Doreen S. Boyd

Predictive relations of tropical forest biomass from Landsat TM data and their transferability between regions

The full realization of the potential of remote sensing as a source of environmental information requires an ability to generalize in space and time. Here, the ability to generalize in space was investigated through an analysis of the transferability of predictive relations for the estimation of tropical forest biomass from Landsat TM data between sites in Brazil, Malaysia and Thailand. The data sets for each test site were acquired and processed in a similar fashion to facilitate the analyses. Three types of predictive relation, based on vegetation indices, multiple regression and feedforward neural networks, were developed for biomass estimation at each site. For each site, the strongest relationships between the biomass predicted and that measured from field survey was obtained with a neural network developed specifically for the site (r>0.71, significant at the 99% level of confidence). However, with each type of approach problems in transferring a relation to another site were observed. In particular, it was apparent that the accuracy of prediction, as indicated by the correlation coefficient between predicted and measured biomass, declined when a relation was transferred to a site other than that upon which it was developed. Part of this problem lies with the observed variation in the relative contribution of the different spectral wavebands to predictive relations for biomass estimation between sites. It was, for example, apparent that the spectral composition of the vegetation indices most strongly related to biomass differed greatly between the sites. Consequently, the relationship between predicted and measured biomass derived from vegetation indices differed markedly in both strength and direction between sites. Although the incorporation of test site location information into an analysis resulted in an increase in the strength of the relationship between predicted and actual biomass, considerable further research is required on the problems associated with transferring predictive relations.

One-Class Classification for Mapping a Specific Land-Cover Class: SVDD Classification of Fenland

Remote sensing is a major source of land-cover information. Commonly, interest focuses on a single land-cover class. Although a conventional multiclass classifier may be used to provide a map depicting the class of interest, the analysis is not focused on that class and may be suboptimal in terms of the accuracy of its classification. With a conventional classifier, considerable effort is directed on the classes that are not of interest. Here, it is suggested that a one-class-classification approach could be appropriate when interest focuses on a specific class. This is illustrated with the classification of fenland, a habitat of considerable conservation value, from Landsat Enhanced Thematic Mapper Plus imagery. A range of one-class classifiers is evaluated, but attention focuses on the support-vector data description (SVDD). The SVDD was used to classify fenland with an accuracy of 97.5% and 93.6% from the users and producers perspectives, respectively. This classification was trained upon only the fenland class and was substantially more accurate in fen classification than a conventional multiclass maximum-likelihood classification provided with the same amount of training data, which classified fen with an accuracy of 90.0% and 72.0% from the users and producers perspectives, respectively. The results highlight the ability to classify a single class using only training data for that class. With a one-class classification, the analysis focuses tightly on the class of interest, with resources and effort not directed on other classes, and there are opportunities to derive highly accurate classifications from small training sets

Size and frequency of natural forest disturbances and the Amazon forest carbon balance

Forest inventory studies in the Amazon indicate a large terrestrial carbon sink. However, field plots may fail to represent forest mortality processes at landscape-scales of tropical forests. Here we characterize the frequency distribution of disturbance events in natural forests from 0.01 ha to 2,651 ha size throughout Amazonia using a novel combination of forest inventory, airborne lidar and satellite remote sensing data. We find that small-scale mortality events are responsible for aboveground biomass losses of ~1.7 Pg C y−1 over the entire Amazon region. We also find that intermediate-scale disturbances account for losses of ~0.2 Pg C y−1, and that the largest-scale disturbances as a result of blow-downs only account for losses of ~0.004 Pg C y−1. Simulation of growth and mortality indicates that even when all carbon losses from intermediate and large-scale disturbances are considered, these are outweighed by the net biomass accumulation by tree growth, supporting the inference of an Amazon carbon sink.

An overview of recent remote sensing and GIS based research in ecological informatics

Abstract This article provides an overview of some of the recent research in ecological informatics involving remote sensing and GIS. Attention focuses on a selected range of issues including topics such as the nature of remote sensing data sets, issues of accuracy and uncertainty, data visualization and sharing activities as well as developments in aspects of ecological modelling research. It is shown that considerable advances have been made over recent years and foundations for future research established.

Mapping specific habitats from remotely sensed imagery: Support vector machine and support vector data description based classification of coastal saltmarsh habitats

Abstract Remote sensing has considerable potential for the provision of information on the distribution of habitats that may be used to inform a variety of activities such as those required through the European Unions Habitats Directive. Such programmes are often resource-limited with a need for innovative methods that optimise resource use. This paper explores two approaches to resource savings when mapping habitats from remotely sensed imagery. The first approach realises that in an area of study often interest is focused on a specific habitat with the remaining land cover classes in the region of no importance. In such circumstances conventional statistical supervised classification analyses may be inefficient and yield a map of sub-optimal accuracy. The second approach seeks to further reduce the training requirements of a supervised classification. For this support vector machine (SVM) based approaches to classification are explored to map coastal saltmarsh habitats in North Norfolk, UK from a Landsat Enhanced Thematic Mapper (ETM+) image. A series of classifications using SVM based approaches and the Maximum Likelihood classifier (MLC) were undertaken. Classification accuracies were significantly higher using the SVM based approaches (e.g., 92.0% overall accuracy) than the MLC (64.8% overall accuracy). The SVM based classifications were demonstrated to be attractive for mapping a priority habitat in that the focus is on the habitat of interest to be mapped throughout the classification process resulting in a reduced need for training data. Moreover, it was shown that this can be further optimised through the use of intelligent training. This approach, based on the use of the support vector data description (SVDD), saved resource requirements even further in that training data were required only for the class of interest and yet still obtained high classification accuracies (95.2% overall accuracy) . The wider adoption of SVM based classification of remotely sensed imagery is advocated for use in conservation activities.

Assessing the Accuracy of Volunteered Geographic Information arising from Multiple Contributors to an Internet Based Collaborative Project

The recent rise of neogeography and citizen sensing has increased the opportunities for the use of crowdsourcing as a means to acquire data to support geographical research. The value of the resulting volunteered geographic information is, however, often limited by concerns associated with its quality and the degree to which the contributing data sources may be trusted. Here, information on the quality of sources of volunteered geographic information was derived using a latent class analysis. The volunteered information was on land cover interpreted visually from satellite sensor images and the main focus was on the labeling of 299 sites by seven of the 65 volunteers who contributed to an Internet-based collaborative project. Using the information on land cover acquired by the multiple volunteers it was shown that the relative, but not absolute, quality of the data from different volunteers could be characterized accurately. Additionally, class-specific variations in the quality of the information provided by a single volunteer could be characterized by the analysis. The latent class analysis, therefore, was able to provide information on the quality of information provided on an inter- and intra-volunteer basis.

The relationship between the biomass of Cameroonian tropical forests and radiation reflected in middle infrared wavelengths (3.0-5.0 mu m)

The use of middle infrared (MIR) radiation (3.0-5.0 mu m) at the regional scale may be unreliable for biophysical estimation, should be corrected for thermal emission and MIR reflectance used in its place. This study considered the potential use of MIR reflectance for studying tropical forests, with the relationship between MIR reflectance and estimated total biomass of the tropical forests of Cameroon derived. Comparisons were drawn with relationships between estimated total biomass and visible reflectance, near infrared reflectance, MIR radiation and surface temperature. Relationships between two vegetation indices, the NDVI and VI3, and estimated total biomass were also explored. It was found that correcting MIR radiation for thermal emission increased the strength of the relationship between radiation acquired in MIR wavelengths and estimated total biomass. The use of MIR reflectance, either alone or within the vegetation index VI3, provided the strongest relationship with estimated total biomass. Thi...

Mapping a specific class with an ensemble of classifiers

Often, in remote sensing, interest is focused on just one of the many classes that are typically represented in the area covered by an image. Various binary classifiers may be used to separate this specific class of interest from all others. It can, however, be difficult to identify the most appropriate classifier in advance. The selection of classifier is also often further complicated by a desire for information on classification uncertainty to indicate the spatial variation in classification quality. Here, five classifiers (a discriminant analysis, decision tree, support vector machine, multi‐layer perceptron, and radial basis function neural network) were used to map fenland, an important class for conservation activities, from Landsat ETM+data. The classifications derived ranged in accuracy from 81.2 to 96.8%. The outputs of the classifications were also combined using a simple voting procedure to determine class allocation. The accuracy of this ensemble approach was 95.6%. Although marginally, but insignificantly (at 95% level of confidence), less accurate than the most accurate individual classifier, it is difficult to specify the most appropriate classifier in advance. In addition, the ensemble approach yielded class‐allocation uncertainty information that may be used to help post‐classification refinement operations and later analyses. For example, as only a small proportion of cases were allocated with a high degree of uncertainty and these contained most of the mis‐classifications, targeting such sites for fieldwork could be one simple and efficient means of increasing classification accuracy. Alternatively, the cases for which all five classifiers agreed on an allocation could be treated as being correctly labelled with a high degree of confidence.

Evaluation of approaches for forest cover estimation in the Pacific Northwest, USA, using remote sensing

The transformation of land cover, in particular coniferous forest, constitutes one of the most notable agents of regional-to-global-scale environmental change. Remote sensing provides an excellent opportunity for providing forest cover information at appropriate spatial and temporal scales. The optimal exploitation of remote sensing relies on the link between known forest cover and the remotely sensed dataset. This paper explores the accuracy of three methods – vegetation indices, regression analysis and neural networks – for estimating coniferous forest cover across the United States Pacific Northwest. All methods achieved a similar accuracy of forest cover estimation. However, in view of the benefits and limitations of each, the neural network approach is recommended for future consideration.

Mapping a specific class for priority habitats monitoring from satellite sensor data

The European Unions Habitats Directive aims to protect biodiversity through the conservation of habitats. If a habitat of interest corresponds to spectrally separable land cover class(es), then this activity can benefit from the production of accurate land cover maps from remotely sensed imagery. Traditionally, the image classification techniques used assume that the set of classes has been defined exhaustively, which requires all of the classes in the region to be included explicitly in the analysis. Often, however, interest focuses on just one or a small sub‐set of the classes occurring in the region that represent the habitat(s) of particular interest. Moreover, given that the size of a training set required for an image classification is typically taken to be a function of the number of classes and discriminating variables (e.g. wavebands) used in the classification, the satisfaction of the assumption of an exhaustively defined set of classes requires that much effort is directed wastefully on classes of little, if any, direct interest. Savings in training could be achieved by focusing on the class(es) of specific interest. A more appropriate approach for mapping a specific class may be to adopt a binary classification analysis that simply seeks to separate the class of interest from all others. In this way the analysis focuses on the class(es) of interest and a small training set may be used. An attractive means to achieve this is through the adoption of decision tree‐ and support vector machine‐based approaches to classification. This paper evaluates the accuracy with which a habitat of interest to the EU Habitats Directive, fen, can be mapped from Landsat ETM+ imagery of the Norfolk Broads using such classifiers as well as, for comparative purposes, a standard maximum likelihood decision rule implemented by a discriminant analysis. All analyses yielded accurate classifications, with a conventional approach based on a maximum likelihood allocation providing an overall classification accuracy of 88.4%. However, both the decision tree‐ and support vector machine‐based approaches provided classifications that were significantly more accurate than conventional maximum likelihood classification (p<0.05), with overall accuracies of 91.6 and 93.6%, respectively (table 3). The results highlight the ability to focus the analysis on the class of interest in a manner that is less wasteful of resources and effort and that yields a more accurate classification than the standard approach.