source of calculation errors is data gaps.For example, one missing measurement of the precip�itation amount can result in a systemic increase of aWFD index, while in fact it can be several times lower.The measurement time of meteorological values is alsovery important; 13:00–15:00 is optimal. Nighttimedata correspond to the process of FF wetting, whichyields errors in the calculations. A sparse network ofweather stations results in differences between weatherparameters measured at the nearest station to the firearea and weather conditions causing the fire occur�rence.

RESULTS AND DISCUSSION

Several peaks in the dynamics of the number andareas of wildfires can be distinguished. Fires can beconventionally divided into spring, summer, andautumn fire danger, i.e., fire outbreaks during a season(Fig. 4). It is natural that the number of outbreaksdecreases from south to north from three to one.

We have revealed that fires can occur even at Тair <0°С and in the presence of snow pack due to a mosaic�like cover. The computability of the FWI indices atnegative air temperatures makes them more suitablefor description of spring and autumn fires if the calcu�lation start dates and conditions are precisely speci�fied.

7 × 104

6 × 104

5 × 104

4 × 104

3 × 104

2 × 104

1 × 104

0

Number of fires4 × 107

3 × 107

2 × 107

1 × 107

0

Fire area, ha

EN

F

Bro

adle

af f

ore

st

Mix

ed f

ore

st

DB

F

Sh

rubl

and

s

Gra

ssla

nd

s�st

epp

e

Wet

lan

ds

Tu

nd

ra

Bu

rned

are

a

Cro

pla

nd

s

Fo

rest

in c

omp

lex

AreaNumber

Fig. 3. Distribution of fire areas (left) and their number (right) for 1996–2008 versus vegetation types according to the GLC2000map. Abbreviations: ENF—evergreen needleleaf forest; DBF—deciduous broadleaf forest. The areas are summed according tothe vegetation percentage within fire outlines; the number of fires has been calculated only for fires where this fraction is largerthan 25%.

Page 6: System analysis of weather fire danger in predicting large fires in Siberian forests

1054

IZVESTIYA, ATMOSPHERIC AND OCEANIC PHYSICS Vol. 47 No. 9 2011

RUBTSOV et al.

6 × 104

5 × 104

4 × 104

3 × 104

2 × 104

1 × 104

0

Fire area, ha2.0 × 104

1.5 × 104

1.0 × 104

5.0 × 103

0

MI1/MI2

April May June July August September October November

2520151050

Precipitation, mm60

302010

FWI

April May June July August September October November

(a)

(b)

0

4050

6005004003002001000

DC250200150100

0

BUI/DMC

April May June July August September October November

(c)

50

25201510

0

ISI

April May June July August September October November

(d)

5

Fig. 4. Dynamics of WFD indices in 2007 for the weather station in Ulety (51.35° N, 112.47° E): (a) MI1 index (solid curve) andMI2 (dashed curve); dynamics of the fire area: daily sums (columns) and accumulated (dotted curve); (b) FWI index and precip�itation; (c) BUI index (solid curve), DCM index (dashed curve), and DC index (dotted curve); (d) ISI index.

Table 2. Averaged coefficients of correlation between the WFD indices and fire areas

Years of fires, number of w.s. Area MI1 MI1mean MI2 FWI FWImean DMC DC BUI ISI

Composite se�ries for 1996–2008, 370 w.s.

Sday 0.45 (87) 0.49 (64) 0.49 (55) 0.45 (94) 0.47 (76) 0.46 (64) 0.51 (33) 0.47 (54) 0.43 (103)

Saccum 0.5 (163) 0.5 (167) 0.48 (122) 0.44 (115) 0.44 (115) 0.49 (128) 0.52 (59) 0.51 (101) 0.42 (97)

Sample of large�scale fires, 1996–2008, 183 w.s.

Sday 0.6 (117) 0.58 (115) 0.59 (117) 0.63 (122) 0.61 (120) 0.61 (119) 0.55 (107) 0.61 (122) 0.63 (123)

Saccum 0.65 (133) 0.66 (130) 0.63 (135) 0.61 (126) 0.62 (126) 0.66 (132) 0.6 (118) 0.65 (135) 0.57 (117)

Note: The number of weather stations (w.s.) providing significant (α = 0.95) correlations from which the average values have been cal�culated is given in parenthesis.

The total number of weather stations for which firesoccurred in a square of 0.5° × 0.5° over 13 years is 370.The yearly analysis has shown an insignificant correla�tion between fire areas and all indices in any year forthe majority of cases (40–70% of weather stations).This depends on the number of “fire days” in a year,

and, hence, observations in one year are insufficient toascertain the correlation. The next step was calcula�tion of point correlations using the composite timeseries of all years when fires occurred in the vicinity ofone weather station. It takes into account the interan�nual ratio of the WFD indices and fire areas. This

Page 7: System analysis of weather fire danger in predicting large fires in Siberian forests

IZVESTIYA, ATMOSPHERIC AND OCEANIC PHYSICS Vol. 47 No. 9 2011

SYSTEM ANALYSIS OF WEATHER FIRE DANGER IN PREDICTING LARGE FIRES 1055

technique has revealed cases where lower WFD valuesof one year in comparison with other years are accom�panied by large fire areas and, hence, there is no directrelationship between these variables. The averaged val�ues of significant (p < 0.05) correlation coefficients (r)are given in Table 2. We have experimentally ascer�tained that consideration of only fires larger than aspecified size threshold results in a significant increasein r (from ~0.45 to ~0.62) for many WFD indices(Table 2). The sampling threshold is within the 1000–5000 ha range for Sday and 1000–10000 ha for Saccum .

The spatial distribution of r shows that the MI1 andMI2 indices agree with Saccum on almost the whole ter�ritory. The CFFWDIS components have better r inforest vegetation (conifers). Negative r can beexplained by two reasons: (1) the relation is not alwayslinear; (2) the influence of other factors of fire spread�ing. The use of the moving average for FWI results in abetter agreement with Saccum. This technique worksworse in general for MI1.

CONCLUSIONS

The results imply that all WFD indices are ambig�uously related with fire areas. The WFD indicesmainly reflect the FF moisture content, which is a keyfactor of onset, development, and spreading of a fire.However, a linear increase in the FF moisture contentdoes not result in a proportional daily growth in thefire area, since the fire area is a result of a complex offactors (wind, relief, surface irregularity, ignitionsource, etc.), which affects this parameter.

The use of WFD in prediction of changes in firebehavior is more suitable for large areas. The FWI andISI indices describe the daily dynamics of fire sizemore accurately in comparison with other WFD indi�ces; however, this relation is not linear. The RussianMI1 and MI2 and Canadian BUI and DMC indicesbetter predict variations in the total fire area.

The estimate can be considered more accurate onthe regional level when data of several weather stationsare considered jointly over a territory, since a localWFD can differ from a continuously increasing WFDon a territory of larger area.

The Canadian WFD system has been designed forforest types of vegetation, and our analysis confirmsthis. However, to enhance the WFD prediction effi�ciency, it should be adapted to the types and growingconditions of Siberian vegetation and local specificityof fire behavior. The detailed analysis of the results ofthe study allows the conclusion that WFD indicesshould be used in a complex. The Russian systemrequires improvement and supplementation withrespect to description (prediction) of diurnal dynam�ics of the growth in fire area, which is an importantfactor in the choice of facilities and tactics of fire sup�pression.

ACKNOWLEDGMENTS

The work was supported by the Russian Founda�tion for Basic Research (project no. 09�05�00900�a).

REFERENCES

Bartalev, S.A., et al., A New SPOT4�VEGETATIONDerived Land Cover Map of Northern Eurasia, Int.J. Remote Sens., 2003, vol. 24, no. 9, pp. 1977–1982.

Bartholomé, E., Belward, A.S., and Achard, F., GLC2000—Global Land Cover Mapping for the Year2000—Project Status November 2002, Publication ofthe European Commission, JRC, Ispra, Italy, EUR20524 EN, 2002.

Belward, A.S., Estes, J.E., and Kline, K.D., The IGBP�DIS 1�km Land�Cover Data Set DISCover: A ProjectOverview, Photogram. Engin. Rem. Sens., 1999, vol. 65,no. 9, pp. 1013–1020.

Bradshaw, L.S., et al., The 1978 National Fire Danger RatingSystem: A Technical Documentation, General TechnicalReport INT�169, Ogden, UT: U.S. Dep. of Agricul�ture, Forest Service, Intermountain Forest and RangeExperiment Station, 1984.

Friedl, M.A., McIver, D.K., Hodges, J.C.F., et al., GlobalLand Cover Mapping from MODIS: Algorithms andEarly Results, Rem. Sens. Environ., 2002, vol. 83, no. 1(2), pp. 287–302.

Hansen, M.C., DeFries, R.S., Townshend, J.R.G., et al.,Global Percent Tree Cover at a Spatial Resolution of500 Meters: First Results of the MODIS VegetationContinuous Field Algorithm, Earth Interactions, 2003,vol. 7, pp. 1–15.

Kasischke, E.S., Hyer, E.J., Novelli, P.C., et al., Influencesof Boreal Fire Emissions on Northern HemisphereAtmospheric Carbon and Carbon Monoxide, Glob.Biogeochem. Cycl., 2005, vol. 19, pp. 1–16. GB1012.doi: 10.1029/2004GB002300.

Lott, N., Data Documentation for Federal Climate ComplexIntegrated Surface Data, Asheville, N.C.: National Cli�matic Data Center, 2005.

Riggs, G.A., Hall, D.K., and Salomonson, V.V., MODIS SnowProducts User Guide for Collection 5 Data Products, http://modis�snow�ice.gsfc.nasa.gov/sug_main.html. 2003.

Sofronov, M.A., Goldammer, I.G., et al., Pozharnaya opas�nost’ v prirodnykh usloviyakh (Fire Hazard in NaturalConditions), Krasnoyarsk: Inst. Lesa im. V.N. SukachevaSO RAN, 2005.

Sofronov, M.A. and Volokitina, A.V., The Canadian Systemof Evaluation of Fire Danger in Forests, in Lesnoekhozyaistvo za rubezhom (Forestry Abroad), Moscow:VNIITsLesresurs, 1996, vol. 5, pp. 2—22.

Sukhinin, A.I., French, N.H.F., Kasischke, E.S., et al.,AVHRR�Based Mapping of Fires in Russia: New Prod�ucts for Fire Management and Carbon Cycle Studies,Rem. Sens. Env., 2004, vol. 93, pp. 546–564.

Trenberth, K.E., Jones, P.D., Ambenje, P., et al., Atmo�spheric Climate Change, in Observations: Surface andClimate Change 2007: The Physical Science Basis. Con�tribution of Working Group I, 4th Assessment Report ofthe Intergovernmental Panel on Climate Change,Solomon, S., Qin, D., Manning, M., Chen, Z., Mar�

Page 8: System analysis of weather fire danger in predicting large fires in Siberian forests

1056

IZVESTIYA, ATMOSPHERIC AND OCEANIC PHYSICS Vol. 47 No. 9 2011

RUBTSOV et al.

quis, M., Averyt, K.B., Tignor, M., and Miller, H.L.,Eds., Cambridge: Cambridge Univ. Press, 2007.

Van Wagner, C.E., Equations and FORTRAN Program for theCanadian Forest Fire Weather Index System, CanadianForestry Service. Forestry Techn. Rep. no. 33, Ottawa,1985.

Van Wagner, C.E., Development and Structure of the Cana�dian Forest Fire Weather Index System, Canadian For�estry Service, Forestry Techn. Rep. no. 35, Ottawa,1987.

Viegas, D.X., Bovio, G., Ferreira, A., et al., ComparativeStudy of Various Methods of Fire Danger Evaluation inSouthern Europe, Int. J. Wildland Fire, 1999, vol. 9,no. 4, pp. 235–246. doi: 10.1071/WF00015

Vonskii, S.M. Zhdanko, V.A., et al., Opredelenie prirodnoipozharnoi opasnosti v lesu (Determination of Naturaland Other Fire Hazards in the Forest), Leningrad:LNIILKh, 1981.

Yefremov, D.F. and Shvidenko, A.Z., Long�Term Environ�mental Impact of Catastrophic Forest Fires in Russia’sFar East and Their Contribution to Global Processes,Int. Forest Fire News, 2004, vol. 32, pp. 43–49.

Yurganov, L.N., Duchatelet, P., Dzhola, A.V., et al.,Increased Northern Hemispheric Carbon MonoxideBurden in the Troposphere in 2002 and 2003 Detectedfrom the Ground and from Space, Atm. Chem. Phys.,2005, vol. 5, pp. 563–573.