(Step 3) Correction of NDVI bottom part (Post TWO processing)
At the bottom part of time series NDVI, TWO method may convert NDVI values higher than actual values. In order to eliminate this effect, NDVI is corrected after TWO processing based on the assumption that, coming close to NDVI bottom, the decrease of NDVI is shrinking. This assumption just means that NDVI temporal curve at its bottom is convex downward. If NDVI value near bottom was converted by TWO method and also satisfies this assumption, conversion by TWO method was canceled and this NDVI value is considered as an actual NDVI without noise.

Investigation of low temporal frequency effects
The potential cause of low temporal frequency effects is the change of solar zenith angle(SZA) at the time of observation. In order to investigate these effects, NDVI for the same land cover condition with different SZA must be analyzed. These NDVI values after the conversion by TWO method were prepared by the following ways.

• homogeneous land cover areas consisting of about 300 4-minute pixels(approx. 1.5 by 1.5 degree) were extracted using global lad cover dataset (Tateishi 1997). The extracted land cover types are evergreen forest, deciduous forest, shrubland, Taklimakan desert, White Sand desert, and vegetation.
• The mode NDVI at a certain time for the extracted area was used.
• The maximum NDVI, minimum NDVI, NDVI at the maximum SZA, NDVI at the minimum SZA, in a temporal NDVI mode curve, were used.

By the above method, effects by misregistration and yearly shift of season were removed.

Before examining the effect by SZA, whether the effect by the change of satellite exists or not must be examined. In order to examine effect by the change of satellite from NOAA-7 to -9 and -11, the mode NDVI values at the above mentioned five types of land cover at the time of minimum Solar Zenith Angle(MinSZA), spring Solar Zenith Angle(Mid1SZA), fall Solar Zenith Angle(Mid2SZA), and maximum NDVI(MaxNDVI) for 12 years(1982-1993) were compared. Though we found the case of NDVI change due to the change of SZA at the time of satellite change, we could not find clear NDVI change for the same SZA value, the same land cover, the same season but the different satellites. That is, we cannot recognize the effect by the change of satellite.

Then, the relationship between NDVI and SZA was examined using 12 year's data as follows. Similarly to the above investigation, the mode NDVI values for five types of land cover at the same season such as the time of minimum Solar Zenith Angle(MinSZA), maximum Solar Zenith Angle(MaxSZA), spring Solar Zenith Angle(Mid1SZA), fall Solar Zenith Angle(Mid2SZA), minimum NDVI(MinNDVI), and maximum NDVI(MaxNDVI) were investigated. Lines in Figure 2 show approximated lines from (SZA, NDVI) plots for the same land cover and same season for 12 years. When SZA is less than 60 degrees, no apparent NDVI decrease was found. But when SZA exceeds 60 degrees, there are some NDVI decreases. Since the rate of these NDVI decrease varies area by area, correction of NDVI decrease is not possible using only SZA value. One of the reason of NDVI decrease at large SZA can be assumed by terrain shade effect.


Figure 2. Variation of PAL NDVI for different Solar Zenith Angle during 1982-1993 at various land cover types
Conclusions
By this study, authors developed Temporal Window Operation(TWO) method to remove high temporal frequency noises. Authors corrected NOAA/NASA Pathfinder AVHRR Land(PAL) 10-day composite NDVI data from 1981 to 1994 by the TWO method. And the use of the NDVI data only within 60 degree SZA is recommended for temporal analysis. Figure 3 shows the region with SZA less than 60 degrees. For example, most part of lands in the north hemisphere have less than 60 degree SZA from February to October, which is the recommended period for temporal analysis using PAL NDVI. The remaining problem of preprocessing for global land cover change monitoring using time series PAL NDVI data is how to remove atmospheric effects due to volcanic eruption.
Figure 3 (a) The region with SZA less than 60 degrees in June from NOAA-11


Figure 3 (b) The region with SZA less than 60 degrees in December from NOAA-11
References

• Tateishi, R and C.G. Wen, (CD-ROM) AARS Global 4-minute Land Cover Data Set, 1997
• J. Cihlar, and J. Howarth, Detection and Removal of Cloud Contamination from AVHRR Images , IEEE Trans.Geosci.and remote sensing, Vol. 32, pp. 58- 589, 1994
• B. Holben, Characteristics of maximum-value composite images from temporal AVHRR data , International Journal of Remote Sensing, Vol. 7, pp. 1417-1434, 1986
• James, M. E., Kalluri, S. N. V.,“ The Pathfinder AVHRR land data set: An improved coarse resolution data set for terrestrial monitoring,” International Journal of Remote Sensing, Vol. 15, pp. 3347-3363,1994
• Price, J. C.,“ Timing of NOAA afternoon passes,” International Journal of Remote Sensing, Vol. 12, pp. 193-198, 1991
• Singh, S. M.,“ Simulation of solar zenith angle effect on Global Vegetation Index(GVI) data,” International Journal of Remote Sensing, Vol. 9, pp. 237-248,1988a
• Singh, S. M.,“ Lowest order correction for solar zenith angle to Global Vegetation Index(GVI) data,” International Journal of Remote Sensing, Vol. 9, pp. 1565-1572,1988b
• Singh, S. M.,“ Lowest-order correction to GVI data for solar zenith angle effect,” International Journal of Remote Sensing, Vol. 10, pp. 819-825,1989
• Rao, C. R. N., Chen, J.,“ Inter-satellite calibration linkages for the visible and near-infrared channels of the Advanced Very High Resolution Radiometer on the NOAA-7, -9, and –11 spacecraft,” International Journal of Remote Sensing, Vol. 16, No. 11, pp. 1931-1942,1995
• Los, S. O.,“ Estimation of the Ratio of Sensor Degradation Between NOAA AVHRR Channels 1 and 2 from Monthly NDVI Composites,” IEEE Transactions on geoscience and remote sensing, Vol. 36, No. 1, pp. 206-213,1998