|کد مقاله||کد نشریه||سال انتشار||مقاله انگلیسی||ترجمه فارسی||نسخه تمام متن|
|4464586||1621807||2016||11 صفحه PDF||سفارش دهید||دانلود کنید|
• Both, CORINE Land Cover and NDVI turned to be useful tools, in combination with other variables, in landscape classification.
• Landscape classifications based on CORINE were sensitive, in terms of typology and number of classes, to changes in spatial resolution.
• Landscape classifications based on NDVI showed little sensitivity to changes across spatial resolutions.
• Landscape classifications based on NDVI provided landscape patterns geographically structured.
• The maximum overall accuracy of landscape classifications was never obtained at the original resolution of land cover data.
The European Landscape Convention encourages the inventory and characterization of landscapes for environmental management and planning actions. Among the range of data sources available for landscape classification, remote sensing has substantial applicability, although difficulties might arise when available data are not at the spatial resolution of operational interest. We evaluated the applicability of two remote sensing products informing on land cover (the categorical CORINE map at 30 m resolution and the continuous NDVI spectral index at 1 km resolution) in landscape classification across a range of spatial resolutions (30 m, 90 m, 180 m, 1 km), using the Cantabrian Mountains (NW Spain) as study case. Separate landscape classifications (using topography, urban influence and land cover as inputs) were accomplished, one per each land cover dataset and spatial resolution. Classification accuracy was estimated through confusion matrixes and uncertainty in terms of both membership probability and confusion indices. Regarding landscape classifications based on CORINE, both typology and number of landscape classes varied across spatial resolutions. Classification accuracy increased from 30 m (the original resolution of CORINE) to 90m, decreasing towards coarser resolutions. Uncertainty followed the opposite pattern. In the case of landscape classifications based on NDVI, the identified landscape patterns were geographically structured and showed little sensitivity to changes across spatial resolutions. Only the change from 1 km (the original resolution of NDVI) to 180 m improved classification accuracy. The value of confusion indices increased with resolution. We highlight the need for greater effort in selecting data sources at the suitable spatial resolution, matching regional peculiarities and minimizing error and uncertainty.
Journal: International Journal of Applied Earth Observation and Geoinformation - Volume 50, August 2016, Pages 95–105