Final Activity Report Summary - MONITORING LAND USE (Mapping, monitoring and modelling land use and land cover change in the Mediterranean region using remote sensing and ancillary data ...)
This project investigated the performance of regional scale mapping, monitoring and modelling techniques of land use and land cover and their changes through time. The research focussed in three Mediterranean test sites, namely an area in the south of the Spanish Province of Valencia, the Aegean island of Lesvos and an area in the outskirts of Athens in Greece.
The project brought together some of the latest advances in the field of remote sensing data processing methods, Geographic information systems (GIS) and land use change modelling. More specifically, it applied novel approaches and invested considerable effort in:
1. the accurate integration, by means of conditional probability networks or Bayesian expert systems, of multiple sources of geographical data in a way that allowed for the assessment, representation and propagation of uncertainty through the process;
2. the calibration to 'like-values' of satellite images recorded at different times, in order to enable meaningful comparisons of the digital data;
3. the accurate mapping of the land use or cover changes that occurred in the last decades by means of post-classification comparison;
4. the identification of the most vulnerable areas, i.e. 'hot-spots'; and
5. the investigation of the role of spatial metrics in the performance of an Artificial neural network (ANN) model for predicting land use and cover changes.
More specifically, in the case of the site between the Spanish provinces of Valencia and Alicante, multi-temporal mapping of the woody perennial vegetation with Landsat data spanning 15 years was the objective in question. Overall, accuracies were high and ranged between 97 and 99 %. It is interesting to note that the percentage correct for woody cover for the two Landsat ETM+ images reached that of perfect estimates (100 %).
In the case of the Greek island of Lesvos, the multi-temporal mapping of the changes in land use and cover was the main objective. The accuracy assessment involved land use classification accuracy and land use change accuracy before and after the multi-temporal mapping approach which involved the use of Conditional probability networks (CPN). The land use and cover mapping results showed that the CPN improved accuracies for almost all types and years. For the accuracy assessment of land cover and use changes a comparison between the satellite images, the land use maps and the reference data was made, in order to draw conclusions on the plausibility of the changes and create a summary table that provided a quantitative description of the accuracy of the mapped changes. The accuracy achieved prior to CPN (pre-CPN) was weighed against the one acquired by the fuller dataset which allowed for the application of the multiple-year approach (post-CPN). Accuracy figures for the areas mapped as 'changed' and 'not-changed' for the period between 1987 and 2000 were higher when the multi-temporal processing was applied. The percentages were high for the unchanged areas, pre-CPN and post-CPN alike.
In the case of the Athens, post-Olympics urban expansion mapping exercise, two land use and cover classification methods were applied in the area of Messogia in the east periphery of the city, namely a pixel-based methodology based on a supervised Maximum likelihood (ML) classification and an object-based segmentation approach. The obtained results showed best performance for the object-based classification. It was also concluded that both classification schemes were improved by the rule-based multi-temporal approach.
Finally, the role of spatial metrics in the performance of an ANN model for predicting land use and cover changes was investigated. The model outputs were validated against a land-use change map, which was derived from ortho-rectified aerial photographs, Landsat TM and Quckbird data. The results from the application of the model before and after the use of spatial metrics indicated that the model could predict the patterns of change in the island's olive-groves reasonably well when parameterised with spatial metrics.
The project brought together some of the latest advances in the field of remote sensing data processing methods, Geographic information systems (GIS) and land use change modelling. More specifically, it applied novel approaches and invested considerable effort in:
1. the accurate integration, by means of conditional probability networks or Bayesian expert systems, of multiple sources of geographical data in a way that allowed for the assessment, representation and propagation of uncertainty through the process;
2. the calibration to 'like-values' of satellite images recorded at different times, in order to enable meaningful comparisons of the digital data;
3. the accurate mapping of the land use or cover changes that occurred in the last decades by means of post-classification comparison;
4. the identification of the most vulnerable areas, i.e. 'hot-spots'; and
5. the investigation of the role of spatial metrics in the performance of an Artificial neural network (ANN) model for predicting land use and cover changes.
More specifically, in the case of the site between the Spanish provinces of Valencia and Alicante, multi-temporal mapping of the woody perennial vegetation with Landsat data spanning 15 years was the objective in question. Overall, accuracies were high and ranged between 97 and 99 %. It is interesting to note that the percentage correct for woody cover for the two Landsat ETM+ images reached that of perfect estimates (100 %).
In the case of the Greek island of Lesvos, the multi-temporal mapping of the changes in land use and cover was the main objective. The accuracy assessment involved land use classification accuracy and land use change accuracy before and after the multi-temporal mapping approach which involved the use of Conditional probability networks (CPN). The land use and cover mapping results showed that the CPN improved accuracies for almost all types and years. For the accuracy assessment of land cover and use changes a comparison between the satellite images, the land use maps and the reference data was made, in order to draw conclusions on the plausibility of the changes and create a summary table that provided a quantitative description of the accuracy of the mapped changes. The accuracy achieved prior to CPN (pre-CPN) was weighed against the one acquired by the fuller dataset which allowed for the application of the multiple-year approach (post-CPN). Accuracy figures for the areas mapped as 'changed' and 'not-changed' for the period between 1987 and 2000 were higher when the multi-temporal processing was applied. The percentages were high for the unchanged areas, pre-CPN and post-CPN alike.
In the case of the Athens, post-Olympics urban expansion mapping exercise, two land use and cover classification methods were applied in the area of Messogia in the east periphery of the city, namely a pixel-based methodology based on a supervised Maximum likelihood (ML) classification and an object-based segmentation approach. The obtained results showed best performance for the object-based classification. It was also concluded that both classification schemes were improved by the rule-based multi-temporal approach.
Finally, the role of spatial metrics in the performance of an ANN model for predicting land use and cover changes was investigated. The model outputs were validated against a land-use change map, which was derived from ortho-rectified aerial photographs, Landsat TM and Quckbird data. The results from the application of the model before and after the use of spatial metrics indicated that the model could predict the patterns of change in the island's olive-groves reasonably well when parameterised with spatial metrics.