Land cover classification has traditionally been based on imagery
from one or two points in time. The pan-European CORINE land cover
classification system, for example, is largely derived from optical datasets
acquired at most twice in one year. These temporal restrictions fail to fully
capture and indicate fluctuations in vegetation over the season, which can
potentially result in misclassification of land cover surfaces. A lack of
information on seasonal variability within a land cover class, for example
different pasture management practices, can further limit the use of the data.
This research explored an alternative approach to deriving
information about Irish landcover, using the seasonal cycle of vegetation
growth retrieved from a time-series of medium resolution satellite imagery. The
study used the Enhanced Vegetation Index (EVI) product for 2006 from the
MODIS-TERRA satellite sensor. A time-series of 250m spatial resolution, 16-day
composite EVI images was processed, using time-series analysis methods. The
seasonal profile per pixel was first modelled using TIMESAT to reduce the
effect of cloud interference, and modelled pixels were then clustered using a
divergence-guided ISODATA clustering algorithm. This grouped pixels of similar
seasonality together using a data-driven approach. A Jeffries-Mathusita threshold
analysis was then applied to evaluate cluster distinctiveness.
Clear patterns in vegetation seasonality were detected across the
island, with the Jeffries-Mathusita analysis identifying ten groups
characterised by different seasonal profiles. Analysis of the seasonality
pattern attributed to individual clusters within these groups revealed that
certain areas exhibited two seasonal peaks in photosynthetic activity, with a
marked decrease in activity in June/July. Subsequent comparison with the CORINE
landcover 2006 dataset for the Republic of Ireland confirmed their location
within “pasture” areas. Land use such as silage cutting, practiced in
intensively-managed grasslands, is suggested as the most likely cause of the detected
double-season cycle. This finding demonstrates the potential of time-series
analysis for improving our knowledge of seasonal attributes of landcover in
Ireland, and of grassland management in particular. Furthermore, the finding
has important implications for carbon accounting under Ireland’s commitments to
the Kyoto protocol.