Aigner, Edgar 
  
satellites of the NOAA-series. The high temporal resolution, however is at the expense of the spatial 
resolution. The AVHRR-sensor has a spatial resolution of 1.1 km in nadir at a temporal resolution of one 
daytime overpass. One of the objectives of this research was to take advantage of the AVHRR's high temporal 
resolution at the largest possible scale. Another objective was to develop a method to predict crop yield, that 
meets the requirements of the local farmers in the Eastern Wimmera. 
2. DATABASE 
2.1. The AVHRR-Data 
A system summary of the AVHRR-sensor is given on the World Wide Web site of NOAA's polar data user 
guide (NOAA POD Guide). Observations by the NOAA- 14 were used exclusively. 
Data for the growing seasons of the Eastern Wimmera (May Table 1: Spectral Bands of the NOAA-14 
to December) from 1995 to 1997 were processed using AVHRR (http://www.ncdc.noaa.gov). 
  
  
  
  
  
  
CSIRO standard routines for calibration navigation, Channel # [Band width [um] [Spectrum 
geometric correction and cloud masking (DILLEY, AC, 1 0.58-0.68 VIS red 
ELSUM, C.C. (1994), DILLEY, A.C., EDWARDS, M. 2 0.734.10 NIR 
(1998)). The short—wave channels 1 and 2 (see table 1) were 3 3.55-3.93 MIR 
used for calculating the NDVI, 4 10.3-11.3 TIR 
5 11.542.5 TIR 
  
  
  
  
  
NDVI = (Ch2 — Ch1) / (Ch2 + Ch1) (1). 
For atmospheric correction a maximum value 
composite (MVC) technique was applied to the 
NDVI time series (see e.g. B.N. HOLBEN 
Figure 1: NDVI — MVC time series for the 3*3 pixel target 
ID 26 1997 47% wheat. 
  
  
  
  
  
  
  
  
  
  
(1986)). For daily values, NDVI was NDVI MVC Tire Series 
interpolated linearly between the cloud free to Peer 
observations. Figure 1 shows the MVC of > l 
NDVI of a 3*3 pixel subset (47 % of the area 074 
was wheat) in 1997. It represents the typical | 064 — NDVHWC 
course of the NDVI-MVC of wheat in 1997. 3057 
This is also valid for the sudden decrease o A, HE RR 
around day of the year (DoY) 210, which was n. 
due to drought. The size of single paddocks in 01} Was 
the Eastern Wimmera is usually in the order 1 ght ie 
120 150 180 210 240 270 300 330 360 
DoY 
km?. 3*3 pixel subsets, thus guarantee that the 
paddock of interest is covered by the 
observation subset at the estimated geometrical 
accuracy of one pixel (DILLEY, A.C., ELSUM, C.C. (1994)). In the AVHRR channels 4 and 5, thermal 
infrared radiance emitted by the earth-atmosphere system is measured. The atmosphere's influence was 
removed using a split window technique (e.g. refer to A.J. PRATA 1994). As split window coefficients, those 
determined empirically by A.J. PRATA (1994) for wheat paddocks in south—eastern Australia were applied. 
Coefficients were available for “bare soil”, “maximum green vegetation cover” and “mature crop”. Within the 
NDVI — MVC curvature, the bare soil signal corresponds to the low NDVI at the beginning of the growing 
season, after sowing and before emergence; maximum cover corresponds to the maximum NDVI occurring 
and mature crop corresponds to a NDVI value just before harvest, early in December. Y.H. KERR et al. (1992) 
describe how to use the NDVI signature for deriving split window coefficients adapted to the changing states 
of plant growth by calculating the fraction between two stages and re—calculating the coefficients using this 
fraction. This idea was applied to the data of the Eastern Wimmera. Thus, a good approximation of daily land 
surface temperatures was available. 
  
  
  
2.2. Meteorological Data 
The weather data used, was made available by the Australian Bureau of Meteorology (BoM), situated in 
Melbourne. Information applied was daily maximum, average and minimum temperatures from three stations 
and daily rainfall from seven stations within or close to the study area. To approximate the meteorological 
conditions within the targets, spatial interpolation between the station was carried out. As there was no 
significant spatial variation in temperature, nearest neighbor interpolation was sufficient. Rainfall, however, 
had a very high spatial variability, so the data had to be interpolated using the inverse of the squared distance 
as a weight. 
  
20 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.