The correction for haze is done after the classification of the subscene. thirty 
For. the most part, preliminary classification of normally bright and th 
Canola/rapeseed fields in non-hazy areas results in rectangular or square field Albert: 
areas being classified. With haze, the regularity and relationship to the 
survey pattern decreases somewhat and the area classified increases 
  
dramatically. A sudden, local increase in area classified as Canola/rapeseed procedi 
was also used as a clue to the presence of haze. The edge of the haze area is Statist 
determined by the local change in classification and the general brightening of and sal 
the image. 
4 
Rather than correct for the haze, we modify the classification of the (Ryers« 
crop under the haze. This is done by visually following the variability of extend 
rapeseed fields into the haze to locate a "normal-hazy" field (which is usually involv: 
in a subset of what has already been classified). This field is trained on and 
the result checked against the general pattern in the non hazy area and against 
fields visually interpreted as ‘hormal-hazy Canola/rapeseed. When an acceptable 
classification within the haze is obtained, usually after further modification, s 
the previous non-hazy classification is deleted within the haze and the new project 
classification is inserted in its place. Areas can then be recorded for the techno! 
whole scene for both normal and hazy areas. shared 
reache« 
Clouds: Eight of 29 DICS scenes contained cloud. Using the USDA criteria | sensing 
employed by Hanuschak et al. (1979), whole counties under the cloud would be itemize 
removed. - Since the Crop District is the primary reporting base, we could not 
delete portions very easily. An alternate method (Ryerson et al., 1981)" was 1 
therefore used to impute under clouds using the percent crop in areas similar includi 
to those covered by cloud as representative of the crop under the cloud, under been « 
cloud shadow and, under associated haze. informs 
governi 
Reduction of Noise in the Classification: One of the problems with before 
classification of Landsat data using the parallelepiped classifier is the 
"salt-and-pepper" effect caused by single pixels either incorrectly classified 4 
as the crop, or cells missed within fields. The net effect is pixels randomly techno] 
confused with the crop and voids in fields that often (but not always) should system. 
be fully classified fields. Although it could be assumed that such pixels trainiı 
would be accounted for by the regression estimator, it was decided to reduce ' the wc 
the noise to permit production of acceptable crop type maps. The reduction of conduct 
noise is accomplished by applying a 3x3 pixel weak filter (a CIAS option) to should 
the Canola/rapeseed classification result. The result is a spatially tighter course 
classification with no isolated single cells -- and fewer voids within fields. doing | 
To assess the result of the filter, the classification is stored as two themes 
areas, but only one is filtered. In the Peace River region the filter 4 
typically resulted in a ten to twenty per cent reduction in area classified, Comp 
although the reduction was not a constant percentage. The variability in project 
percent area changed as a result of the spatial filter is being investigated. questi. 
provide 
RESULTS satell: 
indivic 
For the British Columbia portion of the Peace River District, the project 
standard Farm Enumerative Survey (FES) estimate of 50 914 acres of rapeseed was necess: 
based on 30 of 49 segments (excluding those without cropland) and had a respec! 
co-efficient of variation (C.V.) of 29.9%. The Landsat estimate of 51 603 SUPpori 
acres for the same area, available in early September, was based on only 12 task h 
segments (of 49) and had a C.V. of 31.2%. In this area, crop was imputed under involve 
large areas (up to 10x20 km each) of cloud. The high C.V. for Landsat is for ini 
believed to be a result of missing ground data and missing satellite data (due 
to cloud) over segments. At the time of writing, final results are not / 
available for the rest of the Peace River District or for Alberta's Crop form 9 
District 4a in that Census results have not yet been released and all partici 
statistics have not yet been calculated. In the Alberta portion of the Peace was pl 
392 
— °°... \ NM A mm A 
A 
EP