Intensity Correction

Pixels with lower and higher reflectance factors in the red and the near-infrared channels, respectively, give higher NDVI values, which result in higher NPP estimation in equation (5). Therefore it is very important to adjust image intensity at lower reflectance factors of ch1. The reason why setting the regressions, which cross 50% of reflectance factor, was to keep better intensity adjustment in dark pixels to make NPP estimation more accurate. This method will reduce the effect of misregistration of pixels and mixels caused by a coarse resolution since the coefficients of determination were between 0.821 and 0.972 for ch1 (P<0.01). For the case of ch2, it is important to adjust intensity of dense vegetation pixels with higher reflectance factors. Since differences of reflectance factors are great among land cover and vegetation types, effects of misregistration and mixels are serious in the regression analysis. We selected spatially and temporary homogeneous areas for the regression analysis to reduce these effects. As a result, the coefficients of determination were between 0.899 and 0.980 for ch2 (P<0.01).

These high values suggest that the pixels were sampled appropriately, and that the regression lines provided reliable intensity corrections.

Inter-annual changes of the averages were compared between the original and the corrected imagery. The averages of ch1, ch2 and NDVI of original data clearly changed with season (Figs. 1a, 2a and 3a). However, there were some exceptions to this pattern. For the case of ch1, there were a clear drop of the average from mid-1983 until early 1985, and a clear rise followed by another clear drop from the middle of 1991 until the end of 1993. The former and latter anomalies were caused probably by the eruption of Mt. El Chichon and Mt. Pinatubo, respectively. Atmospheric turbidity or optical thickness increased after these volcanic eruptions (Meteorological Agency of Japan, 1999), and this probably caused errors in atmospheric corrections during the PAL data processing.

The average reflectance factors of the original ch2 data shifted up at the end of 1984 and 1988, and dropped remarkably in 1995. These changes agreed with replacements of NOAA satellites (Kramer, 2002), and suggested poor calibration of each AVHRR sensor. Above all, reflectance factors of the original ch2 increased from 1989 until 1994 before a change of NOAA satellite (Kramer, 2002), and the reflectance factors dropped suddenly after the a) 16

11 —I_I_I—I_I—i—i—i—i—i—i—i—i—i—i—i—i—l

0 24 48 72 96 120 144 168 192 216 1982 84 86 88 90 92 94 96 98 00 Months since January of 1982 / Year

11 —I_I_I—I_I—i—i—i—i—i—i—i—i—i—i—i—i—l

0 24 48 72 96 120 144 168 192 216 1982 84 86 88 90 92 94 96 98 00 Months since January of 1982 / Year b) 16 r-15 -

0 24 48 72 96 120 144 168 192 216 1982 84 86 88 90 92 94 96 98 00 Months since January of 1982 / Year

Figure 1: Changes in average reflectance in the standard area (channel 1). Averages within areas between 40°N and 35°S were computed. (a) Original PAL data, (b) the corrected data. Anomalies were observed clearly in the original PAL data, while the anomalies were removed after the correction.

0 24 48 72 96 120 144 168 192 216 1982 84 86 88 90 92 94 96 98 00 Months since January of 1982 / Year

Figure 1: Changes in average reflectance in the standard area (channel 1). Averages within areas between 40°N and 35°S were computed. (a) Original PAL data, (b) the corrected data. Anomalies were observed clearly in the original PAL data, while the anomalies were removed after the correction.

Months since January of 1982 / Year

1982 84 86 88 90 92 94 96 98 00 Months since January of 1982 / Year

Figure 2: Changes of average reflectance in the standard area (channel 2). Averages within areas between 40°N and 35°S were computed. (a) Original PAL data, (b) the corrected data. The reflectance shifted up in 1985 and 1989, and the reflectance was increasing gradually. The reflectance dropped in 1993 suddenly and increased again after a change of NOAA satellite at the end of 1994. Similar anomalies with channel 1 also appeared in (a). On the other hand, the anomalies were removed after the correction.

1982 84 86 88 90 92 94 96 98 00 Months since January of 1982 / Year

Figure 2: Changes of average reflectance in the standard area (channel 2). Averages within areas between 40°N and 35°S were computed. (a) Original PAL data, (b) the corrected data. The reflectance shifted up in 1985 and 1989, and the reflectance was increasing gradually. The reflectance dropped in 1993 suddenly and increased again after a change of NOAA satellite at the end of 1994. Similar anomalies with channel 1 also appeared in (a). On the other hand, the anomalies were removed after the correction.

change. Then the intensity increased again after 1995. The same trend with ch1 appeared in late 1984 and 1993 in ch2, namely, a clear drop followed by a clear increase followed by another clear drop. These inter-annual trends suggested that the quality of the original ch1 and ch2 images was unreliable, and that the original NDVI, which was computed using the original ch1 and ch2, was also unreliable. As a result, the original NDVI showed a trend that was opposite to that observed in the original ch1 and ch2 in 1984 and between 1992 and 1993. We pointed out that the original NDVI tended to increase since 1982, although there were many uncertainties in the PAL data processing.

On the other hand, inter-annual change patterns of the corrected ch1, ch2 and NDVI were cyclic (Figs. 1b, 2b and 3b). Global vegetation conditions during the study periods were unclear, however, there was no report about a severe vegetation damage on a global scale. Vegetation probably grew as usual in most parts of the world during the study period. Thus, the corrected results were probably reasonable. On the other hand, it should be noted that

1982 84 86 88 90 92 94 96 98 00 Months since January of 1982 / Year

Months since January of 1982 / Year

Figure 3: Changes of average intensity in the standard area (NDVI). Averages within areas between 40°N and 35°S were computed. (a) Original PAL data, (b) the corrected data. Opposite trends of anomalies with channel 1 and channel 2 appeared in the original NDVI. The anomalies were removed after the correction.

Months since January of 1982 / Year

Figure 3: Changes of average intensity in the standard area (NDVI). Averages within areas between 40°N and 35°S were computed. (a) Original PAL data, (b) the corrected data. Opposite trends of anomalies with channel 1 and channel 2 appeared in the original NDVI. The anomalies were removed after the correction.

plants change their spectra greatly season after season in the near-infrared wavelength (Awaya and Tanaka, 1996, 1999; Kodani et al., 2002). There is a risk that the seasonal spectral changes interfere with the correction of ch2, although we sampled spectrally stable pixels for the correction.

Was this article helpful?

0 0

Post a comment