Remote Sensing and Photogrammetry Society Annual Conference

6-7 September, The Macaulay Land Use Research Institute, Craigiebuckler, Aberdeen , UK

Presentation Abstracts

Ignore the ontological aspects of land cover information at your peril

Dr Lex Comber, Department of Geography,  ADAS, Senior Research Consultant, Environment Systems, ADAS Woodthorne, Wergs Road, Wolverhampton, WV6 8TQ, UK

Email lex.comber@adas.co.uk

Abstract:

Third party land cover information (derived from remotely sensed imagery) is increasingly being used to describe landscape structures for a range of applications. Land cover information varies due to methodological artefacts as well as spatial and thematic issues. The land cover ontology how things are identified and what the categorical concepts mean is not reported in land cover meta-data. Users of land cover, in different disciplines and with different perceptions, may be unaware of its origins and meaning. They treat land cover information as if it were data and integrate it into their information systems without fully understanding the conceptual differences / similarities underlying the different categories or classifications. Such ontological discord results in uncertainties that have a much more profound effect upon analyses than do thematic and positional accuracies. We propose that the solution is two-fold. First, land cover meta-data reporting should include descriptions that go beyond the usual class descriptions. They should describe the meaning of the data objects (pixels or parcels) in terms of the specified ontology. This includes descriptions of:

• the landscape concepts that underpin the classes identified in land cover databases;
• the scientific, political, and ontological issues that informed the production of the data;
• who decided what the features of interest were and why.

Second, expert or knowledge-based approaches allow specific (data) integration questions or models to be explored. If the expertise is was available to model data ontologies (including semantics) then learning algorithms such as C4.5 could be applied to the problem. The output would provide expert knowledge. The outcome of these types of analysis would be testing the logical model û data mining. The results would give users a fuller understanding of the nature of the land cover information: integration would allow finer grained characterisation of the landscape, and would identify inconsistencies between different conceptualisations. That is, it would show where either the data or the model was inconsistent, thus allowing the logical model and / or the rule set to be revised. The information that needs to be communicated by producers needs to be expanded from current practise of reporting the easily measurable (scale, resolutions, accuracy etc). CEH have gone some way towards this with the Land Cover Map 2000 that contains extensive object-level meta-data. However ways need to be found that enable logical consistency reporting to be extended to include differences in data meaning. This would allow the benefits of data integration and access to 3rd party information, as proposed by initiatives such as E-Science and the GRID, to be realised.

Towards operational land cover change monitoring from space

Dr Heiko Balzter, Centre for Ecology and Hydrology (CEH), Monks Wood,  Abbots Ripton, Huntingdon, UK

Email hbal@ceh.ac.uk

Abstract:

This presentation gives a review of a number of ongoing projects in the European GMES initiative (Global Monitoring for Environment and Security). Institutional user needs are discussed with a view on the adequacy of current and near-future spcaeborne observing systems. Methodological approaches using multi-sensor concepts (SAR, scatterometer, optical, thermal) at a range of resolutions are presented.

Deriving long term land cover change from aerial photography to assess pressures on biodiversity

Dr Geoff Smith, Section for Earth Observation, Centre for Ecology and Hydrology (CEH), Monks Wood, Abbots Ripton, Huntingdon, UK

Email gesm@ceh.ac.uk

Abstract:

Aerial photography is the only means of deriving long term (~50 years) spatially detailed land cover change information. Historical satellite remote sensing is still relatively coarse spatially (~30 m) when addressing subtle land cover changes and the record only goes back to the mid 1970s.

The Global Monitoring for the Environment and Security (GMES) programme aims to identify the strengths and the weaknesses of the current European capacity for monitoring and information production. This current capacity must be viewed in the light of the ability to derive useful information now and thus to guide the development of future systems. To this end, the BIOPRESS (Linking Pan-European Landcover Change to Pressures on Biodiversity) project is providing decision makers with quantitative information on how changing land cover / use has affected the environment and biodiversity in Europe over the second half of the 20th century.

The project is currently at the midway point having produced historical (1950 * 2000) land cover change information in and around a selection of Natura2000 sites. The sites were selected to represent the range of landscapes and habitats in Europe . A detailed change detection exercise has been applied to 65 transects of 2 km by 15 km along gradients from the centre of a Natura2000 site towards a source of environmental pressure. At each date the transects were interpreted manually against a classification derived from and for integration with CORINE Land Cover.

This work has produced interesting results for land cover change across Europe . It has also identified a range of issues that must be considered when attempting to map land cover change in such a way. These have included the availability and quality of data, the spectral format (panchromatic / colour), the adaptability of the classification and the variability of interpretation across Europe .

The change information will be analysed in the context of 'Driver, Pressure, State, Impact, Response' modelling framework to assess pressures on biodiversity with all results extrapolated to the European level through a biogeographical stratification.

Global land cover mapping using data from Earth Observing satellites; Current status and future perspectives

Dr Hugh Eva, European Commission, Joint Research Centre, 21020 Ispra (VA), Italy

Email hugh.eva@jrc.it

Abstract:

A brief history of global land cover mapping is presented. The Global Land Cover 2000 project is described in detail, and new action including the joint European Commission European Space Agency plans for 250 metre resolution Global Land Cover map for 2005 is introduced.

Remote sensing of growth dynamics of Sitka spruce plantation forests in upland Britain

Dr Daniel Donoghue, Department of Geography, University of Durham, UK

Email danny.donoghue@durham.ac.uk

Abstract:

Monitoring growth patterns over large forest estates using field survey methods is a time-consuming and costly exercise. This study paper evaluates remotely sensed data to help monitor changes in Sitka spruce (Picea sitchensis) plantations in Britain using satellite and airborne laser scanning imagery. The results demonstrate that low-cost satellite image data from the Landsat Enhanced Thematic Mapper (ETM+) sensor can be used to predict and map forest height and basal area characteristics with a good level of accuracy for young crops. Regression analysis of contemporaneous remote sensing data and ground based forest structural measurements is used to predict height and basal area. These models can be applied to any radiometrically normalised image of the same area to give a quantitative description of observed growth in young crops between successive images. This helps to identify and quantify areas of anomalous growth. High resolution airborne laser scanner imagery gives excellent estimation of forest characteristics but is very expensive. These data can, however, be used both to complement field measurements to improve predictions from Landsat data and to assess their quality. We conclude that retrieval of forest structure information is best achieved by the integration of satellite, airborne and ground-based measurements.

The application of digital photogrammetry and crown delineation techniques to derive and monitor tree and stand characteristics

Professor David Miller, Macaulay Land Use Research Institute, Craigiebuckler, Aberdeen AB15 8QH, UK

Email d.miller@macaulay.ac.uk

Abstract:

Ground based forest inventory surveys can provide highly accurate measurements of tree and stand characteristics, but these are expensive to carry out. Aerial photography has been used for several decades as a tool in forest management and inventory. However, conventional methods of interpretation are both time-consuming and costly, with results varying between interpreters. With continuing development of personal computer technology, aerial photographs have become more accessible for digital analysis. This presentation shows the potential operational use of digitised aerial photographs, for the estimation of tree and stand characteristics, and the change in characteristics through time. The digitised aerial photographs were analysed using softcopy photogrammetry and techniques for individual tree crown delineation. For fully grown forest stands, the estimations of stand top height, basal area, stand volume, stand biomass and stand density (-23.7%) were similar with the ground measured stand characteristics ( ± 10%). Applying the approach to multiple dates of imagery provides a means of monitoring forest development and this is discussed as a part of the process of regular monitoring or inventory. The implications of illumination conditions and topography are also discussed.

Mapping land use in NE Scotland with neural networks from remote sensing imagery

Dr Matthew Aitkenhead, Macaulay Land Use Research Institute, Craigiebuckler, Aberdeen AB15 8QH, UK

Email m.aitkenhead@macaulay.ac.uk

Abstract:

Landsat TM imagery can be used to classify different land cover types based on reflectance characteristics in seven wavelength bands. Various methods, including NDVI and other simple mathematical transformations, can be used to show strong variations in band reflecance ratios from different surfaces. However, a neural network trained with the backpropagation method should be able to improve on these simple mathematical calculations by developing complex functions which allow recognition of different land cover or land use types. Landsat imagery of Aberdeen and the surrounding area is used to develop a land use map highlighting areas of residential, commercial and industrial land use, along with various natural and semi-natural land cover classes. Appropriate selection of training sites and categorisation of land cover classes are two aspects highlighted as important to the successful development of a neural network land use mapping system.

Texture analysis of land cover change detection

Miss Fen Yang, Department of Computing Science, University of Aberdeen, Aberdeen, UK

Email fyang@csd.abdn.ac.uk

Abstract:

We wish to detect land cover change for environmental management. The abilities of Laws Masks and Gabor filters to distinguish land cover types are evaluated since they are common texture measures. This paper investigates a texture based image description in which the standardised MPEG-7 Homogeneous Texture Descriptors (HTD) of Gabor filters are used as the textural feature vector. A discriminant classifier is designed to use linear regression analysis to distinguish vectors derived from the filter images. Experiment results show that both Laws Masks and Gabor filters are capable of expressing texture information of land covers, but they have different results on different land cover types.

An evaluation of optimum coherent Digital Elevation Models from interferometric SAR data

Dr Parivash Lumsdon, Macaulay Land Use Research Institute, Craigiebuckler, Aberdeen AB15 8QH, UK

Abstract:

In this paper we propose a new method for the speckle filtering and derivation of an optimum interferogram from fully Polarimetric Interferometric SAR data. This method is based on the phase maps derived from high resolution ESPRIT algorithm. The cross composite covariance matrix elements are filtered using a supervised edge aligned lee filter. In this way the cross polarisation interference is minimised, preserving the information held in Polarimetric phase elements. The results of application of this method to the Glenaffric radar data is illustrated in this paper.

Assessing temporal variability in habitat for vicuñas in Chile

Mr Jerry Laker, Macaulay Land Use Research Institute, Craigiebuckler, Aberdeen AB15 8QH, UK

Abstract:

From a low point of less than 500 individuals in the 1960s, vicuñas in the Chilean altiplano are known to have reached a stable equilibrium of around 16,000 animals (I Region). Behavioural observations suggest the species is territorial and that family groups of males with breeding females are sedentary. However, a 30-year regular census indicates considerable spatial heterogeneity in the observed changes in vicuña population within the study area (5752 km²), consistent with extensive movements of family groups, avoiding locally adverse habitats. To test this theory, data from the SPOT-VEGETATION 1km² programme was used to evaluate change in the amount of green matter during a six-year period from April 1998-April 2004.

Woodland resource assessment in Bangladesh: A case study of deciduous Sal Forests

Mr Sheikh TawhidulIslam, Department of Geography, University of Durham, UK

Email s.t.islam@durham.ac.uk

Abstract:

The models resulted from different forest variables and spectral response patterns of high resolution satellite (Quickbird) image have shown strong association. Band 2 in the visible region and band 4 (NIR) are strongly correlated with forest variables. For instance, high values correlation of coefficients of band 2 with dbh and height (r = -0.65 and r = -0.75 respectively) suggested that Quickbird image could be very useful in this kinds of research. Band 1 and 3 has also revealed significant associations with forest parameters. However, it is evidenced that Quickbird image can most effectively be used for forest variable assessment and predictions at application level even in the third world countries where forest management are hardly practised and forests are heterogenic in nature what, in turn, limits the use of mid resolution TM or ASTER sensor. This attempt provided evidence that analysis of high resolution Quickbird image data coupled with proposed field technique can be used for improving forest statistics in Bangladesh.

Land cover change and its impact on savanna heterogeneity in Kruger National Park, South Africa

Mr Tony Rajan Mathew, School of Geography, The University of Nottingham, UK

Email lgxtrm@nottingham.ac.uk

Abstract:

Vegetation composition, distribution and dynamics underlie the ecosystem patterns we observe ( Packer et al. 1999). Traditional techniques of field survey meant that ecological investigations were mostly confined to the local scale, requiring the ecologist to extrapolate the results to understand the regional dimensions of the process in question. This essentially meant that the knowledge-base developed for ecological patterns and processes was limited, in both its spatial and temporal extent and the data derived remained ‘patchy’ for conservation and management purposes.

Structure and pattern are two of the most common physical characteristics that one sees investigated in studies linking vegetation with ecological processes. The ability to detect changes in pattern and make readings at more than one level of resolution is being increasingly recognised as fundamental for ecosystem science. This has assumed more importance given the realisation that management activities at one scale often have unexpected or undesirable effects at other scales. Achieving this goal requires identification of the physical and biological processes of interest, estimation of the variables and parameters that affect those processes at different scales and development of rules to translate information across scales (Gardner 1998). Through its objective view and synoptic, repeatable coverage, satellite remote sensing has made it possible to map large areas with varying spatial and temporal resolutions relatively easily. However remotely sensed data are generally collected at a single spatial resolution, in contrast to the many scales at which nature operates, making it necessary to have a multi-scalar analytical approach when such data are used for ecological applications.

The savanna constitutes the largest biome in southern Africa occupying 46% of its area and over one-third of the area of South Africa (GRID 2004). It is the primary habitat type in two large national parks: Kruger and Kalahari Gemsbok. With an area of 1,898,458 ha Kruger occupies almost 2.5 percent of the total land surface area in South Africa . Heterogeneity at multiple spatial and temporal scales is a key attribute of savanna ecosystems, as is the relative dominance of woody and herbaceous species. Variation in the proportion of these is driven by the changes in the disturbance regime (fire, herbivory, floods) as well as climate. Disturbance is considered to be a major factor influencing landscape pattern and vegetation composition (Forman and Godron 1986), especially so in savanna, which has fire as one of the major drivers of ecosystem change. We hypothesise that the frequent and extensive fires in the Kruger savanna landscape lead to a shortening of the temporal scale of patchiness, leading to homogenisation of the native vegetation. In terms of scaling, as spatial and temporal resolutions become coarser, the patchiness fuses into homogeneity.

In this study we examine the change in the spatial pattern of vegetation in Kruger National Park over a 30-year time period (1972 – 2002) through the analysis of time-series images from Landsat sensors – Multi-Spectral Scanner (MSS), Thematic Mapper (TM), and Enhanced Thematic Mapper Plus (ETM+). We propose to investigate the relationship between land cover change and savanna heterogeneity at multiple spatial and temporal scales. Different vegetation types manifest particular spectral reflectance and emittance properties, which result in distinctive spectral response patterns ( Packer et al. 1999). The differential absorption of solar energy by green vegetation at the red and near-infrared region of the solar spectrum – the basis for vegetation indices – has long been one of the techniques used for vegetation mapping (Justice et al. 1985; Townshend and Justice 1986). Among the indices proposed Normalised Difference Vegetation Index (NDVI) has seen the most use, especially in global-scale change-detection studies. Given the structural openness that is typical of savanna (compared with the closed-canopy structure of tropical forests), we propose to use Soil Adjusted Vegetation Index (SAVI), which effectively normalises soil substrate variations (Huete 1988) so as to not influence the vegetation measure.

Poster Abstracts

Field survey methodology for a versatile land cover mapping methodology combining remote sensed data sources.

Miss Elizabeth Farmer, Cranfield University at Silsoe, Barton Road, Silsoe, BEDS, MK45 4DT, UK

Email e.a.farmer@cranfield.ac.uk

Abstract:

This study proposes a land cover mapping approach which places less emphasis on the delineation of the landscape into pre-defined classes. Instead, the parameters typically used in boundary delineation, landscape monitoring and of interest to stakeholders will be determined continuously across the study area. Such an approach aims to provide the user with a widely applicable, flexible and detailed vegetation database.

Initially six parameters, such as species and percentage cover, will be considered across a 210km² study area located in the North York Moors National Park . This study area was chosen as it encompasses a diversity of land cover types and moorland management regimes.

Although it is planned to develop a methodology for biophysical parameter interpolation at the next stage based on the integration of field and remotely sensed data, the present work focuses on effective field survey implementation. The field survey, although based on standard ecological measurement techniques differs from that of many land cover mapping schemes as it requires a dis-aggregation of measurements from the polygon to the point. This transitional to point measurements has led to extensive research into the most appropriate and effective field survey design. Primarily based on a pilot study work has been conducted to investigate sample arrangement, sample clustering, measurement techniques, automated data collection and survey logistics culminating in a methodology which is both technically and logistically viable while providing sufficient data to facilitate parameter interpolation.

Mapping and modelling land cover change in Priorat, Catalonia

Miss Roser Cots Folch, University of Lleida, Department of Environment and Soil Science, Rovira Roure 191, 25198 Lleida, Spain

Abstract:

The region of Priorat in Catalonia is in one of the oldest wine-producing areas in Spain , having produced vintages since the 12th century. Recent successes including L'Ermita, which sells for over €150 per bottle, have highlighted the continuing success of the region's vineyards. However, since the beginning of the 20th century until 1992, most of the steeply-terraced vineyards in the region were abandoned. It is only in the last decade that there has been an increase in production and development of new terraced vineyards. This has been partly due to a recognition of the quality of the wine produced, and also through the introduction of subsidies for local farmers and the interest of large companies. Priorat therefore is a prime candidate for the study of land cover and land use change, containing as it does a history of drivers both natural and human. We have used remote sensing imagery taken from 1956 onwards, in concert with neural network methods, to carry out two tasks: (1) to investigate the possibility of mapping land cover from remote sensing imagery, in particular aerial photography; (2) to develop a model of land cover change using natural and human drivers.

Application of Landsat TM images to map long term cropping patterns

Dr José A Martínez-Casasnovas, University of Lleida, Department of Environment and Soil Science, Rovira Roure 191, 25198 Lleida, Spain

Email j.martinez@macs.udl.es

Abstract:

The present work proposes a method that allows mapping long term cropping patterns using time-series of crop maps derived from supervised classification of remote sensing data. These maps allow implicit spatial and temporal relationships among the crops growth in an agricultural area. Cropping pattern is understood as the spatial distribution of associations between crops or crops and uncultivated land in the same fields (although not in a definite succession) along the years of the analysed time-series.