Munro, Duncan 
  
3.1.2. Feature Analysis and Scale Comparison 
A multi-scale approach/analysis was chosen to establish scale limits for the identification of given surface elements in the 
LANDSAT-7 and LANDSAT-5 + SPOT products. A set of three scales (1:50,000, 1:35,000 and 1:20,000) was chosen 
and a comparison was then performed analysing the possibility of clearly identifying elements belonging to two main 
groups of objects: (1-10) built-up anthropic features and (11-16) natural landscape features. They represent a heterogene- 
ous set of objects with a degree of identification likelihood varying with the scale. They are: 
1 - highway 9 - isolated building (housing) 
2 - hard surface road 10 - isolated building (industrial) 
3 - loose/light surface road 11 - orchards 
4 - paths 12 - crops 
5 - railroad 13 - pasture and grassland 
6 - bridges 14 - scattered trees/bushes 
7 - viaducts 15 - woods (low density) 
8 - built-up areas 16 - woods (high density) 
The scale selection derives from a simple decision process: at the scale of 1:50,000 the two products are very similar pro- 
viding excellent performances while at 1:20,000 the pixel effect begins to degrade the outline of different features and the 
resolution of the LANDSAT-7 product becomes insufficient to detect the features. The 1:35,000 scale represents an inter- 
mediate situation. 
As this part of the comparison is the most subjective, the possible error has been minimized by choosing readily identifi- 
able features in order to reduce the possibility of getting different evaluations due to subjective interpretation which 
becomes more common as complex elements are introduced. Additionally, the possibility of discriminating between for- 
est cover types, etc. was not considered a priority because this issue was addressed in the multi-spectral classification part 
of this study. The simplification of feature classes was also adopted to avoid case specific results both in terms of scale 
and location/environment. The recognition likelihood - for each of the 16 classes and 3 scales - was rated using three pos- 
sible levels: (1) low; (2) medium; (3) high. It has to be considered that (a) the results represent an average of the overall 
situation and (b) no other intermediate/heterogeneous possibilities were taken into account. For example different ele- 
ments of a feature class may not be consistently identifiable all over the area but when detected cannot be confused with 
other feature classes. Alternatively, there may be class objects that most of the time are readily detectable but are difficult 
to assign to a feature class because of some overlap between similar classes. Apart from the role of subjectivity in such 
estimations, it was considered impossible to field check, in a reasonable time, all possible occurrences. The topographic 
maps were therefore used as a reference. Considering the 16 possible elements listed previously, the results are pre- 
sented, according to the 3 previously chosen scales, in the following table (note that only the 1999 LANDSAT-5 + SPOT 
merge and the LANDSAT-7 ETM+ image fusion products were considered in this comparison): 
  
  
  
  
  
  
  
  
Features-> 1d1213l4]5]617.].84] 9.]410].1]|12] 13] 14 |. 15 ] 16 | avg. 
1:50,000 ETM+ 3 | 1/1 1212} 2/1} 31 1.) 2/1 } 3 | 2 | 1 [ 4 | 2 | 1.75 
1:35,000 ETM+ 312 | 1 |} 2112} 3} 2) 53 | 1°} 2° | 2} 3} 2 |} 1! 2 3 } 2.15 
1:20,000 ETM+ 31 3/2 | 3 | 2} 3/3 | 3 | 2 | 2 |. 2.3 |. 3-4 2.1 3:13 1 2.65 
1:50,000 TM+SPOT | 3 | 1 ; {| 2 2) 3.2} 3} ll ° 2; 1 | 3 2.1. 2.| 1} 2 [194 
1:35,000 TM+SPOT! 3 |} 3 2 | 2! 2 5313} 3 {| 2 | 3 {| 2 {3} 3 [3 | 2} 3 | 2,63 
1:20.000 TM+SPOT | 3! 3 | 3{|3|3}3) 3) 3/3} 3) 3/3! 3/3} 3 | 3 3 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Taking into account the limited consistency of these results due to the subjectivity issue, some general conclusions could 
be drawn by considering any introduced error to be homogeneously distributed. The incorporation of SPOT panchro- 
matic data clearly leads to a better definition of objects, a difference with LANDSAT-7 that becomes more perceptible at 
larger scales. In a scale to scale comparison the LANDSAT-5 + SPOT product is never poorer than the LANDSAT-7 
image fusion product. Further comparative studies could be defined according to different interpretation purposes. Figure 
1 shows an example of LANDSAT-7 Band 8 and SPOT PAN over the area of Almaraz. 
  
936 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 
  
  
cot 
eve 
ind 
wit 
An 
En 
rel: 
LA 
the