Publications (157)360.24 Total impact
 [Show abstract] [Hide abstract]
ABSTRACT: Spatial predictions of soil properties are needed for various purposes. However, the costs associated with soil sampling and laboratory analysis are substantial. One way to improve efficiencies is to combine measurement of soil properties with collection of cheapertomeasure ancillary data. There are two possible approaches. The first is the formation of classes from ancillary data. A second is the use of a simple predictive linear model of the target soil property on the ancillary variables. Here, results are presented and compared where proximally sensed gammaray (γray) spectrometry and electromagnetic induction (EMI) data are used to predict the variation in topsoil properties (e.g. clay content and pH). In the first instance, the proximal data is numerically clustered using a fuzzy kmeans (FKM) clustering algorithm, to identify contiguous classes. The resultant digital soil maps (i.e. k = 2–10 classes) are consistent with a soil series map generated using traditional soil profile description, classification and mapping methods at a highly variable site near the township of Shelford, Nottinghamshire UK. In terms of prediction, the calculated expected value of mean squared prediction error (i.e. σ2p,C) indicated that values of k = 7 and 8 were ideal for predicting clay and pH. Secondly, a linear mixed model (LMM) is fitted in which the proximal data are fixed effects but the residuals are treated as a combination of a spatially correlated random effect and an independent and identically distributed error. In terms of prediction, the expected value of the mean squared prediction error from a regression (σ2p,R) suggested that the regression models were able to predict clay content, better than FKM clustering. The reverse was true with respect to pH, however. We conclude that both methods have merit. In the case of the clustering the approach is able to account for soil properties which have nonlinearity's with the ancillary data (i.e. pH), whereas the LMM approach is best when there is a strong linear relationship (i.e. clay).Geoderma 11/2014; s 232–234:69–80. · 2.51 Impact Factor  [Show abstract] [Hide abstract]
ABSTRACT: Soil bulk density (BD) is measured during soil monitoring. Because it is spatially variable, an appropriate sampling protocol is required. This paper shows how information on shortrange variability can be used to quantify uncertainty of estimates of mean BD and soil organic carbon on a volumetric basis (SOCv) at a sampling site with different sampling intensities. We report results from two contrasting study areas, with mineral soil and with peat. More sites should be investigated to develop robust protocols for nationalscale monitoring, but these results illustrate the methodology. A 20 × 20m2 monitoring site was considered and sampling protocols were evaluated under geostatistical models of our two study areas. At sites with local soil variability comparable to our mineral soil, sampling at 16 points (4 × 4 square grid of interval 5 m) would achieve a root mean square error (RMSE) of the sample mean value of both BD and SOCv of less than 5% of the mean (topsoil and subsoil). Pedotransfer functions (PTFs) gave predictions of mean soil BD at a sample site, comparable to our study area on mineral soil, with similar precision to a single direct measurement of BD. On peat soils comparable to our second study area, the mean BD for the monitoring site at depth 0–50 cm would be estimated with RMSE to be less than 5% of the mean with a sample of 16 cores, but at greater depths this criterion cannot be achieved with 25 cores or fewer.European Journal of Soil Science 10/2014; · 2.39 Impact Factor  [Show abstract] [Hide abstract]
ABSTRACT: We briefly describe three methods of seabed characterization which are ‘fit for purpose’, in that each approach is well suited to distinct objectives e.g. characterizing glacial geomorphology and shallow glacial geology vs. rapid prediction of seabed sediment distribution via geostatistics. The methods vary from manual ‘expert’ interpretation to increasingly automated and mathematically based models, each with their own attributes and limitations. We would note however that increasing automation and mathematical sophistication does not necessarily equate to improve map outputs, or reduce the time required to produce them. Judgements must be made to select methodologies which are most appropriate to the variables mapped, and according to the extent and presentation scale of final maps. http://www.earthdoc.org/publication/publicationdetails/?publication=7778909/2014;  [Show abstract] [Hide abstract]
ABSTRACT: Soil in a changing world is subject to both anthropogenic and environmental stresses. Soil monitoring is essential to assess the magnitude of changes in soil variables and how they affect ecosystem processes and human livelihoods. However, we cannot always be sure which sampling design is best for a given monitoring task. We employed a rotational stratified simple random sampling (rotStRS) for the estimation of temporal changes in the spatial mean of saturated hydraulic conductivity (Ks) at three sites in central Panama in 2009, 2010 and 2011. To assess this design's efficiency we compared the resulting estimates of the spatial mean and variance for 2009 with those gained from stratified simple random sampling (StRS), which was effectively the data obtained on the first sampling time, and with an equivalent unexecuted simple random sampling (SRS). The poor performance of geometrical stratification and the weak predictive relationship between measurements of successive years yielded no advantage of sampling designs more complex than SRS. The failure of stratification may be attributed to the small largescale variability of Ks. Revisiting previously sampled locations was not beneficial because of the large smallscale variability in combination with destructive sampling, resulting in poor consistency between revisited samples. We conclude that for our Ks monitoring scheme, repeated SRS is equally effective as rotStRS. Some problems of smallscale variability might be overcome by collecting several samples at close range to reduce the effect of smallscale variation. Finally, we give recommendations on the key factors to consider when deciding whether to use stratification and rotation in a soil monitoring scheme.European Journal of Soil Science 09/2014; · 2.39 Impact Factor  [Show abstract] [Hide abstract]
ABSTRACT: Deficiency or excess of certain trace elements in the soil causes problems for agriculture, including disorders of grazing ruminants. Geostatistics has been used to map the probability that trace element concentrations in soil exceed or fall below particular thresholds. However, deficiency or toxicity problems may depend on interactions between elements in the soil. Here we show how cokriging from a regional survey of topsoil geochemistry can be used to map the risk of deficiency, and the best management intervention, where both depend on the interaction between two elements. Our case study is on cobalt. Farmers and their advisors in Ireland use index values for the concentration of total soil cobalt and manganese to identify where grazing sheep are at risk of cobalt deficiency. We use topsoil data from a regional geochemical survey across six counties of Ireland to form local cokriging predictions of cobalt and manganese concentrations with an attendant distribution which reflects the joint uncertainty of these predictions. From this distribution we then compute conditional probabilities for different combinations of cobalt and manganese index values, and so for the corresponding inferred risk to sheep of cobalt deficiency and the appropriateness of different management interventions. We represent these results as maps, using a verbal scale for the communication of uncertain information. This scale is based on one used by the Intergovernmental Panel on Climate Change, modified in light of some recent research on its effectiveness.Geoderma 08/2014; s 226–227:64–78. · 2.51 Impact Factor  [Show abstract] [Hide abstract]
ABSTRACT: Marine spatial planning and conservation need underpinning with sufficiently detailed and accurate seabed substrate and habitat maps. Although multibeam echosounders enable us to map the seabed with high resolution and spatial accuracy, there is still a lack of fitforpurpose seabed maps. This is due to the high costs involved in carrying out systematic seabed mapping programmes and the fact that the development of validated, repeatable, quantitative and objective methods of swath acoustic data interpretation is still in its infancy. We compared a wide spectrum of approaches including manual interpretation, geostatistics, objectbased image analysis and machinelearning to gain further insights into the accuracy and comparability of acoustic data interpretation approaches based on multibeam echosounder data (bathymetry, backscatter and derivatives) and seabed samples with the aim to derive seabed substrate maps. Sample data were split into a training and validation data set to allow us to carry out an accuracy assessment. Overall thematic classification accuracy ranged from 67% to 76% and Cohen’s kappa varied between 0.34 and 0.52. However, these differences were not statistically significant at the 5% level. Misclassifications were mainly associated with uncommon classes, which were rarely sampled. Map outputs were between 68% and 87% identical. To improve classification accuracy in seabed mapping, we suggest that more studies on the effects of factors affecting the classification performance as well as comparative studies testing the performance of different approaches need to be carried out with a view to developing guidelines for selecting an appropriate method for a given dataset. In the meantime, classification accuracy might be improved by combining different techniques to hybrid approaches and multimethod ensembles.Continental Shelf Research 08/2014; 84:107119. · 2.12 Impact Factor 
Article: An index to represent lateral variation of the confidence of experts in a 3D geological model
[Show abstract] [Hide abstract]
ABSTRACT: A Confidence Index is proposed that expresses the confidence of experts in the quality of a 3D model as a representation of the subsurface at particular locations. The Confidence Index is based on the notion that the variation of the height of a particular geological surface represents general geological variability and local variability. The general variability comprises simple trends which allow the modeller to project surface structure at locations remote from direct observations. The local variability limits the extent to which borehole observations constrain inferences which the modeller can make concerning local fluctuations around the broad trends. The general and local geological variability of particular contacts are modelled in terms of simple trend surfaces and variogram models. These are then used to extend measures of confidence that reflect expert opinion so as to assign a confidence value to any location where a particular contact is represented in a model. The index is illustrated with an example from the East Midlands region of the United Kingdom.Proceedings of the Geologists Association 07/2014; · 1.33 Impact Factor  [Show abstract] [Hide abstract]
ABSTRACT: The multivariate cumulants characterize aspects of the spatial variability of a regionalized variable. A centred multivariate Gaussian random variable, for example, has zero thirdorder cumulants. In this paper it is shown how the thirdorder cumulants can be used to test the plausibility of the assumption of multivariate normality for the porosity of an important formation, the Bunter Sandstone in the North Sea. The results suggest that the spatial variability of this variable deviates from multivariate normality, and that this assumption may lead to misleading inferences about, for example, the uncertainty attached to kriging predictions.Spatial Statistics. 02/2014;  [Show abstract] [Hide abstract]
ABSTRACT: Marine spatial planning and conservation need underpinning with sufficiently detailed and accurate seabed substrate and habitat maps. Although multibeam echosounders enable us to map the seabed with high resolution and spatial accuracy, there is still a lack of fitforpurpose seabed maps. This is due to the high costs involved in carrying out systematic seabed mapping programmes and the fact that the development of validated, repeatable, quantitative and objective methods of swath acoustic data interpretation is still in its infancy. We compared a wide spectrum of approaches including manual interpretation, geostatistics, objectbased image analysis and machinelearning to gain further insights into the accuracy and comparability of acoustic data interpretation approaches based on multibeam echosounder data (bathymetry, backscatter and derivatives) and seabed samples with the aim to derive seabed substrate maps. Sample data were split into a training and validation data set to allow us to carry out an accuracy assessment. Overall thematic classification accuracy ranged from 67% to 76% and Cohen׳s kappa varied between 0.34 and 0.52. However, these differences were not statistically significant at the 5% level. Misclassifications were mainly associated with uncommon classes, which were rarely sampled. Map outputs were between 68% and 87% identical. To improve classification accuracy in seabed mapping, we suggest that more studies on the effects of factors affecting the classification performance as well as comparative studies testing the performance of different approaches need to be carried out with a view to developing guidelines for selecting an appropriate method for a given dataset. In the meantime, classification accuracy might be improved by combining different techniques to hybrid approaches and multimethod ensembles.Continental Shelf Research 01/2014; · 2.12 Impact Factor  [Show abstract] [Hide abstract]
ABSTRACT: It is important to understand how and where pollution and other anthropogenic processes compromise the ability of urban soil to serve as a component of the natural infrastructure. An extensive survey of the topsoil of the Greater London Area (GLA) in the United Kingdom has recently been completed by a nonprobability systematic sampling scheme. We studied data on lead content from this survey. We examined an overall hypothesis that land use, as recorded at the time of sampling, is an important source of the variation of soil lead content, and we examined specific orthogonal contrasts to test particular hypotheses about land use effects. The assumption that the residuals from land use effects are independent random variables cannot be sustained because of the nonprobability sampling. For this reason modelbased analyses were used to test the hypotheses. One particular contrast, between the lead content in the soil of domestic gardens and that in the soil under parkland or recreational land, was modelled as a spatially dependent random variable, predicted optimally by cokriging. We found that land use is an important source of variation in lead content of topsoil. Industrial sites had the largest mean lead content, followed by domestic gardens. Detailed contrasts between land uses are reported. For example, the lead content in soil of parkland did not differ significantly from that of recreational land, but the soil in these two land uses, considered together, had significantly less lead than did the soil of domestic gardens. Local cokriging predictions of this contrast varied substantially, and were larger in outer parts of the GLA, particularly in the south west.Geoderma 11/2013; s 209–210:65–74. · 2.51 Impact Factor  [Show abstract] [Hide abstract]
ABSTRACT: As soils come under increasing pressure to maintain a range of ecosystem services, there is interest in how soils change over time in response to factors such as change in landuse. Many studies examining long and shortterm soil change have focused on soils with relatively high mineral and fertility status. Therefore, the aims of this study are to explore regional change on a marginal sandy soil formed over the Sherwood Sandstone outcrop in Nottinghamshire, U.K. (750 km2) and to assess changes in soil fertility as a function of the natural weathering process and land use change. The study uses data from three sources to examine differences between soil fertility properties under two major landuses through the depth of the soil/mobile regolith (~ 1.6 m) and into the saprolite. It is proposed that the differences reflect in part the result of historical change in landuse. From old maps we identify the landuse changes back to 1781. This allowed us to compare soils that have been under woodland cover at least since 1781 with those that were converted to arable use in major deforestation between 1781 and 1881. Soils now under woodland have low concentrations of base cations, an acid pH and a mean organic carbon concentration (0–15 cm) of 2.7%. In contrast soils now under arable use have large concentrations of base cations, pH close to neutral and mean organic carbon concentration (0–15 cm) of 1.7%. There is evidence in the arable soils of leaching to depth of materials from applied fertilisers and lime. These results show the rapid change in properties of soil formed in bedrock, with small concentrations of nutrients and weatherable minerals, which can result from landuse change.Geoderma 10/2013; s 207–208:35–48. · 2.51 Impact Factor  [Show abstract] [Hide abstract]
ABSTRACT: We analyzed data on nitrous oxide emissions and on soil properties that were collected on a 7.5km transect across an agricultural landscape in eastern England using the discrete wavelet packet transform. We identified a wavelet packet "best basis" for the emission data. Wavelet packet basis functions are used to decompose the data into a set of coefficients that represent the variation in the data at different spatial frequencies and locations. The "best basis" for a set of data is adapted to the variability in the data by ensuring that the spatial resolution of local features is good at those spatial frequencies where variation is particularly intermittent. The best basis was shown to be adapted to represent such intermittent variation, most markedly at wavelengths of 100 m or less. Variation at these wavelengths was shown to be correlated particularly with chemical properties of the soil, such as nitrate content. Variation at larger wavelengths showed less evidence of intermittency and was found to be correlated with soil chemical and physical constraints on emission rates. In addition to frequencydependent intermittent variation, it was found that the variance of emission rates at some wavelengths changed at particular locations along the transect. One factor causing this appeared to be contrasts in parent material. The complex variation in emission rates identified by these analyses has implications for how emission rates are estimated.Journal of Environmental Quality 07/2013; 42(4):10709. · 2.35 Impact Factor  [Show abstract] [Hide abstract]
ABSTRACT: We develop an algorithm for optimizing the design of multiphase soil remediation surveys. The locations of observations in later phases are selected to minimize the expected loss incurred from misclassification of the local contamination status of the soil. Unlike in existing multiphase design methods, the location of multiple observations can be optimized simultaneously and the reduction in the expected loss can be forecast. Hence rational decisions can be made regarding the resources which should be allocated to further sampling. The geostatistical analysis uses a copulabased spatial model which can represent general types of variation including distributions which include extreme values. The algorithm is used to design a hypothetical second phase of a survey of soil lead contamination in Glebe, Sydney. Observations for this phase are generally dispersed on the boundaries between areas which, according to the first phase, either require or do not require remediation. The algorithm is initially used to make remediation decisions at the point scale, but we demonstrate how it can be used to inform over blocks.Spatial Statistics. 05/2013; 4:1–13.  [Show abstract] [Hide abstract]
ABSTRACT: The revised Environmental Protection Act Part 2A contaminated land Statutory Guidance (England and Wales) makes reference to 'normal' levels of contaminants in soil. The British Geological Survey has been commissioned by the United Kingdom Department for Environment, Food and Rural Affairs (Defra) to estimate contaminant levels in soil and to define what is meant by 'normal' for English soil. The Guidance states that 'normal' levels of contaminants are typical and widespread and arise from a combination of both natural and diffuse pollution contributions. Available systematically collected soil data sets for England are explored for inorganic contaminants (As, Cd, Cu, Hg, Ni and Pb) and benzo[a]pyrene (BaP). Spatial variability of contaminants is studied in the context of the underlying parent material, metalliferous mineralisation and associated mining activities, and the built (urban) environment, the latter being indicative of human activities such as industry and transportation. The most significant areas of elevated contaminant concentrations are identified as contaminant domains. Therefore, rather than estimating a single national contaminant range of concentrations, we assign an upper threshold value to contaminant domains. Our representation of this threshold is a Normal Background Concentration (NBC) defined as the upper 95% confidence limit of the 95th percentile for the soil results associated with a particular domain. Concentrations of a contaminant are considered to be typical and widespread for the identified contaminant domain up to (and including) the calculated NBC. A robust statistical methodology for determining NBCs is presented using inspection of data distribution plots and skewness testing, followed by an appropriate data transformation in order to reduce the effects of point source contamination.Science of The Total Environment 04/2013; 454455C:604618. · 3.16 Impact Factor  [Show abstract] [Hide abstract]
ABSTRACT: Threedimensional framework models are the state of the art to present geologists’ understanding of a region in a form that can be used to support planning and decision making. However, there is little information on the uncertainty of such framework models. This paper reports an experiment in which five geologists each produced a framework model of a single region in the east of England. Each modeller was provided with a unique set of borehole observations from which to make their model. Each set was made by withholding five unique validation boreholes from the set of all available boreholes. The models could then be compared with the validation observations. There was no significant betweenmodeller source of variation in framework model error. There was no evidence of systematic bias in the modelled depth for any unit, and a statistically significant but small tendency for the mean error to increase with depth below the surface. The confidence interval for the predicted height of a surface at a point ranged from ±5.6 m to ±6.4 m. There was some evidence that the variance of the model error increased with depth, but no evidence that it differed between modellers or varied with the number of closeneighbouring boreholes or distance to the outcrop. These results are specific to the area that has been modelled, with relatively simple geology, and reflect the relatively dense set of boreholes available for modelling. The method should be applied under a range of conditions to derive more general conclusions.Proceedings of the Geologists Association 04/2013; · 1.33 Impact Factor  01/2013; Routledge., ISBN: 9781849713672
 [Show abstract] [Hide abstract]
ABSTRACT: This paper illustrates the potential for statistical mapping of seabed sediment texture classes. It reports the analysis of legacy data on the composition of seabed sediment samples from the UK Continental Shelf with respect to three particle size classes (sand, mud, gravel). After appropriate transformation for compositional variables the spatial variation of the sediment particle size classes was modelled geostatistically using robust variogram estimators to produce a validated linear model of coregionalization. This was then used to predict the composition of seabed sediments at the nodes of a fine grid. The predictions were backtransformed to the original scales of measurement by a Monte Carlo integration over the prediction distribution on the transformed scale. This approach allowed the probability to be computed for each class in a classification of seabed sediment texture, at each node on the grid. The probability of each class, and derived information such as the class of maximum probability could therefore be mapped. Predictions were validated at a set of 2000 randomly sampled locations. The class of maximum probability corresponded to the observed class with a frequency of 0.7, and the uncertainty of this prediction was shown to depend on the absolute probability of the class of maximum probability. Other tests showed that this geostatistical approach gives reliable predictions with meaningful uncertainty measures. This provides a basis for rapid mapping of seabed sediment texture to classes with sound quantification of the uncertainty. Remapping to revised class definitions can also be done rapidly, which will be of particular value in habitat mapping where the seabed geology is an important factor in biotope modelling.Sedimentary Geology 12/2012; 281:35–49. · 2.13 Impact Factor  [Show abstract] [Hide abstract]
ABSTRACT: This paper develops and demonstrates a model of stochastic spatial variation. It is proposed that this model may represent soil variability according to a particular mode under which the soil varies continuously, showing shortrange lateral trends induced by local effects of the factors of soil formation which vary across the region of interest in an unpredictable way. The trends in soil variation are therefore only apparent locally, and the soil variation at regional scale appears random. Such variation might be expected in a landscape where the soil varies along topographic catenas which repeat across the region in response to a drainage pattern which is not entirely regular in spacing or orientation, and is therefore unpredictable. The Continuous Local Trend (CLT) mode of soil variation may also be expected where gradients of soil properties are induced around individual plants, or plant roots.In the stochastic model the local trend is assumed to be described by a function of distance to the nearest event in a realisation of a random spatial point process. A model is developed here in which the point process shows complete spatial randomness, so it is called the Poisson Continuous Local Trend (PCLT) model. The covariance function for the PCLT with a general distance function is developed and some hypothetical examples are shown, including one in which the variogram of a soil property is inferred by using a published topofunction. The PCLT model is then fitted to the empirical variogram of some data on soil water content in a gently undulating clay landscape, and the multiple point statistics of the PCLT model for these data are compared with those of a corresponding multivariate normal model.Geoderma 11/2012; s 189–190:661–670. · 2.51 Impact Factor  [Show abstract] [Hide abstract]
ABSTRACT: In most spatial analysis of soil variation it is assumed that the random variation not captured by fixed effects (class means or continuous covariates) is spatially dependent. It is proposed that this should be tested formally, both to justify the kriging component in subsequent spatial prediction and as evidence of the extent to which the included fixed effects have succeeded in accounting for soil variation that is spatially dependent at the scales resolved by the soil sampling. A formal test is possible by computing the log ratio of the likelihoods for a full model with spatially dependent random effects and a null model which is pure nugget. It is shown that the sampling distribution of the log likelihoodratio under the null model is not χ2(p) where p is the number of additional random effects parameters in the model with spatial dependence. This is because, while the null model is nested in the full model, parameters of the full model take bounding values in the null case. The sampling distribution may be computed by Monte Carlo simulations. It is shown that the power to reject the null model by the log likelihoodratio test depends on the importance of the nugget effect in the underlying model, and on the sampling scheme. In many circumstances it may be hard to demonstrate spatial dependence. The recommended procedure was applied to some data on the organic carbon content of the topsoil and subsoil of a field in England. This was modelled either with the overall mean the only fixed effects, or with separate means for different soil map units as fixed effects. There was significant evidence for spatial dependence in the random effects at both depths when the overall mean was the only fixed effect. When map unit means were used as fixed effects there was significant, though weaker, spatial dependence in the topsoil, but the null model could not be rejected for the subsoil. This has implications for any further sampling to map organic carbon in the subsoil.Geoderma 09/2012; s 185–186:102–109. · 2.51 Impact Factor 
Article: Sedimentary Geology
[Show abstract] [Hide abstract]
ABSTRACT: This paper illustrates the potential for statistical mapping of seabed sediment texture classes. It reports the analysis of legacy data on the composition of seabed sediment samples from the UK Continental Shelf with respect to three particle size classes (sand, mud, gravel). After appropriate transformation for compositional variables the spatial variation of the sediment particle size classes was modelled geostatistically using robust variogram estimators to produce a validated linear model of coregionalization. This was then used to predict the composition of seabed sediments at the nodes of a fine grid. The predictions were backtransformed to the original scales of measurement by a Monte Carlo integration over the prediction distribution on the transformed scale. This approach allowed the probability to be computed for each class in a classification of seabed sediment texture, at each node on the grid. The probability of each class, and derived information such as the class of maximum probability could therefore be mapped. Predictions were validated at a set of 2000 randomly sampled locations. The class of maximum probability corresponded to the observed class with a frequency of 0.7, and the uncertainty of this prediction was shown to depend on the absolute probability of the class of maximum probability. Other tests showed that this geostatistical approach gives reliable predictions with meaningful uncertainty measures. This provides a basis for rapid mapping of seabed sediment texture to classes with sound quantification of the uncertainty. Remapping to revised class definitions can also be done rapidly, which will be of particular value in habitat mapping where the seabed geology is an important factor in biotope modelling.Sedimentary Geology 08/2012; 281:3549. · 2.13 Impact Factor
Publication Stats
2k  Citations  
360.24  Total Impact Points  
Top Journals
Institutions

2012–2014

British Geological Survey
 NERC Isotope Geosciences Laboratory (NIGL)
Nottigham, England, United Kingdom


1970–2013

Rothamsted Research
Harpenden, England, United Kingdom


2010

Universidad Politécnica de Madrid
 Chemistry and Agricultural Analysis
Madrid, Madrid, Spain


2009

AgriFood and Biosciences Institute
Béal Feirste, N Ireland, United Kingdom


2008–2009

University of Florida
 Department of Soil and Water Science
Gainesville, FL, United States


2006–2008

Cranfield University
Cranfield, England, United Kingdom


2007

University of Reading
Reading, England, United Kingdom


1995–1998

University of Oxford
 Department of Plant Sciences
Oxford, ENG, United Kingdom


1994–1996

University of Wales
Cardiff, Wales, United Kingdom
