ect, 
dol- 
Jefi- 
s in 
| be 
ure. 
tern 
rest 
it is 
ore 
and 
s in 
rors 
10d- 
ancy 
t for 
be 
ancy 
cant 
ven- 
licy. 
>) is 
lods 
> SO 
the 
both 
eris- 
con- 
e. A 
vere 
 fig- 
ure 1). Each test site was 3km. x 3km. in size. These 
test sites were selected to cover as much variety of land 
cover as possible and they ranged from concentrated ur- 
ban areas (Sites 5, 6), to agricultural areas (Sites 3, 7, 8, 
10, 11), forested areas (Sites 1, 2, 9, 10 ) and natural 
vegetation areas (Sites 2, 4). 
On each site two separate types of survey were made. 
The first was concerned with producing a new land cover 
map following the convention of the CORINE project. 
The aim of this was mostly to understand the accuracy 
of the original (1985) map and its reproducability when a 
different mapping team is used. The second type of sur- 
vey was concerned with recording areas of landscape 
homogeneity that could be used in experiments on im- 
age classification. For this type of survey very detailed 
information was recorded such as vegetation type, can- 
opy conditions, stand height, soil type, soil colour, soil 
wetness, and percentage cover etc. 
All data obtained in the field surveys were digitised and 
entered into an ARC/INFO system. The original Portu- 
guese CORINE map was also available in ARC/INFO 
format allowing inter-comparisons to be made. 
  
    
  
N PORTUGAL 
| 11 10 
a, 
1991 EXPERIMENTAL 
: LAGOA DE ALBUFEIRA 
SETUBAL 
AZEITAO 
GAMBIA 
BELEM 
LOURES 
ALENQUER 
SAMORA GORREIA ‘A 
10: CORUCHE 
11: SAMORA CORREIA ‘B’ 
ATLANTIC 
OCEAN 
m 
NO 00-30 Uu coto rà 
30 km. 
  
  
  
Figure 1. Test Sites Used in 1991 Field Experiments 
Multitemporal sequences of digital Landsat Thematic 
Mapper imagery were also acquired from 1985 (i.e. at 
the same time as the original mapping exercise) and 
from 1991 (coincident with our own revised field survey). 
The images were rectified and co-registered with the 
digital map information in the UTM projection. 
3.2 Land Cover Map Reproducability Studies 
Following the assembly of all the above-mentioned map 
and image datasets with a 6 year time gap, it has been 
possible to undertake a variety of studies on change de- 
tection and map revision besides on original map accu- 
racy. Most of the work to date has been concerned with 
the inter-comparison of the 1991 survey CORINE-type 
land cover map and the original 1985 Land Cover map 
product on the the eleven mapped test sites. Typical re- 
sults from this are shown in figures 2 and 3. 
Figure 2 shows such a comparison on a coastal zone 
test site which contains a lagoon (running from upper 
right to lower left of each box) and areas of forest, heath- 
land and agriculture. In fact this area has seen little land 
cover change over the 85-91 time period and the lagoon 
and beach front (to the far left of each box) impose a ba- 
sic similarity on the two maps. However in the zones 
545 
  
  
  
  
  
  
       
  
  
      
924 312 
322 242 \ 312 
211 
P unm 
523 3 
^h 331 512 
312 
322 
243 311 
1985 CORINE MAP 
Subjective mapping 
differences in this area 312 
321 
523 ^ 
N 
EN 
  
  
Major subjective 
class interpretation 
difference 
  
  
  
  
1991 JRC - EMAP SURVEY MAP 
Figure 2. Intercomparison of Original Land Cover Map 
and JRC 1991 Survey - Lagoa de Albufeira Test Site 
above and below the lagoon there are considerable 
mapping differences. These differences involve both dif- 
ferent parcel boundaries and different land cover class 
labels. In general these differences arise from the sub- 
jectivity which is inherent to the CORINE mapping proc- 
ess. In many areas the European landscape, especially 
in the Mediterranean zone, contains very mixed terrain 
which is not easily categorised by the 44 classes of the 
standard nomenclature. Hence a considerable degree of 
subjective judgement comes into play when humans are 
forced to make the landscape conform to this model. 
This subjectivity is readily revealed by the intercompari- 
son of figure 2. One of the most striking differences be- 
tween the two maps of figure 2 concerns the parcel 
which was labelled class 243 (land principally occupied 
by agriculture with significant areas of natural vegeta- 
tion) in the 1985 survey. This is in fact an area of holiday 
homes with large gardens and natural vegetation areas 
as ground inspection reveals. The original map class is 
not a completely valid description of the area. In the new 
survey of 1991 the 'recreational' nature of the area was