With global food demand on course to double in the next 50 years the pressures of agricultural intensification on
ecosystem services in highly managed landscapes are increasing. Within an agricultural landscape non-cropped areas are
a key component of ecological heterogeneity and the sustainability of ecosystem services. Management of the landscape
for both production of food and ecosystem services requires configuring the non-cropped areas in an optimal way, which,
in turn requires large scale information on the distribution of non-cropped areas. In this study the Canny edge detection
algorithm was used to delineate 93% of all boundaries within 422 ha of agricultural land in south east England. The
resulting image was used in conjunction with vegetation indices derived from Color Infra Red (CIR) aerial photography
and auxiliary landuse data in an Object Orientated (OO) Knowledge Based Classifier (KBC) to identify non-cropped
areas. An overall accuracy of 94.27% (Kappa 0.91) for the KBC compared favorably with 63.04% (Kappa 0.55) for a
pixel based hybrid classifier of the same area.
In this study satellite data from five different multispectral sensors were used in a change detection study of vegetation
disturbance on an Irish active raised bog. Radiometric normalisation was performed using Temporally Invariant Clusters
(TIC) and cross calibration applied using linear regression of radiometrically stable ground-based targets. Erdas
Imagine's Spatial Modeller was used to create a change detection model using pixel-to-pixel based subtraction with a
Standard Deviation (SD) threshold. The effectiveness of the cross calibration process was shown with the aid of
Kolmogorov Smirnov sample tests which showed a reduced D value between master and slave cumulative distribution
curves after cross calibration. The spatial accuracy of various SD threshold levels was assessed, with 1.5 SD producing
0.19% error when compared to actual ground truth boundary data of change. An error matrix of change/ no change
verified 1.5 SD as the optimum threshold for change detection, with user, producer, overall and kappa values all above
95%. Vegetation disturbance in the study was predominantly attributed to turf cutting on the boundaries of the bog.
However in May 2008 a large burn event occurred on the northeastern side of the bog which removed all surface
vegetation, equating to an area of 36ha (or 7.85% of total area).