GIS and Spatial Decision Support System for Environmental Degradation Monitoring


2. Methodology
Considering the above-mentioned scenario, it was a consensus that we should not study the flood process (consequences) in the low lands, but the causes related to that phenomenon, which were located in the upper Taquari watershed. In this way, the geographical approach was delimited. The erosion process at the highlands are very dynamic, consequently it was realized that we should model the problem as a multitemporal one.

The use of static Geographical Information System (GIS), where the information would be organized (diagnoses) and some pre-selected analyses (prognoses) done only at one static time, would not be enough.

It was necessary to have a GIS system linked to a monitoring system, where we could spatially visualize and update the spatial database.

In this way, a system that could automatically extract land use and land cover changes from multitemporal satellite images was considered to be essential. This system aims to be simple enough so that the municipalities do not need highly qualified technicians to deal with Remote Sensing or Digital Image Processing Techniques.

Algorithms for automatic land degradation extraction from multitemporal satellite images (MATHER, 1999; RICHARDS et al 1999; CHAVEZ, 1996; ALVES et al 1996) were developed as part of the SDSS information input.

A hybrid algorithm, consisting basically of the usual unsupervised classification method, where an existing classification of the same area is explored to estimate the initial position of the centroids of each cluster, was implemented. Experiments were carried out to compare the performance of the proposed hybrid method with the conventional unsupervised method.

The evaluation uses images obtained in 1996 and 1999 of two representative sub-basins subjected to a rapid process of environmental degradation during the last few years. Another method based on the use of linear regression procedure was also implemented (JOHNSON et al, 1998; SONG et al, 2001).

This method is based on the assumption of a linear relationship between image bands across time. Pixels that change from one class in the earlier image to a different class in the later image (change pixels) behave like outliers and affect the accuracy of these techniques. In this way, the linear regression approach investigates methods to filter the training data from outliers before estimating the coefficients of the linear relationship between image bands. The procedure discriminates between change and stable pixels by evaluating how well the linear model fits to them.

In order to gather information that will feed the Spatial Decision Support System, the erosion processes and water contamination by agrochemicals are being monitored in three selected micro-watersheds. This information is being used to model the behavior of each representative micro-watershed.

The information gathered feeds a GIS system that, along with Remote Sensing data sources form an automatic land degradation detection system, monitoring land use changes and erosion processes throughout the years. In order to unravel the complexity of the Taquari issue we adopted a problem-oriented approach (DE GROOT, 1992).

However, what is considered a major problem to people located in the lowlands (siltation and floods, for example) may not be considered of utmost importance to those in the highlands (soil erosion). In order to streamline our problem analysis, we considered the highland portion of the Taquari watershed in our studies.

Page 3 of 5
| Previous | Next |