GLOBAL BIOCHEMICAL CYCLES, VOL. 16, NO. 1, 10.1029/2000GB001357, 2002

4. GAC Observations of the 1997 Fire Development

[19] Having established the relative reliability of GAC fire statistics for Borneo in the El Niño year of 1997, the Satellite Active Archive was further interrogated to construct a data time series with which to analyze the spatiotemporal development of the 1997 Borneo fires. Relatively cloud free observations were again required, and the archive was found to contain generally two or three suitable nighttime NOAA 14 GAC scenes per month, where Borneo was largely cloud free and located around the center of the AVHRR swath. The selected image dates were 6, 21, and 31 July; 1 and 19 August; 2 and 11 September; 12, 17, and 21 October; 2 and 11 November; and 13 December 1997. The relatively sparse nature of this data set when compared to daily overpass rate of the POES satellite highlights the cloudy nature of this tropical region, even during the intense drought conditions existing in 1997.

[20] Each of the 13 GAC scenes was subject to the fire detection procedure outlined in section 3.2 using a T_3.7-11 threshold of 6 K. Fire detection and statistical analyses were carried out using calibrated, raw geometry GAC data, followed by geocoding of the results to indicate temporal changes in fire spatial distribution. Geocoding was carried out using the automatically generated control points present in the AVHRR data header [Kidwell, 1995], followed by manual ground control pointing using coastlines. Root-mean-square error in the geocoding was consistently reported as less than one GAC pixel. Geocoded data were compared to the land cover map Forest Cover in Central Indonesia [World Conservation Monitoring Centre (WCMC), 1996].

Figure 9. Temporal development of the 1997 Borneo fires as determined using nighttime AVHRR GAC data. Image dates are indicated by data points. Fire pixel numbers are expressed in LAC data equivalent pixels using the inverse GAC sampling procedure, as outlined by Belward et al. [1994].

[21] Figure 9 shows the variation in fire pixel number as determined from the 13 GAC images listed previously. The data indicate a 3 month temporal envelope for the period of major fires (August–October), with no detectable fires in the 6 or 21 July data and all fires ceasing by 13 December. Unlike the data of Legg and Laumonier [1999], only relatively cloud free scenes were used to construct these statistics, but we concur with their finding that the cessation date of the fires appears directly related to the arrival of significant rainfall. This also agrees with the detailed climatic data shown by Toma et al. [2000]. Fire activity peaks in the 2 September image, with 9733 LAC pixel equivalent fires identified. The fivefold increase in identified fire pixels between the 19 August and the peak of 2 September highlights this 2 week period as one of the most important in determining the final huge extent of the resulting forest damage. Though they will not be studied here, it is also known that major fires again occurred during the first half of 1998, and in fact, it is believed that forest damage was even greater during this second wave [Mori, 2000].

Figure 10. Spatiotemporal evolution of the 1997 Borneo fires illustrated using analysis of eight nighttime AVHRR GAC images of the island taken during the second half of 1997. Each white dot represents a GAC pixel determined to have contained one or more active fires at the time of the satellite overpass.

[22] Figure 10 shows examples of the geocoded fire pixels extracted from the GAC imagery. At the start of the activity in August 1997 the fires are seen to be active mostly 100–200 km north of the southern coast, an area characterized by the interface between cleared, cultivated land and lowland rain forest. This appears to provide further evidence for the primary cause of the initial fires, namely, land clearance at the forest margin. Over the following months the areas of peak fire concentration move successively southward toward the coast, impacting areas progressively characterized by peat swamp forest [WCMC, 1996]. In 1996 this area had been identified as a likely focus for future deforestation by Achard et al. [1998], and one region of peak fire activity during September 1997 is concentrated just east of Palangkaraya, an area directly associated with the “Million hectares-Mega-Rice Project” [Lim et al., 1999]. From 1996 this project drained large areas of south Kalimantan's tropical peat swamps using a vast network of canals, with the ultimate but unsuccessful aim of large-scale rice cultivation. The drying and deforestation of the swamp forest associated with this activity most likely acted to directly increase fire intensity in this area. Other fire activity peaks occurred at the same time near the towns of Memala and Panahan in the south and around Ketapang on the eastern coast. The timing of peak fire activity found using GAC imagery concurs with the results of Nakajima et al. [1999] and Legg and Laumonier [1999], who used data from the Total Ozone Mapping Spectrometer to confirm September 1997 as the period of maximum spatial extent and density of the fire-related haze. Furthermore, Davies and Unam [1999] showed that in September 1997 levels of PM₁₀ particulates recorded near ground level reached a maximum of 995 µg m^-3 as far away as Kuching, Malaysia. This is ~20 times the normal background level, and these concentrations, believed to be the highest ever recorded for an urban area, are directly attributable to the peak in Indonesian fire activity occurring at the time. Interestingly, Nakajima et al. [1999] report that the 0.5 µm Angstrom parameters for the haze peaked later, in October 1997, and one reason maybe the increased burning of peat that occurred as the fires progressed farther into the southern coastal swamps. It seems likely that the smoke from burning peat would have different chemical and particulate characteristics to that produced by burning vegetation, and this may account for the later maxima in the haze optical absorption reported by Nakajima et al. [1999].

Figure 11. Cumulative map of fire pixels derived from the 13 nighttime AVHRR GAC scenes identified in the data of Figure 9 (July–December 1997). Each white dot represents a GAC pixel determined to have contained at least one active fire at the time of one or more of the satellite overpasses.

Figure 12. Fire analysis in relation to land cover. Points indicate the area of each respective land cover type identified on the island of Borneo by WCMC [1996]. Bars indicate the percentage of each land cover affected by fires according to the cumulative fire pixel map shown in Figure 11.

[23] Figure 11 shows a cumulative fire hot spot map derived from all 13 GAC images. Comparison of this map with the cumulative 1997 ATSR and AVHRR LAC hot spot map of south Kalimantan, included by Legg and Laumonier [1999], provides very good agreement, further confirming the legitimacy of the GAC analysis. Comparison with Figure 3 indicates that fires were concentrated in the low-lying areas of Kalimantan, primarily in a latitudinal band below 2°S between Ketapang in the west and Tanahgrogot in the east. Some fire activity is also present in western and eastern Kalimantan, but activity is absent from the central mountain range and from Malaysia and Brunei. The massive impact that the 1997 fires had on south Kalimantan can be clearly perceived from this image. Overlaying this cumulative fire map with the land cover map of WCMC [1996] enables a simple calculation of the percentage area of each land cover type that was directly impacted by the 1997 fires (Figure 12). In terms of absolute area, the land cover categories of swamp forest, lowland rain forest, and areas of cleared cultivation were all similarly affected by fire, with only an 8% difference in the number of fire pixels. However, because swamp forest covers a smaller absolute area than these other two land cover types, the percentage of swamp forest impacted by the 1997 fires was far greater, being <20% of the total area of this ecosystem type. The development of such severe fire activity was presumably encouraged by the drying out of the normally moist peat layers during the intense drought, most likely exacerbated by previous draining of large areas in preparation for agriculture. Siegert and Rücker [2000] indicate that the postfire survivability of trees in burned swampland was typically far lower than in lowland forest, further highlighting the swampland ecosystem as the most seriously impacted area.

Citation: Wooster, M. J., and N. Strub, Study of the 1997 Borneo fires: Quantitative analysis using global area coverage (GAC) satellite data, Global Biogeochem. Cycles, 16(1), 10.1029/2000GB001357, 2002.