INVENTORY ASSESSMENT OF GRASSLANDS IN IRELAND USING HYPERTEMPORAL OPTICAL DATA

摘要

Grassland is the dominant land-cover in Ireland,rnbut so far no image based operational land-coverrnclassification system is available for therndiscrimination of the variety of grassland types.rnThe resulting knowledge gap on grassland coverandrn-management figures in Ireland needs to bernaddressed if accurate terrestrial carbon estimatesrnare to be derived. For this purpose we present arnmethodology for the classification of general landcoverrnand different grassland management regimes.rnMODIS composite time-series data for the yearsrn2002 to 2010, with an emphasis on 2008, werernclassified with different machine learning methods,rnnamely Support Vector Machine, Random Forestrnand Extremely Randomized Trees in two studyrnareas in Ireland and compared with the output fromrnthe Maximum Likelihood classifier. Inrnhomogenous areas (pure pixels), the 5-fold-crossvalidatedrnclassification accuracies exceeded 96 %rnfor all machine learning classifiers, with SVMrnreaching 99.1 % in 2010 (κ=0.989). However, duernto the low spatial resolution of the sensor (250 m)rnand the high fragmentation of the Irish landscape,rnmixed pixels are a common occurrence, but can tornsome extent be analysed using a class probabilityrnapproach. However, for detailed land cover andrnuse, it is necessary to incorporate additional higherrnspatial resolution datasets by data-fusion.