Correction of time series NDVI by the method of Temporal Window Operation (TWO)
The temporal window operation (TWO) procedure on the NDVI trajectory of each pixel by finding "low" NDVI value and replacing these by linearly interpolated. Values of the NDVI profile. Two algorithm start at the beginning of the NDVI (start point) curve and checks whether the NDVI for the current period is equal to or greater than the previous NDVI value within the window. If it is higher, current value is assigned as the start point of next window (window 1,s1-> s2). If there is no higher value within the window, select the biggest value as a next start point and replacing these by linearly interpolated value from current start point to next start point (window 6,s6 -> s7).
Fig.1 Correction of time series of NDVI by the method Temporal Window Operation (TWO)
Data
NDVI time series for thirteen years, from 1981 to 1994, were taken 10-day composite data NOAA/NASA pathfinder AVHRR Land Data Set. These data were calculated from the calibrated channel 1 and 2 data:
Where R
ch1 and R
ch2 represent top- of the atmosphere reflection of AVHRR channel 1 and 2, respectively. Each full composite image covered whole global land contained over 1.458 x 10
7 pixels.
Result
Two is sensitive to length of the window size; too short period will retain too much noise, while too long period will tend to smooth out important changes. Fig. 2. Shows the difference between original NDVI data when window size is 5 (every 5 10-day composite data). The result of this correction method is different depending on the window size. For the vegetation type shows that almost no change of NDVI profile through year, such as desert area and tropical forest area, the total vegetation biomass change can be big, the extraction of short-period vegetation change is difficult. The window size of 4-5 is appropriate for the two times planted area. For the area where there is a peak vegetation time in one year, like deciduous area, the window size of 5-6 is better. Usually for global vegetation research using 10-day composite data, the window size 4-5 is suitable.
Fig 2. NDVI profile from AVHRR 10-day composite data : 10-day composite raw data and profile using the TWO method (window size=5), a) deciduous area , b) desert area , c) tropical forest area , d) two times planting area.
Comparison of the MVC, MVI and TWO techniques
The problem of the commonly used MVC to vegetation monitoring is that no matter how long the maximum value occurs in the composite period, it is assigned for the value of the 16
th in each month. To solve this problem, the Maximum Value Interpolation (MVI) was proposed.
