Zahra Asadolahi; Mostafa Keshtkar; Zia Badehian
Abstract
Introduction: Due to the importance of ecosystem services mapping in decision making, different approaches have been developed for mapping ecosystem services. InVEST software is based on models that combine land use/cover with ecosystem services, and so land use is considered an essential component of ...
Read More
Introduction: Due to the importance of ecosystem services mapping in decision making, different approaches have been developed for mapping ecosystem services. InVEST software is based on models that combine land use/cover with ecosystem services, and so land use is considered an essential component of all models. Currently, remote sensing technology is one of the top techniques in land use/cover map extraction using both pixel-based and object-oriented methods. While the pixel-based method is based on the classification of numerical values of images, object-oriented image processing is more accurate in the classification process, due to the use of spectral information, texture, and content information, being widely used in all sectors including environmental sciences. In this regard, the present study aimed to apply satellite images and object-oriented processing in land use/cover mapping and habitat services modeling. Material and methods: The research was carried out using eCognition 9.01 and InVEST3.0 software in four steps including preparing information, object-oriented processing of satellite images, object-oriented classification, and finally, habitat modeling. Images of Landsat and Sentinel satellites were fused to the eCognition software and processed in conjunction with ASTER digital elevation model data. Segmentation was performed as the first step of object-oriented classification using multi-resolution segmentation algorithm. Due to the size of the study area and the average spatial resolution of Landsat images, the images were segmented with 30 scales, 0.4 coefficient, and 0.5 compression. Geometry, vegetation (NDVI), Pixel Gray Surface Composition (GLCM), and lighting degrees were classified using the Assign class classification algorithm. Then, by matching the extracted map with 130 teaching points, the accuracy of the kappa coefficient was determined. Next, land use/cover map was introduced into InVEST software. Results and discussion: The results and statistics obtained from object-oriented classification accuracy presented acceptable results with a kappa coefficient of 0.93. In the study area, land use/cover classes were prepared in six categories including irrigated and rain-fed agriculture, forest, rangeland, man-made areas, and water resources. According to the results, rangeland and forest types with an area of 39.8% and 33.0% covered more than 72.8% of Lorestan Province’s area. Major problems with land use mapping were the inability of 30-m pixel Landsat satellite imagery to distinguish between rain-fed agriculture, rangeland, and low-density forest types due to their spectral similarity as well as rural areas due to their small surface area. Hence, we attempted to overcome this limitation by modifying the segment characteristics such as shape, tone, texture, and other information. Habitat suitability was considered for each land use class. The susceptibility of each habitat type to the threats in the study area was also weighted. Human threats affecting habitat quality were classified into three groups of agricultural lands, residential areas, and roads. Finally, the model was implemented and a habitat quality index was obtained with values ranging from zero to one. Conclusion: There are different approaches to ecosystem services mapping, one of which is extracting ecosystem services information directly from land use/cover maps. Such an approach is appropriate for large-scale areas that are restricted in terms of available data and expert knowledge, and the service is directly related to land use.
