Knowledge-based classification of multitemporal data-set for automatic land use/land cover mapping in rainfed agricultural areas of central Spain
Guerrero, I.
In spite of its main economic and environmental impact on Europe, geographic information of agricultural figures is usually presented in broad categories and more detailed geographic data on different agricultural substrates seems desirable. But one of the limitations when mapping agricultural areas is their high between-year variability caused, among others, by rotation systems. In this study, an Expert System-based image classification protocol is developed to provide yearly updated and detailed geographic information on different substrates of agricultural areas in Central Spain. The protocol is applied in two different medium resolution optical sensors (ASTER and Landsat ETM+) and two different years (2001 and 2002). To enable the repeatability of the process and the multi-source feeding of the protocol, the following strategies have been used: Multi-seasonal data-set: two dates selected after analysing the cropping pattern of the study area. Image segmentation and object-oriented classification. A pixel-based classification was also applied to compare the results. Feature space based on spectral indices. Overall accuracies between 74.55 percent and 83.28 percent were obtained. Pixel-based classifications showed lower accuracies than the corresponding object-oriented ones. Both maps obtained from Landsat ETM+ imagery (30m cell size) showed lower accuracies than those derived from ASTER imagery (15m cell size). Major agricultural substrates presented on the area (i.e. fields under two-year rotation system of rainfed cereal crops) were effectively classified (accuracies from 78.17 percent to 92.96 percent).
Показать больше [+] Меньше [-]Ключевые слова АГРОВОК
Библиографическая информация
Эту запись предоставил Mediterranean Agronomic Institute of Chania