Barrett, Rachel 
  
temporal analyses produced a report of the various crops grown in that season. The results of the different iterations of 
the program were amalgamated to identify the crop or crops grown in each field area during the season. 
  
Raw TM, XS, XI Data 
Resampling, alignment 
Y 
Stacking 
Y 
Digitise AOI 4« 1:5000 orthophoto 
i map series 
v p Masking 
Initial image processing 
  
Principal Components Analysis 
(PCA) 
| 
Unsupervised classification process 
(maximum likelihood) 
. 
Training dataset 
PaddockID « 
(iterative classification) 
y Class types expected 
Classified field area content 
  
  
  
Figure 3. Diagrammatic representation of the image classification process. 
Different operators scored the results, so that the identity of the "unknown" field areas was not available to those 
performing the analysis. 
2.4 Accuracy Assessment 
The prediction results achieved were entered in a matrix table and errors of omission, comission, and normalised overall 
accuracy were calculated as described by Congalton (1991), with actual crop identity displayed as y coordinate (vertical 
axis) and the prediction displayed as x coordinate (horizontal axis) This representation of the data therefore reported 
not only the predictive accuracy of individual crop types but also exemplifies the outcomes of unsuccessful predictions. 
  
136 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.