Çok-zamanlı çok-bantlı uydu görüntüleri kullanarak Akdeniz çalı sınıflandırması

Shrublands, which have a crucial role in retaining the ecological balance, constitute an important part of the Mediterranean ecosystems. However, their composition, distribution and dynamics are not well understood. It is necessary to know the distribution of the alliances at regional scale in order to construct models that explain their dynamics. Such models will help researchers to evaluate their role in ecosystems and predict their responses to climate change. Necessary alliance distribution maps can only be produced by employing remote sensing techniques. This study presents a methodology that generates alliance-level woodland/shrubland maps of the Mediterranean region in southern Turkey from satellite images using various machine learning techniques with different parameter combinations. Multi-temporal images are used to extract information from vegetation phenology. Topographic and meteorological data are also used for improving classification. Cross-validation is performed using a ground-truth data set of 7452 polygons. Results show that detailed and accurate maquis shrubland classification is possible using a combination of environmental features and multi-spectral and multi-temporal satellite images. Addition of the environmental features to remotely sensed ones improved classification accuracy by 16%. The Random Forest (RF) algorithm is found to improve classification accuracy by 35.9% and 13.9% relative to Support Vector Machine and Quadratic Discriminant Analysis algorithms, respectively. Alliance-level classification maps of maquis acquired from RF classification are produced with 64.0-82.1% overall accuracy. Large-scale shrub classification method will have important implications on natural resource management and other ecological applications.