In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
582
Photo 4 (Part of Semi-Arid Region of Piaui State)
Geographic Information Required
Data about various components required for this research have
been gathered from various, Federal, State, and Municipality
Agencies, such as, Land Use, Soil, Soil Conservation, Slope and
Elevation, Drought and Flood, Climate (Precipitation, Temperature
and Humidity), Geology and Hydrology, Vegetation and Forest,
Irrigation and Drainage, Socio-Economic, Municipality and State
boundaries etc.
3. Individual land use and cover classifications should be
customized to facilitate interpretations of digital images with
different resolutions.
Image Processing
A 1000 by 1000 pixel sub-scene of LANDSAT-TM and SPOT
multi-spectral data (band 3,4, 5 & 1,2,3) were used for image
analysis. More than 40 sites were visited in the study area, and
reference data, such as soil, vegetation, geology, topography,
climate and others are made to assist in supervised classification
(Maximum Likelihood Classification-MAXCLAS). Various
field trips served as a basis for accuracy assessment to derive
various earth resources information. The digital interpretation was
checked by three field trips. The relevant statistics, such as mean,
mode, medium, standard deviation, variance and co-variance
matrices wwere applied for our study. After inspection of the
digital classification combined with the field work, finally
resulted into 15, 17 and 12 categories of land use/land cover
classification in Paraiba, Piaui and Ceara states at the Level II.
(Anderson et al., 1976). The accuracy assessments of the
transformed and no-transformed LANDSAT-TM and SPOT image
were concluded to compare the best areas of known reference
data with the same areas on Level II land use and land cover
classification on a pixel by pixel base produced by supervised
classification. The over all accuracy was found more than 85% in
all the three areas. By using RECODE program of ERDAS
Software on land use/land cover information resulted into 11,11
and 6 categories of soil associations in each area. Re-coding was
possible because of the high degree of correlation of land use and
land cover with the features of other maps. Field observations
conducted at the sites confirmed this relationship.
Programs of ERDAS used for study
Following programs of ERDAS Software in systematic sequence
were used for unsupervised, supervised classification & accuracy
assessment.
For Unsupervised Classification:
READ-CLUSTR-DISPLAY-COLORMOD-CLASNAM-
RECODE-COLORMOD-CLASNAM-ANNOTAT-CLASOVR-
BSTATS-LISTIT
For Supervised Classification:
READ-SEED-SIGDIST-SIGMAN-ELLIPSE-CLASNAM-
MAXCLAS-DISPLAY-COLORMOD-CLASNAM-ANNOTAT-
CLASOVR-RECODE-INDEX-RECODE-INDEX-COLORMOD-
CLASNAM-ANNOTAT-SCAN-BSTATS-LISTIT.
For Accuracy Assessment:
READ-DISPOL-DIGSCRN-GRDPOL-CLASOVR-CLASNAM-
SUMMARY.
Criteria used for land use classification
For our study of semi-arid regions of NE Brazil, the land use
and cover classification system (Anderson et ah, 1976) is
modified in accordance with the local climate, local needs and
existing conditions. During the conduct of our project, we used
the following important criteria:
1. The interpretation accuracies in the identification of land use
and land cover categories from remote sensor data should be 85%
or greater.
2. The multiple use of land should be recognized where possible.
Digital Image Interpretation Procedure
Subset from SPOT Scene
Ground Truth
Unsupervised Classification
n^n unsupervised <^1;
Select Training Areas ^ Select Test Areas
Supervised Classificatic
Derive Level I & II with Modifications
Accuracy Assl Usinent
▼
Geometric Correction
I
Derive Land Use/Land cover Map
Derive Soil Associations Map
1
Assess Utility of Product