Browsing by Author "Sannier, C."
Now showing 1 - 2 of 2
Results Per Page
Sort Options
Item Open Access Comparison of AVHRR, MODIS and VEGETATION for land cover mapping and drought monitoring at 1 km spatial resolution(Cranfield University, 2007-05) Toukiloglou, Pericles; Sannier, C.Low spatial resolution remote sensors are one of the best data sources for large area land cover mapping and drought monitoring. This study was concerned with identifying which of the three most operational such sensors (AVHRR, MODIS, and VEGETATION), were likely to help produce the best results within the mentioned applications. A rigorous review of the sensors’ characteristics led to the hypothesis that in land cover mapping and drought monitoring applications MODIS is most likely to achieve the best results followed by VEGETATION and lastly by AVHRR. This hypothesis was tested against experimental results generated within this study. A methodology was developed allowing for unbiased relative comparison of the capacity of the sensors’ Solar Reflective Bands (SRBs) to map land cover, and was applied to data collected over the UK and Greece, for which maps were produced using data collected by each sensor over the same dates and sites, and accuracy estimated using reference data. In the majority of cases the most accurate maps were produced by MODIS data; however, there were cases when maps produced by AVHRR and particularly VEGETATION data were more accurate. Drought monitoring methodologies for low resolution data require historical Normalised Difference Vegetation Index (NDVI) records extending longer than MODIS and VEGETATION operational times. Towards solving this limitation, the relationships between the sensors’ NDVI measurements over the same targets were investigated. It was found that NDVI data for one sensor could be predicted from NDVI data collected by another sensor with considerable accuracy. Consequently, MODIS and VEGETATION historical NDVI records could be extended based on past AVHRR data, and applications could be benefited by interchanging sensors for provision of NDVI data in the event of a sensor failure. These extended datasets were used to assess drought conditions over Ethiopia with the aim of using the Vegetation Productivity Indicator (VPI) methodology. The sensors’ NDVI data responsiveness to rainfall was assessed, finding MODIS NDVI data to best reflect rainfall conditions, and likely to produce more accurate VPI results. Overall the experimental results generated in this study supported the initial hypothesis.Item Open Access A Critical Evaluation of Remote Sensing Based Land Cover Mapping Methodologies(Cranfield University, 2008-06-05) Farmer, Elizabeth A.; Sannier, C.; Brewer, Timothy R.A novel, disaggregated approach to land cover survey is developed on the basis of land cover attributes; the parameters typically used to delineate land cover classes. The recording of land cover attributes, via objective measurement techniques, is advocated as it eliminates the requirement for surveyors to delineate and classify land cover; a process proven to be subjective and error prone. Within the North York Moors National Park, a field methodology is developed to characterise five attributes: species composition, cover, height, structure and density. The utility of land cover attributes to act as land cover ‘building blocks’ is demonstrated via classification of the field data to the Monitoring Landscape Change in the National Parks (MLCNP), National Land Use Database (NLUD) and Phase 1 Habitat Mapping (P1) schemes. Integration of the classified field data and a SPOT5 satellite image is demonstrated within per-pixel and object-orientated classification environments. Per-pixel classification produced overall accuracies of 81%, 80% and 76% at the field samples for the MLCNP, NLUD and P1 schemes, respectively. However, independent validation produced significantly lower accuracies. These decreases are demonstrated to be a function of sample fraction. Object-orientated classification, exemplified for the MLCNP schema at 3 segmentation scales, achieved accuracies approaching 75%. The aggregation of attributes to classes underutilises the potential of the remotely sensed data to describe landscape variability. Consequently, classification and geostatistical techniques capable of land cover attribute parameterisation, across the study area, are reviewed and exemplified for a sub-pixel classification. Land cover attributes provide a flexible source of field data which has been proven to support multiple land cover classification schemes and classification scales (sub-pixel, pixel and object). This multi-scaled/schemed approach enables the differential treatment of regions, within the remote sensing image, as a function of landscape characteristics and the users’ requirements providing a flexible mapping solution.