Mapping and measuring the troposphere pollutants originated from the 1997 forest fire in south east asia.
2.1 Method
All the procedures followed to map and obtain the amount of fire emission constituents were performed using the ERDAS Imagine (Digital image processing software) and SPSS softwares.
(i) Data calibration
Raw digital numbers (DN) of channels 1 and 2 of AVHRR data were first converted into percent albedo values. Such conversion is important because the resulting reflectance compensating for the in-orbit degradation of DN as a result of weather changes before and after the launch of AVHRR sensor into the space. The conversion of DN to radiance governs the following relationship.
Ai/ii = Si/iiC+ Ii/ii (1)
where, Ai/ii is the percent albedo measured by AVHRR channels (channels 1 and 2 in this study), C is the input data value in counts (DN) and Si/ii and Ii/ii are respectively the slope and intercept values for bands 1 and 2. These calibration coefficients were obtained from Rao and Chen, 1998. These values are updated at NOAA/NESDIS in the 1B data stream at approximately one- month intervals.
Table 1. The Si and Ii values for AVHRR channels 1 and 2
| Satellite | Si | Ii | S2 | I2 |
| NOAA- 14 | 0.1318 | -5.4050 | 0.1657 | -6.7938 |
(Source: Rao and Chen, 1998)
(ii) Geometric correction
Image to image registration technique was carried out to register all the 3 calibrated images (dated 17, 28 and 29 September, 1997) to a master image. The master image is a digital map of the corresponding area captured originally by digitizing a topo map and then rasterized to 1 km grid size to match the NOAA AVHRR LAC data.
(iii) Atmospheric correction
The atmospheric correction for this study was performed using radiative transfer model with ground truth parameters like temperature, relative humidity, atmospheric pressure, visibility and height from sea level and zenith angle (satellite parameter) determined from historic data (Table 2). An assumption of a Lambertian surface was made because it considers a perfect diffusion and therefore facilitate the calculation. Reflectance values (after compensating for scattering and absorption of atmosphere) were obtained after performing the sequence of the following calculations;
-
estimation of total optical thickness,
- estimation of atmospheric transmittance,
- estimation of total irradiance at the surface of the earth,
- estimation of the path radiance,
- estimation of the radiance sensed by the sensor and,
- estimation of the reflectance.
where,
R channel = reflectance after compensating for atmospheric attenuation, L
s = radiance sensed by the sensor, Lp= path radiance, T
q= atmospheric transmittance at
q zenith, E
g= global irradiance reaching the surface of the earth.
Table 2. Ground-truth parameters used in radiative trasfer model to estimate the reflectance values of AVHRR data.
| Parameters |
Averaged value |
| Satellite zenith angle (°) |
36.52 |
| Temperature (° C) |
27.28 |
| Relative humidity (%) |
83.40 |
| Atmosphere pressure (mbar) |
1011.86 |
| Visibility (km) |
3.74 |
| Altitude from sea level (m) |
14.16 |
(Source: Asmala Ahmad and Mazlan Hasim, 1999)
(iv) Cloud masking
It is necessary to identify and separate the cloud pixels from non- cloud pixels for the retrieval of smoke plume originated from forest fire. This is because of the enormous errors arising from cloud contamination when deriving atmospheric pollutants present in the smoke plume. In this study a simple technique (Q technique) which was proposed by Saunders and Kriebal (1988) was used.This technique is based on the ratio between the reflectance in the NIR and visible bands of AVHRR data.
Q = NIR / Visible (3)
Q values over cloud pixels are close to unity due to quite similar Mie scattering effects of the reflectance for both channels (Franca and Cracknell, 1995).
(v) Extraction of haze constituents
The measurement of space-borne remote sensing of gases and aerosol particles in this study is based on an assumed relationship between the ground measured aerosols and the spectral path radiance. Path radiance is detected by AVHRR sensor above a non-reflective surface and is the result of backscattering to space by particles and molecules in the atmosphere (Kaufman 1993). The amount of each of the 5 atmospheric constituents acquired at the 5 monitoring stations (dependent variables) and the calibrated reflectance from satellite data (independent variables) at the corresponding locations were used to establish empirical relationships between the aerosol and the scattered spectral path radiance. A multiple regression analysis (using both channels 1 and 2 of AVHRR) was carried out to predict some of the common but not validated assumptions about the spatial distribution and estimation of the amount of the particles over Peninsular Malaysia. The best model based on the best regression coefficients (r
2) for each of the constituents was selected to be applied to every pixel of images dated 17, 28 and 29 of September 1997. The regression line/model expresses the best prediction of the dependent variable (haze constituents) at other locations all over Peninsular Malaysia.
(a)
(b)
Figure 3. Cloud masking using Q technique, (a) image of Peninsular Malaysia before cloud masking and (b) after cloud masking.
