Correction of non-uniform aerosol effect of Landsat TM image by using blockwise approached atmospheric correction model (BACM)
Data description
There subseted Landsat TM images (1000*1000 pixels) which are the test sites of NSF LTER (National science Foundation, Long-Term Ecological Reseach) project are used to testify the model. Two images scanned in July 12 and July 28 1993 are located in Hog Island (HI), the other one scanned in June 22 1995 is located in Madison (MD). Their solar Zentith angles are all bout 30.50. the sunphotomether-measured aerosol optical detpths which are available from the AERONET (Holben et al. 1997) denote that HI-12jul 93 image with
t (TM1) =0/69 is much hazier than HI-28jul93 with
t (TM1)=0.25 (table 1), HI scenes are all in uniform atmospheric effect, whereas MD scene shows non-uniform aerosol effect (hazier in the lower part). HI scenes contain near sea-shore oceans and most of agricultural lands combined with crops and bare soils. The MDS scene contains different laandcovers, such as forest lake, urban and bare soil.
Results and discussions
Table 1 shows the derived aerosol optical depths (with /without iteration) and their error
D t to the field-measurements in TM1 and TM3 bands. Highly abnormal derived in (>4.0) by using the
u iterative procedure results large
D t(-0.14~0.36 ) in TM1 band. Reasonable
u should be in range of 2 (dust) ~ 4 (smoke) (Kaufman and Tanre 1992). Mean errors of
D t by iterated method are 0.25 and 0.12 in TM1 and TM3 bands which are larger than non-iterated method (0.14 and 0.05). In evaluation the accuracy of accuracy of retrieved
t in MD-22 jun 95 scene with non-uniform aerosol effect, the retrieved
t s of central four blocks are averaged (table 2). The ranges of retrieved
t (TM1) and
t (TM3) are 0.62 ~ 0.84 and 0.29-0.53 respectively, which account for the spatial variation of aerosol concentrations in MD-22jun95 scene. Retrieved aerosol optical depths are larger in lower part of image than upper part (e.g. 0.82 and 0.62 in lower left and upper left corner in TM1 band), which are also conforming to diverse haziness of MD-22jun 95 scene. After atmospheric correction, the non-uniform aerosol effect is well removed.
After above discussions, therefore, the accuracies of retrieved aerosol optical depth (i.e. 0.14 and 0.05 in TM1 and TM3 bands) should be satisfactory, since the error of retrieved surface reflectance is about 0.01 ~ 0.02 as shown in the previous study (Liu 1995).
Although surface reflectance measurements are not available, the typical bare soil and vegetation reflectance spectra can be obtain after atmospheric correction even in the very hazy part (e.g.
t (TM1) = 0.84 , visibility ~ 5km) of the MD -22jun95 (figure 1).
Figure 1. The apparent and surface spectra of bare and vegetation extracted from the very hazy part (visibiligy~5km) of MD -22jun95 scene.
Conclusion
An atmospheric correction model (BACM) is developed for correcting the non-uniform aerosol effect of Landsat TM images by using the blockwise approach. A sensitivity study of aerosol characteristics such as optical depth, single scattering albedo,phase function and Junge parameter has shown that large error of retrieved aerosol optical depth (~0.3) can be caused due to he error of derived Junge parameter in the iterative procedure. Therefore, a fixed model is used.
The results show that the accuracies of the retrieved aerosol optical depths are encouraging (
D t <0.2), by comparing the error of retrieved surface reflectance (0.01 ~ 0.02) made in the previous study. Non-uniform aerosol aerosol effect of image is also well corrected. After atmospheric correction, typical bare soil and vegetation reflectance spectra can e acquired.
