Christopher D. Lippitt

Object-based classification of residential land use within Accra, Ghana based on QuickBird satellite data

A segmentation and hierarchical classification approach applied to QuickBird multispectral satellite data was implemented, with the goal of delineating residential land use polygons and identifying low and high socio‐economic status of neighbourhoods within Accra, Ghana. Two types of object‐based classification strategies were tested, one based on spatial frequency characteristics of multispectral data, and the other based on proportions of Vegetation–Impervious–Soil sub‐objects. Both approaches yielded residential land‐use maps with similar overall percentage accuracy (75%) and kappa index of agreement (0.62) values, based on test objects from visual interpretation of QuickBird panchromatic imagery.

Can Error Explain Map Differences Over Time

This paper presents methods to test whether map error can explain the observed differences between two points in time among categories of land cover in maps. Such differences may be due to two reasons: error in the maps and change on the ground. Our methods use matrix algebra: (1) to determine whether error can explain specific types of observed categorical transitions between two maps, (2) to represent visually the differences between the maps that error cannot explain, and (3) to examine how the results are sensitive to possible variation in map error. The methods complement conventional accuracy assessment because they rely on standard confusion matrices that use either a random or a stratified sampling design. We illustrate the methods with maps from 1971 and 1999, which show seven land-cover categories for central Massachusetts. The methods detect four transitions from agriculture, range, forest, and barren in 1971 to built in 1999, which a 15 percent error cannot explain. Sensitivity analysis reveals that if the accuracy of the maps were less than 77 percent, then error could explain virtually all of the observed differences between the maps. The paper discusses the assumptions behind the methods and articulates priorities for future research.

Mapping Selective Logging in Mixed Deciduous Forest: A Comparison of Machine Learning Algorithms

This study assesses the performance of five Machine Learning Algorithms (MLAs) in a chronically modified mixed deciduous forest in Massachusetts (USA) in terms of their ability to detect selective timber logging and to cope with deficient reference datasets. Multitemporal Landsat Enhanced Thematic Mapperplus (ETM+) imagery is used to assess the performance of three Artificial Neural Networks ‐ Multi-Layer Perceptron, ARTMAP, Self-Organizing Map, and two Classification Tree splitting algorithms: gini and entropy rules. MLA performance evaluations are based on susceptibility to reduced training set size, noise, and variations in the training set, as well as the operability/transparency of the classification process. Classification trees produced the most accurate selective logging maps (gini and entropy rule decision tree mean overall map accuracy � 94 percent and mean per-class kappa of 0.59 and 0.60, respectively). Classification trees are shown to be more robust and accurate when faced with deficient training data, regardless of splitting rule. Of the neural network algorithms, self-organizing maps were least sensitive to the introduction of noise and variations in training data. Given their robust classification capabilities and transparency of the classselection process, classification trees are preferable algorithms for mapping selective logging and have potential in other forest monitoring applications.

Connecting the Dots Between Health, Poverty and Place in Accra, Ghana

West Africa has a rapidly growing population, an increasing fraction of which lives in urban informal settlements characterized by inadequate infrastructure and relatively high health risks. Little is known, however, about the spatial or health characteristics of cities in this region or about the spatial inequalities in health within them. In this article we show how we have been creating a data-rich field laboratory in Accra, Ghana, to connect the dots between health, poverty, and place in a large city in West Africa. Our overarching goal is to test the hypothesis that satellite imagery, in combination with census and limited survey data, such as that found in demographic and health surveys (DHSs), can provide clues to the spatial distribution of health inequalities in cities where fewer data exist than those we have collected for Accra. To this end, we have created the first digital boundary file of the city, obtained high spatial resolution satellite imagery for two dates, collected data from a longitudinal panel of 3,200 women spatially distributed throughout Accra, and obtained microlevel data from the census. We have also acquired water, sewerage, and elevation layers and then coupled all of these data with extensive field research on the neighborhood structure of Accra. We show that the proportional abundance of vegetation in a neighborhood serves as a key indicator of local levels of health and well-being and that local perceptions of health risk are not always consistent with objective measures.

Geographic Object-based Delineation of Neighborhoods of Accra, Ghana using QuickBird Satellite Imagery

The objective was to test GEographic Object-based Image Analysis (GEOBIA) techniques for delineating neighborhoods of Accra, Ghana using QuickBird multispectral imagery. Two approaches to aggregating census enumeration areas (EAs) based on image-derived measures of vegetation objects were tested: (1) merging adjacent EAs according to vegetation measures and (2) image segmentation. Both approaches exploit readily available functions within commercial GEOBIA software. Image-derived neighborhood maps were compared to a reference map derived by spatial clustering of slum index values (from census data), to provide a relative assessment of potential map utility. A size-constrained iterative segmentation approach to aggregation was more successful than standard image segmentation or feature merge techniques. The segmentation approaches account for size and shape characteristics, enabling more realistic neighborhood boundaries to be delineated. The percentage of vegetation patches within each EA yielded more realistic delineation of potential neighborhoods than mean vegetation patch size per EA.

Urban Vegetation Cover and Vegetation Change in Accra, Ghana: Connection to Housing Quality

The objectives are to (1) quantify, map, and analyze vegetation cover distributions and changes across Accra, Ghana, for 2002 and 2010; and (2) examine the statistical relationship between vegetation cover and a housing quality index (HQI) for 2000 at the neighborhood level. Pixel-level vegetation cover maps derived using threshold classification of 2002 and 2010 QuickBird normalized difference vegetation index images have very high overall accuracies and yield an estimate of 5.9 percent vegetation cover reduction over the study area between 2002 and 2010. A high degree of variance in vegetation cover for individual dates is explained by HQI at the neighborhood level, although minimal covariability between absolute or relative vegetation cover change and HQI for 2000 was observed.

The effect of input data transformations on object-based image analysis

The effect of using spectral transform images as input data on segmentation quality and its potential effect on products generated by object-based image analysis are explored in the context of land cover classification in Accra, Ghana. Five image data transformations are compared to untransformed spectral bands in terms of their effect on segmentation quality and final product accuracy. The relationship between segmentation quality and product accuracy is also briefly explored. Results suggest that input data transformations can aid in the delineation of landscape objects by image segmentation, but the effect is idiosyncratic to the transformation and object of interest.

Time–space radiometric normalization of TM/ETM+ images for land cover change detection

A novel approach to image radiometric normalization for change detection is presented. The approach referred to as stratified relative radiometric normalization (SRRN) uses a time-series of imagery to stratify the landscape for localized radiometric normalization. The goal is to improve the detection accuracy of abrupt land cover changes (human-induced, natural disaster, etc.) while decreasing false detection of natural vegetation changes that are not of interest. These vegetation changes may be associated with such phenomena as phenology, growth and stress (e.g. drought), which occur at varying spatial and temporal scales, depending on landscape position, vegetation type, season, precipitation history and historic episodes of local disturbance. The SRRN approach was tested for a study area on the Californian border between the USA and Mexico using Landsat Thematic Mapper and Enhanced Thematic Mapper Plus satellite imagery. Change products were generated from imagery radiometrically normalized using the SRRN procedure and with imagery normalized using a traditional empirical line technique. Reference data derived from high spatial resolution airborne imagery were utilized to validate the two change products. The SRRN procedure provided several benefits and was found to improve the overall accuracy of detecting abrupt land cover changes by nearly 20%.

Frequency distribution signatures and classification of within-object pixels.

The premise of geographic object-based image analysis (GEOBIA) is that image objects are composed of aggregates of pixels that correspond to earth surface features of interest. Most commonly, image-derived objects (segments) or objects associated with predefined land units (e.g., agricultural fields) are classified using parametric statistical characteristics (e.g., mean and standard deviation) of the within-object pixels. The objective of this exploratory study was to examine the between- and within-class variability of frequency distributions of multispectral pixel values, and to evaluate a quantitative measure and classification rule that exploits the full pixel frequency distribution of within object pixels (i.e., histogram signatures) compared to simple parametric statistical characteristics. High spatial resolution Quickbird satellite multispectral data of Accra, Ghana were evaluated in the context of mapping land cover and land use and socioeconomic status. Results show that image objects associated with land cover and land use types can have characteristic, non-normal frequency distributions (histograms). Signatures of most image objects tended to match closely the training signature of a single class or sub-class. Curve matching approaches to classifying multi-pixel frequency distributions were found to be slightly more effective than standard statistical classifiers based on a nearest neighbor classifier.

Remote Sensing Theory and Time-Sensitive Information

The role of time in remote sensing is complex and, particularly in the case of time-sensitive remote sensing, can impose substantial control on the utility of information derived using remote sensing techniques. A common vernacular is used to define the role of time in remote sensing as it is currently understood. The various manifestations of time in the remote sensing process are then evaluated in terms of their potential effect on the utility of information produced. The concept of information utility, which necessitates acknowledgment of remote sensing’s role as a information production system, provides an orienting construct for evaluating the affect of time on the remote sensing process.