This file was created by scanning the printed publication. Errors identified by the software have been corrected; however, some errors may remain. Real Time AVHRR Detection of Forest Fires and Smoke in Mexico Between January and June 19981 Ignacio Galind02 Ramon 50lan0 3 Abstract-Using satellite imagery, the whole forest regions of Mexico are extensively studied to monitor fires and smoke during the 1998 biomass burning season. The spatial and temporal distribution of fires are examined. Although most of Mexico suffered from forest fires, the largest number of correlated pixels are located in the states ofChiapas (4,394, only during May), Durango (4,363, March to June), Jalisco (3,414, January to March and May), Guerrero (2760, March-May) and, Oaxaca (2093, April-May). The largest number of fires occurred in May and haze and smoke covered most of the country, the Gulf of Mexico and the southern part of the United States. We propose a multispectral detection method that operates in real time. It works on the imagery received before sunrise and after sunset. A flag is shown together with fire coordinates. At present the method depicts all kind of biomass burning including controlled straw and stubble burning for agricultural purposes. Forest fires occur in Mexico every year between December and August during the dry season. Usually forest fires peak on April (Rodriguez-Trejo, 1996). Data for the period 19801997 indicate a yearly average of 6,837 forest fires with a damaged surface of 223,114 ha, i.e., about 33 ha/fire. More than 80% of forest fires correspond to shrubs and scrubs (SEMARNAP, 1998). Although most offorest fires (97%) are due to human negligence or. deliberate action, however natural calamities such as hurricanes or the EI Nino/Southern Oscillation (ENSO) events contribute to accumulate enormous amounts of dry organic matter. These conditions are associated on the next year to a temporal drought now called La Nina (SEMARNAP, 1998). In fact, Table 1 data shows that the largest number offorest fires occur the year after ENSO events. The previous maximum number of fires for the period 1980-1997 happens to be 1988. ' The necessary and sufficient conditions for the development of forest fires were provided by the ENSO event of 1997-98, considered as the most intense of this century, namely: A very severe drought in most of the country associated to out of records high temperatures and strong winds. Maximum temperature for Mexico during the 1998 dry season was higher than the maximum temperature for the period 1941-1997. The burning for agricultural purposes Ipaper presented at the North American Science Symposium: Toward a Unified Framework for Inventorying and Monitoring Forest Ecosystem Resources, Guadalajara, Mexico, November 1-6,1998. 2 Ignacio Galindo is Professor and Principal Researcher at the Centro Universitario de Investigaciones en Ciencias del Ambiente, Universidad de Colima, Mexico. 3 Ramon Solano is Assistant Researcher at the same address. 68 produced also many uncontrolled fires. In spite of the different actions taken to reduce the forest fires risk, the final balance from January to June, indicates 14,302 fires affecting 583,664 ha (0.4% of the total forest surface). 73% corresponded to grazing land, shrubs and scrubs. 27% corresponded to forest burning in different degrees (i.e., 0.3% of the total forest surface). The average is 40.81 halfire, that is about 20% higher than the average for the period 1980-1997. Satellite data have been used increasingly during the past few years to examine burning in remote places. One of the primary sensors on board the NOAA series of polar orbiting satellites is the Advanced Very High Resolution Radiometer (AVHRR). This scanning instrument acquires data in five spectral channels, one in the visual range CO.58 £ I £ 0.68 mm), one in the near infrared range (0.725 £ I £ 1.1 mm) and three in the thermal range (3.53- 3.93, 10.3-11.3 and 11.512.5 mm). Data are sensed by all five channels simultaneously. The instrument has full resolution of 1.1 km at nadir. Although fire detection using satellite data comes back to 1977 when Croft (1977, 1978) presented views of agricultural burning in central Africa, the theory of fire monitoring using channel 3 was developed by Dozier (1981) and Matson et al., (1987). At present improvements on fire detection Table 1.-Forest fires in Mexico, 1980-1997* Year 1980 1981 1982** 1983** 1984 1985 1986** 1987** 1988 1989 1990 1991 ** 1992** 1993 1994** 1995** 1996 1997** Number of fires 4,242 2,740 5,599 6,087 6,120 4,386 8,482 9,263 10,492 9,946 3,443 8,621 2,829 10,251 7,830 7,860 9,256 5,163 Area covered (ha) Area I Fire (ha) 110,709 67,228 137,669 272,000 236,032 152,224 290,815 287,347 518,286 507,471 80,400 269,266 44,401 235,020 141,502 309,097 248,765 107,845 26 25 25 45 39 35 34 31 47 51 23 31 16 23 18 39 27 21 *Oirecci6n General Forestal, Subsecretarfa de Recursos Naturales. SEMARNAP, 1998 **ENSO events USDA Forest Service Proceedings RMRS-P-12. 1999 2. T3 ~ T4 + 10, where T4 is the AVHRR channel 4 temperature. It ensures that the hot bare soils are not confused as fire pixels. 3. Albedo Al < Albedomax (2-4%), this is a masking procedure to avoid pixels having a high albedo due to clouds and ground features. This condition is applied to NOAA 12 images received both near to sunset and sunrise. 4. 268 <T4 < 303 K, this condi tion discriminates false "hot points" such as water, water clouds over land, hot rocks, active volcanoes, etc. methods are made introducing temperature thresholds outlined for channel 3 temperatures with respect to channel 4 temperature (Christopher et al., 1998). Similarly, a technique to estimate satellite-derived burning areas is now in use (Cahoon et al., 1992). In what follows we present a multispectral method of forest fire detection. The method aims a more complete description of burning and smoke identification, geolocation, and area estimation. This method is applied for detection of the 1998 Mexican forest fires. All AVHRR channels are used in order to have a more complete description of burning and the smoke plume :. Data and Methods _ _ _ _ _ __ B) Image Composition Since April 1994 we have in operation a real-time NOAA polar-orbiting satellite ground receiving station. The AVHRR LAC images from NOAA-12 and NOAA-14 are used in this analysis to map fires and smoke. Forest fires points overlay on a visible image (channell + channel 2). Channel 2 offers more ground features and it is more transparent to aerosol. The combination of both channels produces a compensated image with a clear smoke plume over ground features. Since fires can be considerably smaller than the maximum resolution (-1.1 km by 1.1 km) the data set needs to be corrected in order to obtain subpixel size high-temperature sources. Other necessary correction is the removal of water bodies temperatures initially identified as "hot points" determined with channel 3 data. The number of pixels with forest fires is distributed for each month in a matrix array. Each row number corresponds to the day of the month and the column number contains its geographical coordinates distributed by states. Finally, a multispectral data set is constructed containing all forest fires and the smoke detected in Mexico from January to June 1998. ,~ A) Fire Dectection To locate forest fires it is necessary to determine the geographical coordinates from any pixel in the image located by its row, or scan line number S and within the scan line by its column, or pixel number P. The geolocation process, i.e., the identification of the values of W (longitude) and N (latitude) for each pixel (S,P) is electronically made through a third order polynomial function. Channels 1 and 2 data identify smoke loading and surface characteristics. Channel 3 data provide information during night on fires as ''hot spots" (Matson et al., 1987). Surface temperature in the fire neighborhood is determined using channels 4 and 5 data. Several detection schemes are used, however, these methods are dependent on local conditions, a method applicable over the Amazon was developed by Kaufman et al. (1990), recently this method is improved by Christopher et al. (1998). The method here used is adapted to local features. A pixel is classified as fire if the following conditions are met: Results _ _ _ _ _ _ _ _ _ __ The total number of pixels per month and state is shown in Figures 1 to 6. It is noticeable that from January to April the total pixel average per month was from 27 to 90, however for May it went to 652 to decay in June to 100. The first five places for the maximum number offires per state is shown in Table 2. Whereas the temporal forest fire distribution from January to June 1998 is shown in Table 3. 1. T3 ~ T3 min , where T3 is the AVHRR channel 3 temperature, T3 min @ 299 K. It ensures that false hot pixels are counted as biomass burning. 160 149 143 140 en 120 - 100 ~ '0:. 0 ~ 80 80 (I,) .c E 63 60 Z 56 -:tA :\.'i 40 w~ 18 21 20 o 49 47 :::J 0 2 0 25 18 III o 0 ~ • I I I 23 _ 16 24 • II I I I A ! Ii I • 7 0 1 o 0 /. ,,~ ,,'7/ ~~ I'~~~~#~A-~J'rJ>~~V~~f:-0" ~.f-P' ff#,&> ",.#' .:I'/"~~(;o~:#.p~~#""/.~V~~~/.f"~ffi'~rf .f~d' ~' ~ ~ <f<f~';:; <9' /o~ ~#' c;~' ~ # ~c:fF ~ ~f ~./:''f'tf'.:? <:j '" ~~ .t::l' $ ~ #" Figure 1.-Forest fires per state detected on real time using AVH RR channel 3 data. January 1998 USDA Forest Service Proceedings RMRS-P-12. 1999 69 ~ 3+4 350 304 300 I/) 270 Q) ,~ 250 0 200 \... 266 Tl7 Q) ..c E 150 Z 100 127 II .. ::::::I 66 113 ~ • -" ~I I I i I I I 55 .. 7 50 51 9 0 I.l'f 119 ... 53 .. 3 ..2 II 2 2 Ii 0 16 o • 0 Figure 2.-Forest fires per state detected on real time using AVHRR channel 3 data. February 1998 .. 00 343 350 I/) 300 ~C. 250 0 L- 245 241 ,,~ 223 190 200 ID ..c E iAn 150 ::::::I Z 111 100 a 1 2 • I• I 57 44 I I 21 R .. 74 I SO 100 99 u 1 :: .,'" 19 • I ~ 2 45 : :ill a 3 10 Figure 3.-Forest fires'per state detected on real time using AVH RR channel 3 data. March 1998 500 438 ... 50 .. 00 367 361 I/) ~~, (j) 350 X '5. 300 0 250 L- 252 ID ..c E ::J Z 200 143 150 1" R<; 100 44 .44. SO 0 0 0 97 R~ I I I I zo o • .,., :\.<; • -' I .. il 'Ul 1 11 18 0_. 14 • • 45 I 3 a 4 Figure 4.-Forest fires per state detected on real time using AVHRR channel 3 data. April 1998 / 70 USDA Forest Service Proceedings RMRS-P-12. 1999 5000 4394 4500 4000 1!2 3500 .~c..3000 ~ .- -~'" 2676 2500 2188 Q) .0 2000 E ~ 1000 661 500 3 0 0 • 1/;':0 . 1461 1500 A.,., ~"I"> I : - ~ 1118 94 • 941 505 I <>1':1 I~I. • '<:'1" 4 0 ~I <>An 386 "ld.~ 57 • • 12 I 0 67 5 47 _ 0 - 1200 1019 1000 In ~ - 800 c.. ~ 600 Q) .0 5 380 371 400 Z 269 210 1R7 200 0 8 =~ 81 - I 54 25 10 •• 3 • 37 49 _ 61 1 . 65 80 _ __ 24. 3 9 . . 32 24 I 46 1 o • 53 Table 2.-Maximum number of fires per state during the 1998 dry season. January February March -------------------------------Coahuila 149 Sinaloa 344 Jalisco 143 Puebla 304 Tamaulipas 80 Coahuila 270 Guanajuato 63 Jalisco 266 Veracruz 56 Tamaulipas 177 April May June - - - - - - - - - - - - - - - No. of Pixels- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Michoacan Coahuila Guerrero Durango Jalisco Table 3.-Temporal distribution of forest fires during the 1998 dry season. Month Total number of pixels January February March April May June 893 2,603 2,533 2,864 20,870 3,227 Total 32,990 USDA Forest Service Proceedings RMRS-P-12. 1999 343 245 241 230 223 Durango Oaxaca Michoacan Guerrero Edo. de Mexico 438 367 361 331 252 Chiapas Jalisco Durango Guerrero Oaxaca 4,394 2,782 2,676 2,188 1,726 Durango 1,019 Sonora 380 Guanajuato 371 Tamaulipas 269 Edo. de Mexico 210 The geographical distribution of all the detected forest fires in Mexico determined from AVHRR data from January to June 1998 is shown in Figure 7. Figure 8 shows a composite image depicting fire points and the haze and smoke layer originating in the Southeast from Mexico and Central America spreading over the Gulf of Mexico. Smoke trajectories are conformed according with the wind pattern. Figure 9 shows the alignment of smoke plumes from fires detected on February 28, 1998 where wind blows northeast. 71 ~ ~ ? (J> "I' \t~ ~~~" Ii:( ~ • ~ ~ {) Figure 7.-Geographical distribution of forest fires in Mexico detected from AVHRR data. January to June 1998 72 USDA Forest Service Proceedings RMRS-P-12. 1999 Figure S.-Composite image (AVHRR channels 1 + 2,3 and 4) showing forest fires (dot points) and the smoke layer spreading over the Gulf of Mexico. May 8, 1998. USDA Forest Service Proceedings RMRS-P-12. 1999 73 Figure g.-Composite image (AVHRR channels 1 + 2,3 and 4) showing forest fires (circles) and the alignment of the smoke plumes in NE direction. February 26, 1998. 74 USDA Forest Service Proceedings RMRS-P-12. 1999 Conclusions ------------------------------ References ____________________ 1. From January to June 1998 were recorded in Mexico 14,302 fires affecting 583,664 ha (0.4% of the total forest surface). 73% corresponded to grazing land, shrubs and scrubs. 27% to forest burning in different degrees (i.e., 0.3% ofthe total forest surface). The average is 40.81 halfire, that is about 20% higher than the average for the period 19801997. 2. A unique AVHRR data set containing the geographical and temporal distribution of burning for Mexico (JanuaryJune, 1998) is now ready for analysis and assessment of the impacts on the biodiversity and the effects of smoke on the regional radiation balance. 3. The largest number of fires occurred in the states of Chiapas, Durango, Jalisco, Guerrero, and Oaxaca. 4. Burning reach~d a maximum in May covering most of the country. The smoke reached the Gulf of Mexico and the southern part of the United States. 5. The forest fire detection method is now operational, it works on real time basis. Cahoon, D.R. Jr., Stocks, B., J., Levine, J.S., Cofer III, W.R., and Chung, C.C., (1992): Evaluation of a Technique for Satellitederived Area Estimation of Forest Fires. J. of Geophys. Res., 97 D4, 3805-3814. Christopher, S.A, Wang, M.,· Berendes, T.A, Welch, R.A., and Yang, S.K., (1998): The 1985 Biomass Burning Season in South America: Satellite Remote Sensing of Fires, Smoke, and Regional Radiative Energy Budgets. J. Appl. Meteorology, 37, 661-678. Croft, T.A, (1977): Nocturnal images of the earth from space (Order Number 68197) (Reston, Virginia: U.S. Geological survey). Cited in Cracknell, AP., The Advanced Very High Resolution Radiometer (AVHRR). Taylor and Francis, London, pp. 534 Croft, T.A, (1978): Night-time images of the Earth from Space. Scientific American, 239, No.1, 68-79. Cited in Cracknell, A.P., The Advanced Very High Resolution Radiometer (AVHRR). Taylor and Francis, London, pp. 534 Dozier, J., (1981): A method for satellite identification of surface temperature fields of subpixel resolution. Remote Sensing of Environment, 11,221-229. Kaufman, Y.J., Tucker, C.J., and Fung, 1. (1990): Remote sensing of biomass burning in the tropics. J. Geophys. Res., 95, 9927-9939. Matson, M., Stephens, G., and Robinson, J. (1987): Fire detection using data from the NOAA-N satellites. Int. J. ofRemote Sensing, 8,961-970. Rodriguez-Trejo, D. (1996): Incendios Forestales. Universidad Aut6noma de Chapingo y Mundi-Prensa, Mexico. pp 167-170. Secretaria de Medio Ambiente, Recursos Naturales y Pesca (SEMARNAP) (1998): Los incendios forestales en Mexico, 1998. 36pp. Acknowledgments The authors are indebted to Mrs. Myriam Cruz for her assistance in preparing the manuscript. This research was partially sponsored by CONACYT (26001-T; 095PN-1297), Comisi6n Forestal del Estado de Michoacan and, Subsecretaria de Recursos Naturales (SEMARNAP). USDA Forest Service Proceedings RMRS-P-12. 1999 75