Generation of current land cover/use maps using direct and indirect approaches

Classification of land cover and land use has great importance for land planning and land management applications. As remote sensing technologies improve day by day, a wide variety of land cover and land use classification systems have been developed to answer different needs in many application areas. In this study, Coordination of Information on Environment (CORINE) land cover/use classification system which is a European programme established to make an inventory on land cover of the 38 member states, has been evaluated for it's production capability of land cover/use data by using two different approaches. Firstly, land cover/use map is produced by updating former Corine Land Cover Map with change layer using standard CORINE techincal approach and labeled as indirect approach. Then, land cover/use map is created from scratch without using updating methodology and labeled as direct approach. CORINE Program is the biggest European land cover/use monitoring project aiming to generate periodical and standard geographical information on environment in order to support environmental policies in various areas such as agriculture, climate change monitoring, urban planning etc. CORINE land cover/use classification project is mainly based on satellite image interpretation practice. First CORINE Land Cover (CLC) maps were generated in 1990, after that, CLC maps were updated in accordance with several needs specified by CLC users in 2000. CLC2000 was created by using geometrically corrected satellite images acquired by LANDSAT 7 ETM. Ever since the generation of CLC2000, CLC maps are being created in every 6 years. CORINE nomenclature mainly consist of 5 land cover/use classes as artificial surfaces, agricultural areas, forests and semi-natural areas, wetlands, water bodies in 1st level. These 5 land cover/use classes are detailed as 44 land cover/use classes in 3rd level. Minimum width of linear objects and minimum mapping area are 100 m and 25 ha, respectively. Additionally, minimum mapping area of change in change mapping is 5 ha. In this study, CORINE land cover data produced by using two different approaches has been compared in terms of their accuracy. Multi-temporal remote sensing data obtained form IRS-P6 and SPOT 5 satellites of which acquisiton dates are September 2006 and July 2014 were used to create false colour composites with three bands between 0.50-0.89 μm spectral ranges. These false colour images provide high level photo-interpretation at the scale of 1:100 000. The satellite images were resampled to 20 m spatial resolution by using nearest neighbour resampling method and then they were diced to 20 km x 20 km tiles. Mugla, a south western city of Turkey, was selected for the pilot study due to diverse land cover characteristics of the area. Also, Mugla has different types of land cover/use changes between the years of 2006 and 2014. Creating land cover/use maps with CORINE methodology is based on multi-temporal satellite image comparison for generation of land cover changes and adding land cover change areas to previous CLC data. In order to create 2014 CORINE land cover data (CLC2014), CLC2006 data was revised, then CLC-Change 2014 was created making comparative analysis of 2006 and 2014 false colour images using visual interpretation method. Revision process was applied to prevent of error deployment in geometry and/or thematic content from from CLC2006 and correctly generation of land cover change data. In this study, 27 land cover/use classes at 3rd level covering nearly 38 208 ha have been determined in Mugla and changes in the study region consist of 69 polygons with total area of 7 168 ha. Agricultural area changes occured more than the other land cover types in the study area. Agricultural changes arise from one agricultural land cover type changing into another. In addition new built-up areas, reforestration and degradation areas have been identified in the study area. Directly produced CLC2014 land cover data has been generated by depending on CORINE land cover technical standards. When comparing land cover data generated by using direct and indirect approach, approximately %12 of the total area were detected to be different from each other in both maps. These differences are composed of decrease and increase for each land cover class. For example, while coniferous forest land cover area (312) in directly produced CLC2014 data was larger than CLC2014 generated by indirect approach , fruit tree and berry plantations land cover type (222) covers less area than CLC2014 generated by indirect approach. The results shows that these differences are mainly based on generalization rules of CORINE land cover classes and visual interpretation. Two different accuracy assessment for these current land cover/use maps has been performed by using sample areas that are randomly selected as 100 m x 100 m and 500 m x 500 m width grids. Sample grids were generated by Geospatial Modelling Environment (GME) tool. Sample width selection was performed based on two main mapping units that are used in CORINE land cover mapping. While 100 m x 100 m grids were selected based on minimum width of linear objects, 500 m x 500 m grids were selected based on minimum mapping unit. Same sample areas were used for each land cover/use maps. Ground Truth for these samples have been obtained from Google Earth high resolutiın imagery. Acccuracy assessment performed by using 100 m x 100 m width grids were evaluated for each land cover/use classes for two land cover/use maps generated by two different approaches.Confusion matrices have been created using same sample areas for each land cover data. While overall accuracy for directly produced land cover/use map is % 81,73, accuracy for land cover/use map produced by using updating former CORINE Land Cover Map with change layer is % 76,35. Accuracy assessment results of directly produced CLC2014 showed that user's accuracy for 242, 243, 323, 324 and 332 CORINE land cover classes are lower than other land cover/use classes in the study area. On the other hand, accuracy assessment results of CLC2014 produced by using updating former CORINE Land Cover Map with change layer indicated that user's accuracies for 112, 133, 242, 243, 323, 324 and 332 CORINE land cover classes are lower than other land cover/use classes in the study area. Kappa Coefficients for two CLC2014 maps have been calculated as 0,79 for directly produced CLC2014 and 0,73 for CLC2014 produced by using indirect approach. In acccuracy assessment performed by using 500 m x 500 m width grids, each land cover/use classes for each two maps were not evaluated, because polygon structures of some land cover classes are not linear. Therefore, sample grids cover more than one land cover class for these small and non-linear polygons and do not represent real land cover class. Confusion matrices have been created using same sample areas for each CLC2014 land cover data. Overall accuracy results have been obtained as %88,77 for directly generated CLC2014 and %79,30 for CLC2014 generated by using CORINE methodology. Accuracy assessment results performed by using 500 m x 500 m width sample grids indicated that user's accuracy for complex land cover classes (242, 243, 313 etc.) increased, on the other hand, user accuracy for 231, 331, 211, 122 and 131 etc. land cover classes, which contain small areas in the study area and linear widths of them are lower than 500 m, decreased or were not assessed. When Kappa Coefficients for two CLC2014 maps have been taken into account, the results indicated that values for directly produced CLC2014 and CLC2014 produced by using indirect approach 0,87 and 0,77 respectively. Consequently, when directly produced CLC2014 (direct approach) and CLC2014 produced by using updating former CORINE Land Cover Map with change layer (indirect approach) compared with regard to their duration of production, the results showed that if previous CLC data (revision CLC data) has higher land cover accuracy, generation of update CLC by using indirect approach takes less time and has higher time efficiency and higher land cover accuracy. However, if previous CLC data (revision CLC data) has lower land cover accuracy, time durations of generation of update CLC by using CORINE methodology and generation of directy produced CLC will be nearly same. In this study, production duration of CLC data were almost same.