Remotely sensed data have considerable potential for mapping and monitoring tropical forests. For the production of regional scale maps which may be up-dated periodically, relatively coarse spatial resolution remotely sensed data, such as those from the NOAA AVHRR, are an appropriate source of data for such mapping applications. These maps, however, typically depict land cover at the nominal level only and may be unsuitable for the estimation of forest extent and dynamics. In this paper, results of an investigation into the estimation of sub-pixel forest cover and classification at the ordinal level are presented. Based on an analysis of Landsat MSS data that had been degraded spatially to a 1.2-km resolution, a strong correlation, r = 0.94, was observed between predicted and actual sub-pixel forest cover (...)
Typicalities of class membership generated from a maximum likelihood classification may be used to increase classification accuracy by the modification of class allocation on the basis of additional ground surveys. Whilst this increases the amount of ground survey required the additional survey effort is directed to regions where there is potential misclassification. Directing surveys to those cases which displayed, for instance, a typicality of less than 0·05 to their most likely class of membership increased significantly the accuracy of a crop classification with synthetic aperture radar data by 11·70 per cent to 77·27 per cent.
Accuracy assessment should be a fundamental part of a land cover mapping programme but often constrained by the lack of ground data. Here, two sources of volunteered data are used to illustrate the potential of neogeographical activity in map validation. Ground based photographs acquired by an internet-based collaborative project and interpreted by a set of volunteers provided the reference data to support evaluation of the Globcover map's representation of tropical forest in West Africa. The results highlight concerns with volunteered data, notably the low levels of agreement but also show that imperfect data may be used to derive useful information on map properties and accuracy.
The landscape patches that are fundamental to landscape ecology may be considered as objects to be extracted from remotely sensed imagery. The accuracy with which objects may be characterised varies as a function of the spatial resolution of the imagery used. In general terms, a coarsening of the spatial resolution degrades the characterization of objects, notably through an increase in the proportion of mixed pixels which cannot be appropriately represented by conventional hard classification techniques. Accurate landscape mapping may often require either the adoption of fine spatial resolution imagery or use of sub-pixel scale analyses of coarse spatial resolution imagery. As the former is often impractical, the full realization of the potential of remote sensing as a source of information on landscape objects requires developments in sub-pixel scale techniques. In this paper, a new method of superresolution mapping based on a unifying framework of image halftoning, inverse halftoning and Hopfield neural network techniques is proposed as a means of gaining accurate information on landscape patches from coarse spatial resolution images. Fine temporal resolution of coarse spatial resolution remote sensing systems is exploited by fusing the time-series data as an input for the superresolution mapping. The accuracy of the analyses is evaluated relative to conventional a hard classification technique using object characterization. The results show that the proposed hybrid method is considerably more accurate than standard hard analyses in estimating the shape of the objects. The results also demonstrate that objects that are smaller than a pixel, which are missed using the hard classification techniques, can be detected using the super-resolution mapping. * Corresponding author.
Estimates of the area or percent area of the land cover classes within a study region are often based on the reference land cover class labels assigned by analysts interpreting satellite imagery and other ancillary spatial data. Different analysts interpreting the same spatial unit will not always agree on the land cover class label that should be assigned. Two approaches for accommodating interpreter variability when estimating the area are simple averaging (SA) and latent class modeling (LCM). This study compares agreement between area estimates obtained from SA and LCM using reference data obtained by seven trained, professional interpreters who independently interpreted an annual time series of land cover reference class labels for 300 sampled Landsat pixels. We also compare the variability of the LCM and SA area estimates over different numbers of interpreters and different subsets of interpreters within each interpreter group size, and examine area estimates of three land cover classes (forest, developed, and wetland) and three change types (forest gain, forest loss, and developed gain). Differences between the area estimates obtained from SA and LCM are most pronounced for the estimates of wetland and the three change types. The percent area estimates of these rare classes were usually greater for LCM compared to SA, with the differences between LCM and SA increasing as the number of interpreters providing the reference data increased. The LCM area estimates generally had larger standard deviations and greater ranges over different subsets of interpreters, indicating greater sensitivity to the selection of the individual interpreters who carried out the reference class labeling.
RÉSUMÉLes variations topographiques influencent les relations angulaires entre la cible et le capteur. comme la rétrodiffusion des microondes par une cible donnée est fonction de ces relations angulaires, des cibles similaires observées suivant des géométries angulaires différentes produiront une rétrodiffusion différente et, par conséquent, présenteront des niveaux de gris différents sur les images radar. Dans le cas de terrains plats, ces variations sont introduites uniquement par les géométries de visée obliques du capteur. Dans le cas de terrains plus complexes, des cibles observées suivant le même angle de dépression seront quand même observées suivant des angles d'incidence locaux différents en raison des différences topographiques locales. Dans le cas de régions vallonnées, on a constaté que ces effets représentaient entre 15 % et 50 % de la variance des niveaux de gris de l'image. Les effets topographiques sont également fonction de la longueur d'onde, avec des tendances différentes pour les bandes X et L des images RAS (radar à antenne de synthèse).SUMMARYTopographic variations influence the angular relationship between target and sensor. Since microwave backscatter from a target is dependent on this angular relationship, similar targets viewed at different angular geometries will backscatter differently and so exhibit dissimilar tones on radar imagery. For flat terrain these variations result only from the off-nadir viewing geometry of the system. Where the terrain is more complex, targets viewed at the same depression angle may still be viewed at different local angles of incidence because of topographic differences in their siting. For a region of undulating terrain these effects were found to account for around 15% and up to 50% of the variance in image tone. The topographic effects also showed a wavelength dependency, with different trends observed for X- and L- band SAR imagery.
The maximum-likelihood classification of remotely sensed data involves considerable computational effort, in the process calculating a large amount of information on the class membership characteristics for each case (e.g., pixel). Little of this information, however, is made available in the conventional output, which consists simply of the most likely class of membership for each case. More of the information generated in the classification can be output, specifically the a posteriori probabilities and typicalities of class membership.
Volunteered Geographic Information (VGI) offers a potentially inexpensive source of reference data for estimating area and assessing map accuracy in the context of remote-sensing based land-cover monitoring. The quality of observations from VGI and the typical lack of an underlying probability sampling design raise concerns regarding use of VGI in widely-applied design-based statistical inference. This article focuses on the fundamental issue of sampling design used to acquire VGI. Design-based inference requires the sample data to be obtained via a probability sampling design. Options for incorporating VGI within design-based inference include: 1) directing volunteers to obtain data for locations selected by a probability sampling design; 2) treating VGI data as a “certainty stratum” and augmenting the VGI with data obtained from a probability sample; and 3) using VGI to create an auxiliary variable that is then used in a model-assisted estimator to reduce the standard error of an estimate produced from a probability sample. The latter two options can be implemented using VGI data that were obtained from a non-probability sampling design, but require additional sample data to be acquired via a probability sampling design. If the only data available are VGI obtained from a non-probability sample, properties of design-based inference that are ensured by probability sampling must be replaced by assumptions that may be difficult to verify. For example, pseudo-estimation weights can be constructed that mimic weights used in stratified sampling estimators. However, accuracy and area estimates produced using these pseudo-weights still require the VGI data to be representative of the full population, a property known as “external validity”. Because design-based inference requires a probability sampling design, directing volunteers to locations specified by a probability sampling design is the most straightforward option for use of VGI in design-based inference. Combining VGI from a non-probability sample with data from a probability sample using the certainty stratum approach or the model-assisted approach are viable alternatives that meet the conditions required for design-based inference and use the VGI data to advantage to reduce standard errors.
Training patterns are of unequal importance in image classification. For classification by a neural network, training patterns that lie close to the location of decision boundaries in feature space may aid the derivation of an accurate classification. The role of such border training patterns is investigated. A neural network trained with border patterns had a lower accuracy of learning but significantly higher accuracy of generalisation than one trained with patterns drawn from the class cores. Unfortunately, conventional training pattern selection and refinement procedures tend to favour core training patterns.
Training set characteristics can have a significant effect on the performance of an image classification. In this paper the effect of variations in training set size and composition on the accuracy of classifications of synthetic and remotely sensed data sets by an artificial neural network and discriminant analysis are assessed. Attention is focused on the effects of variations in the overall size of the training set, in terms of the number of training samples, as well as on variations in the size of individual classes in the training set. The results showed that higher classification accuracies were generally derived from the artificial neural network, especially when small training sets only were available. It was also apparent that the opportunity of the artificial neural network to learn class appearance was influenced by the composition of the training set. The results indicated that the size of each class in the training set had an effect similar to. that of including a priori probabilities of class membership into the discriminant analysis. In the classification of the remotely sensed data set the classification accuracy was increased significantly as a result of increasing the number of training cases for abundant classes in the image.
Pattern recognition is concerned with a range of information processing issues associated with the description or classification of measurements. It is based on a broad and often loosely related body of literature and techniques (Schalkoff 1992). Although statistical...
Crowdsourcing is a popular means of acquiring data, but the use of such data is limited by concerns with its quality. This is evident within cartography and geographical sciences more generally, with the quality of volunteered geographic information (VGI) recognized as a major challenge to address if the full potential of citizen sensing in mapping applications is to be realized. Here, a means to characterize the quality of volunteers, based only on the data they contribute, was used to explore issues connected with the quantity and quality of volunteers for attribute mapping. The focus was on data in the form of annotations or class labels provided by volunteers who visually interpreted an attribute, land cover, from a series of satellite sensor images. A latent class model was found to be able to provide accurate characterisations of the quality of volunteers in terms of the accuracy of their labelling, irrespective of the number of cases that they labelled. The accuracy with which a volunteer could be characterized tended to increase with the number of volunteers contributing but was typically good at all but small numbers of volunteers. Moreover, the ability to characterize volunteers in terms of the quality of their labelling could be used constructively. For example, volunteers could be ranked in terms of quality which could then be used to select a sub-set as input to a subsequent mapping task. This was particularly important as an identified subset of volunteers could undertake a task more accurately than when part of a larger group of volunteers. The results highlight that both the quantity and quality of volunteers need consideration and that the use of VGI may be enhanced through information on the quality of the volunteers derived entirely from the data provided without any additional information.
Drought, associated with the El Niño Southern Oscillation (ENSO), can have considerable impact on tropical rainforests. Concern over drought, particularly given the possibility of an increase in its occurrence and intensity, has fostered a desire for an increased understanding of drought events and their impact to inform the development of a drought monitoring system. This paper investigates the use of National Oceanic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR) data in a drought monitoring system for the rainforests of Sabah, Borneo. These rainforests are dynamic with respect to their coupling with ENSO processes and in their biophysical properties, and such dynamism may have implications for how NOAA AVHRR data may be used. In particular, this paper explores the transferability of relationships between a drought indicator (rainfall) and the response of the rainforest, as measured by four NOAA AVHRR variables (middle infrared reflectance; VI3; Ts/VI3 and Ts/NDVI), under particular site conditions. It was found that both spatial variability in forest biophysical properties and geographical variability in drought impact had implications for the transferability of relationships developed under local conditions across Sabah rainforests within a drought monitoring system. Suggestions are presented for how NOAA AVHRR data could be used, with a new drought monitoring index – the Ts/VI3 – recommended.
No abstract is provided for this article.
1. School of Geography, University of Nottingham, Nottingham, NG7 2RD, UK 2. Department of Computer Science, Maynooth University, Maynooth, Ireland 3. Ecosystems Services and Management Program, International Institute for Applied Systems Analysis (IIASA), Laxenburg, Austria 4. Faculty of Geo-Information Science and Earth Observation, University of Twente, Enschede, The Netherlands 5. IGN France, COGIT Laboratory, 73 Avenue de Paris, 94160 Saint-Mande, France 6. Department of Mathematics, University of Coimbra / INESC Coimbra, Coimbra, Portugal
