Christopher D. Lloyd

The integration of spectral and textural information using neural networks for land cover mapping in the Mediterranean

The aim of this study was to develop an efficient and accurate procedure for classifying Mediterranean land cover with remotely sensed data. Combinations of artificial neural networks (ANN) and texture analysis on a per-field basis were used to classify a Landsat Thematic Mapper image of the Cukurova Deltas, Turkey, into eight land cover classes. This study integrated spectral information with measures of texture, in the form of the variance and the variogram. The accuracy of the ANN was greater than that of maximum likelihood (ML) when using spectral data alone and when using spectral and textural data. The use of texture measures through the per-pixel and per-field majority rule approaches were found to reduce classification accuracy because the field boundaries were enlarged and so overwhelmed the measures of texture. In contrast, the per-field approach (where the field was specified prior to analysis) combined with texture information increased significantly classification accuracy. However, the accuracy decreased as the variogram lag increased. The accuracy with which land cover could be classified in this region was maximised at 89% by using a per-field, ANN approach in which semivariance at a lag of 1 pixel was incorporated as textural information. This is 15% greater than the accuracy achieved using a standard per-pixel ML classification. The primary limitation of the use of the per-field approach was noted to be the need for prior knowledge of field boundaries which may be resolved using existing data or through some form of edge-detection routine.

Deriving DSMs from LiDAR data with kriging

Abstract Light Detection And Ranging (LiDAR) is becoming a widely used source of digital elevation data. LiDAR data are obtained on a point support and it is necessary to interpolate to a regular grid if a digital surface model (DSM) is required. When the data are numerous, and close together in space, simple linear interpolation algorithms are usually considered sufficient. In this letter, inverse distance weighting (IDW), ordinary kriging (OK) and kriging with a trend model (KT) are assessed for the construction of DSMs from LiDAR data. It is shown that the advantages of KT become more apparent as the number of data points decrease (and the sample spacing increases). It is argued that KT may be advantageous in some instances where the desire is to derive a DSM from LiDAR point data but in many cases a simpler approach, such as IDW, may suffice.

Texture classification of Mediterranean land cover

Maximum likelihood (ML) and artificial neural network (ANN) classifiers were applied to three Landsat Thematic Mapper (TM) image sub-scenes (termed urban, agricultural and semi-natural) of Cukurova, Turkey. Inputs to the classifications comprised (i) spectral data and (ii) spectral data in combination with texture measures derived on a per-pixel basis. The texture measures used were: the standard deviation and variance and statistics derived from the co-occurrence matrix and the variogram. The addition of texture measures increased classification accuracy for the urban sub-scene but decreased classification accuracy for agricultural and semi-natural sub-scenes. Classification accuracy was dependent on the nature of the spatial variation in the image sub-scene and, in particular, the relation between the frequency of spatial variation and the spatial resolution of the imagery. For Mediterranean land, texture classification applied to Landsat TM imagery may be appropriate for the classification of urban areas only.

A comparison of texture measures for the per-field classification of Mediterranean land cover

Land cover of a Mediterranean region was classified within an artificial neural network (ANN) on a per-field basis using Landsat Thematic Mapper (TM) imagery. In addition to spectral information, the classifier used geostatistical structure functions and texture measures extracted from the co-occurrence matrix. Geostatistical measures of texture resulted in a more accurate classification of Mediterranean land cover than statistics derived from the co-occurrence matrix. The primary advantage of geostatistical measures was their robustness over a wide range of land cover types, field sizes and forms of class mixing. Spectral information and the variogram (geostatistical texture measure) resulted in the highest overall classification accuracies.

Assessing uncertainty in estimates with ordinary and indicator kriging

The objective of this paper is to examine the applicability of three geostatistical approaches, ordinary kriging (OK), kriging with a trend model (KT), and indicator kriging (IK), to the assessment of uncertainty in estimates. This paper uses the OK and KT standard error and the conditional standard error of the conditional cumulative distribution function (ccdf) derived through IK to assess uncertainty in estimates of elevation. The mean OK and KT standard error and mean IK standard error, using data sampled from a remotely sensed digital terrain model (DTM), were used to ascertain the uncertainty in estimates. The estimates of elevation were assessed with reference to the complete DTM. Judgement on the success of the three approaches was made on the basis of the difference between the standard error of estimates and the mean kriging standard error. The mean OK and KT standard errors represent the standard error of estimation more accurately than the mean IK standard error, and OK (or KT) estimates of elevation values were more accurate than those for IK. Furthermore, IK may be significantly more costly to implement than OK (or KT) in terms of expenditure of time and effort. Also, the implementation of IK was demonstrated to be problematic in the presence of a low-frequency trend. A modified form of IK was also employed whereby the thresholds for estimation of the ccdfs were adapted locally in the basis of the available observations. This approach markedly reduced the problems encountered with IK employing fixed (global) thresholds. IK with locally adaptive indicator thresholds provided a more accurate guide to uncertainty on a local basis than OK or KT. It is suggested that IK recommended for the assessment of uncertainty in estimates locally where the estimation of accuracy of a specified will need to be implemented with a trend model to further improve results.

Analysing Commuting Using Local Regression Techniques: Scale, Sensitivity, and Geographical Patterning

In this paper, two forms of local regression are employed in the analysis of relations between out-commuting distance and other socioeconomic variables in Northern Ireland. The two regression approaches used are moving window regression (MWR) and geographically weighted regression (GWR). For the first approach different window sizes are applied and changes in results assessed. For the second approach, a Gaussian kernel is used and its bandwidth varied. Seven independent variables are utilised, although a single variable (deprivation) provides the main analytical focus. Differences in results obtained with use of the two approaches are discussed. The relationship between window size or bandwidth size and observed spatial patterning is discussed and elucidated. The results support previous work that indicated severe limitations in using global regressions to examine relationships between socioeconomic variables. Also, the utility of comparing results obtained from MWR and GWR is assessed and the benefits of both approaches are outlined.

Non-stationary variogram models for geostatistical sampling optimisation: An empirical investigation using elevation data

A problem with use of the geostatistical Kriging error for optimal sampling design is that the design does not adapt locally to the character of spatial variation. This is because a stationary variogram or covariance function is a parameter of the geostatistical model. The objective of this paper was to investigate the utility of non-stationary geostatistics for optimal sampling design. First, a contour data set of Wiltshire was split into 25 equal sub-regions and a local variogram was predicted for each. These variograms were fitted with models and the coefficients used in Kriging to select optimal sample spacings for each sub-region. Large differences existed between the designs for the whole region (based on the global variogram) and for the sub-regions (based on the local variograms). Second, a segmentation approach was used to divide a digital terrain model into separate segments. Segment-based variograms were predicted and fitted with models. Optimal sample spacings were then determined for the whole region and for the sub-regions. It was demonstrated that the global design was inadequate, grossly over-sampling some segments while under-sampling others.

Deriving ground surface digital elevation models from LiDAR data with geostatistics

This paper focuses on two common problems encountered when using Light Detection And Ranging (LiDAR) data to derive digital elevation models (DEMs). Firstly, LiDAR measurements are obtained in an irregular configuration and on a point, rather than a pixel, basis. There is usually a need to interpolate from these point data to a regular grid so it is necessary to identify the approaches that make best use of the sample data to derive the most accurate DEM possible. Secondly, raw LiDAR data contain information on above‐surface features such as vegetation and buildings. It is often the desire to (digitally) remove these features and predict the surface elevations beneath them, thereby obtaining a DEM that does not contain any above‐surface features. This paper explores the use of geostatistical approaches for prediction in this situation. The approaches used are inverse distance weighting (IDW), ordinary kriging (OK) and kriging with a trend model (KT). It is concluded that, for the case studies presented, OK offers greater accuracy of prediction than IDW while KT demonstrates benefits over OK. The absolute differences are not large, but to make the most of the high quality LiDAR data KT seems the most appropriate technique in this case.

Are Northern Ireland's Communities Dividing? Evidence from Geographically Consistent Census of Population Data, 1971–2001

There is an extensive literature on various aspects of segregation in Northern Ireland (NI). However, there are no census-based analyses of population change and residential segregation that cover the entire 1971–2001 period using consistent geographical units through time for all NI. This shortcoming is addressed in this paper by an analysis of changes in (i) the spatial distribution of population and (ii) residential segregation between 1971 and 2001 using the NI Grid-Square Product comprising data for a set of 1km2 cells that cover all populated areas in NI. The substantive issue of whether NI has become more segregated through time is addressed as are questions about measuring change through time using the census and the importance of spatial scale. One important conclusion is that NI indeed became more residential^ segregated between 1971 and 2001, but that residential segregation in 2001 remained approximately at its 1991 level according to most indicators.

Exploring population spatial concentrations in Northern Ireland by community background and other characteristics: an application of geographically weighted spatial statistics

Information on how populations are spatially concentrated by different characteristics is a key means of guiding government policies in a variety of contexts, in addition to being of substantial academic interest. In particular, to reduce inequalities between groups, it is necessary to understand the characteristics of these groups in terms of their composition and their geographical structure. This article explores the degree to which the population of Northern Ireland is spatially concentrated by a range of characteristics. There is a long history of interest in residential segregation by religion in Northern Ireland; this article assesses population concentration not only by community background (‘religion or religion brought up in’) but also by housing tenure, employment and other socioeconomic and demographic characteristics. The spatial structure of geographical variables can be captured by a range of spatial statistics including Morans I. Such approaches utilise information on connections between observations or the distances between them. While such approaches are conceptually an improvement on standard aspatial statistics, a logical further step is to compute statistics on a local basis on the grounds that most real-world properties are not spatially homogenous and, therefore, global measures may mask much variation. In population geography, which provides the substantive focus for this article, there are still relatively few studies that assess in depth the application of geographically weighted statistics for exploring population characteristics individually and for exploring relations between variables. This article demonstrates the value of such approaches by using a variety of geographically weighted statistical measures to explore outputs from the 2001 Census of Population of Northern Ireland. A key objective is to assess the degree to which the population is spatially divided, as judged by the selected variables. In other words, do people cluster more strongly with others who share their community background or others who have a similar socioeconomic status in some respect? The analysis demonstrates how geographically weighted statistics can be used to explore the degree to which single socioeconomic and demographic variables and relations between such variables differ at different spatial scales and at different geographical locations. For example, the results show that there are regions comprising neighbouring areas with large proportions of people from the same community background, but with variable unemployment levels, while in other areas the first case holds true but unemployment levels are consistently low. The analysis supports the contention that geographical variations in population characteristics are the norm, and these cannot be captured without using local methods. An additional methodological contribution relates to the treatment of counts expressed as percentages.