Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
1 Linking Remote Sensing and Social Science: The Need and the Challenges Ronald R. Rindfuss and Paul C. Stern There is increased interest today in making scientific progress through the use of remotely sensed datai in social science research. Space-based sensors are scanning the earth's surface and sending back images with increasingly high spatial, spectral, and temporal resolution, and data likely to become publicly available within the next year promise to show considerably improved resolu- tion.2 Government agencies that collect remotely sensed data, such as the Na- tional Aeronautics and Space Administration (NASA) and National Oceanic and Atmospheric Administration (NOAA), have a growing interest in making these data useful to social scientists, and the increased availability of funding for re- search on the human dimensions of global change provides incentives for social scientists to study human activities with a strong spatial component, such as land- use transformations. This confluence of events sets the stage for social scientists to use remotely sensed data and for social scientists and remote sensing experts to collaborate.3 This volume examines the potential for such use. It offers some guidance for researchers and research sponsors in the form of reports of promis- ing research, information on the state of the technology, and reflections on the challenges of linking social science and remotely sensed data. Remote sensing is not a new technology. Aerial photographs have been in widespread use for a half-century (Carls, 1947) and satellite images for a quarter- century (e.g., Estes et al., 1980; Morain, in this volume). These images have been put to various socially useful purposes, including making crop forecasts, predicting severe storms, and planning land development. Despite the apparent usefulness of remotely sensed data for social purposes, however, remotely sensed images have not been a popular data source for social science research, for several reasons.
2 HNKING REMOTE SENSING AND SOCIA:L SCIENCE: THE NEED AND THE CHALLENGES First, the variables of greatest interest to many social scientists are not readily measured from the air. Many social scientists find visible human artifacts such as buildings, crop fields, and roads less interesting than the abstract variables that explain their appearance and transformations. Changing land use, road and build- ing construction, and the like are regarded as manifestations of more important variables, such as government policies, land-tenure rules, distributions of wealth and power, market mechanisms, and social customs, none of which is directly reflected in the bands of the electromagnetic spectrum. Thus social scientists are likely to be skeptical that remote sensing can measure anything considered im- portant in their fields of study (Turner, in press). A related issue is that social science is generally more concerned with why things happen than where they happen (Turner, in press). Even areas of social science in which one might expect a spatial orientation are curiously aspatial. For example, while it seems almost self-evident that spatial propinquity must be a factor in the shaping of social networks, it is only recently that the spatial aspects of social networks have been receiving attention (Faust et al., 1997~. Relatively few social scientists outside the field of geography value the spatial explicitness that remotely sensed data provide, nor do the typical social science data sets contain the geographic coordinates that would facilitate linking social science data and remotely sensed data. Further, the scientists who participate in developing remote sensing tech- niques have overlapped little with social scientists in their backgrounds, theories, methods, jargon, or epistemological approaches, although this situation is chang- ing. Integrating social science and remote sensing will require the fusion not only of data, but also of quite different scientific traditions. Many social scientists do not know what a pixel is, and few have ever considered how clouds may affect data quality. Similarly, the average remote sensing expert is unlikely to be conversant with a wide range of social science problems and solutions, such as why fixed-effects statistical models were developed. It is easy for scientists on one side to underestimate the difficulty of learning the approaches, theories, methods, and jargon of the other. This difficulty is compounded by the fact that those on each side are likely to have some familiarity with the other. Social scientists are likely to have been watching news and weather reports for decades, acquiring what they think is an ability to interpret satellite imagery. Some of the images that appear on the television screen bear a close resemblance to familiar objects, such as maps, making interpretation seem easy. But in fact, these pro- cessed images can be several steps removed from the remotely sensed data on which they are based. People who see only these products may have little appre- ciation of the analysis necessary to produce them. Similarly, socioeconomic patterns and trends are discussed frequently in the mass media. It is easy for those not trained in social science to claim some understanding of it and to think that incorporating it into their research would be straightforward. But as with remote sensing, popularized presentations can mask the detailed analysis that lies
RONALD R. RINDFUSS AND PAUL C. STERN 3 behind the summary data. We return to the issue of different scientific traditions later, in discussing the problem of training future scholars. Finally, bridging the social science and remote sensing fields undoubtedly entails the risks frequently encountered by those who do interdisciplinary re- search. For example, when we discuss the issues raised in this chapter with remote sensing experts who were originally trained in the social sciences, they frequently mention feeling marginalized from their original fields of study be- cause the problems and concepts central to their remote sensing research are not considered core to those disciplines. Estes et al. (1980) have discussed this issue from the perspective of geography the discipline in which social science and remote sensing have most closely converged and we are told that the situation remains much the same in the late l990s. Given this gap between the social sciences and remote sensing, why bother trying to bridge the two fields? The question has different answers for different kinds of scientists. For some remote sensing experts, a compelling answer is social utility: remote sensing is expensive, and government spending on it is more justifiable if it improves our understanding of the social system by being incorporated into social science research. While remotely sensed data have been employed for a variety of socially useful purposes, such as increasing yields through precision farming or weather forecasting, there are, as noted, relatively few examples of those data being used in social science research. The social utility argument posits that remote sensing becomes even more valuable to the extent that social scientists find it useful, and that efforts should be made to identify and overcome the barriers to making this happen. In addition, the contri- butions of social scientists might allow remote sensing experts to "see" landscape features in the remotely sensed data not previously apparent. There are several examples of this in the present volume. From the perspective of social science, one important reason for using re- motely sensed data is to gather information on the context that shapes social phenomena. The role of context has been central to the theories and empirical work of numerous sociologists, economists, and anthropologists. Remote sens- ing offers an additional source of contextual data for multilevel analyses. An- other consideration involves the growing interdisciplinary community of scien- tists interested in sustainable development, pollution prevention, global environmental change, and related issues of human-environment interaction who need to compare data on social and environmental phenomena at the same spatial and temporal scales. This community includes both social and physical scien- tists. For them, fusing social and remotely sensed data should be an attractive strategy. Although linking of remote sensing and social science is difficult, it has been and continues to be done, as evidenced by the case studies presented in this volume. This chapter examines why it is important to join people and pixels, addressing some of the challenges of doing so. Understanding the challenges is
4 HNKING REMOTE SENSING AND SOCIA:L SCIENCE: THE NEED AND THE CHALLENGES essential if progress is to be made. In considering the promise and opportunities offered by collaboration between remote sensing and social science, we want to strike a balance between overpromising and underselling. We do not believe remote sensing will quickly revolutionize social science; rather, we suggest that some progress can be made by joining social science and remote sensing perspec- tives, techniques, and data. Hence, the majority of this volume consists of ex- amples of the use of remotely sensed data mainly from space-based platforms- in social science research.4 However, we do not want to be overly constrained by the present, so we also speculate about additional, as yet untried, applications of remote sensing to social science questions. The volume is intended to stimulate dialogue between social science and remote sensing experts, and in any dialogue, it helps to know something about the participants. The Committee on the Human Dimensions of Global Change at the National Research Council, which is responsible for this volume, consists of social and natural scientists interested in the scientific understanding of human- environment interactions. The majority of the committee members were trained in the social sciences, and the authors of this chapter are typical in this regard. One of us (Rindfuss) is trained in sociology, with research and teaching interests in population studies or demography. He began graduate school wanting to understand American fertility patterns and trends, and then branched out geo- graphically and substantively, while continuing to publish on fertility in the United States. He has used remotely sensed data in work on population migration and social change in Nang Rong district, Thailand. He has worked mainly with micro-level data sets in which individuals or households are the units of analysis. He has been involved in various interdisciplinary activities through the Popula- tion Association of America, the National Research Council, and other multidisciplinary organizations and has directed an interdisciplinary research cen- ter, but approaches the topic of people and pixels from the vantage point of a sociologist/demographer. The other author (Stern) is trained in social psychology, but has long been interested in human-environment interactions, particularly in behaviors at the individual and household levels that affect the use of natural resources and the generation of waste and pollution. He has worked with data on individuals' attitudes, beliefs, and behavior and has studied the effects of interventions aimed at changing environmentally significant behavior at the local and regional levels. He has published research together with colleagues from various disciplines in the social and natural sciences, but despite his experience in interdisciplinary collaboration and the potential of remote sensing to provide data on the environ- mental effects of social interventions, he has not used remotely sensed data in his research.
RONALD R. RINDFUSS AND PAUL C. STERN WHAT CAN REMOTE SENSING DO FOR SOCIAL SCIENCE? s One rationale for linking people and pixels is that doing so might result in better social science research. This could happen in several ways, although the realistic potential for making these improvements is in some dispute. Measuring the Context of Social Phenomena Many social science theories relate individual or household behavior to the context within which the individual or household is located. "Context" can denote a variety of entities, including a political or administrative unit, a social network, a school, or a racial or ethnic group. When the individual is the unit of analysis, the individual's household is also a context. People live their lives in contexts, and the nature of those contexts structures the way they live. Contexts can provide advantages (for example, growing up within a wealthy school dis- trict) or produce constraints (young adults in rural areas with poor soil quality are more likely to out-migrate). Hypotheses from theories of context may involve additive effects (teenagers residing in high-crime neighborhoods are more likely to become involved in crime than are teenagers in low-crime neighborhoods) or interactive effects (the negative effect of education on fertility is stronger for blacks than for whites) but in either event, the hypotheses concern the effects of context on individuals or households. Contexts can be measured in various ways. Censuses, because they obtain information on almost all individuals and households in a country, can be aggre- gated to various units (block, neighborhood, school district, city, county, or state) to provide measures of the demographic or socioeconomic characteristics of those units. The choice of scale and of characteristics depends on the theory and the hypothesis being tested; the effects of the contextual variables are estimated using statistical models. Sometimes individual respondents know the contextual variables well enough to provide them directly. Race, ethnicity, and religion are examples. Sometimes contextual variables must be measured with data that are not gathered from individuals. Examples include public expenditures on educa- tion and laws governing land tenure. In the case of social networks, researchers are experimenting with several approaches to measuring the structure of net- works and positioning individual respondents within that structure. Remote sensing provides an additional means of gathering contextual data, particularly in describing the biophysical context within which people live, work, and play. First of all, remotely sensed data provide an alternative representation of geographical context to that given by maps. Maps always include the map- maker's selection of what is important to represent, and remotely sensed data, though also imperfect representations of reality, have different biases. They can therefore offer a check on what is in maps, additional information, and sometimes a useful alternative perspective. A good general source of information on meth
6 HNKING REMOTE SENSING AND SOCIA:L SCIENCE: THE NEED AND THE CHALLENGES oafs of measuring and understanding geographical contexts is the recent volume Rediscovering Geography (National Research Council, 1997~. In addition, remote sensing has the potential to supplement georeferenced social data by characterizing numerous aspects of the context, ranging from land cover to soil moisture to weather. An example is the work of one of the authors (Rindfuss) in Nang Rong district, Thailand, mentioned earlier. This work (the subject of Chapter 6 in this volume) concerns the determinants of out-migration. The Nang Rong district is a rural, agricultural area, with rice being the predomi- nant crop. Over 90 percent of the adults are farmers, so access to land is essential for young adults seeking employment in the district. Since forested land in Nang Rong tended to have ambiguous legal titles, it was expected that the availability of forested land would reduce the likelihood of out-migration of young adults, and this was indeed found to be the case (Rindfuss et al., 1996~. Another example comes from the work of Geoghegan and colleagues (see Chapter 3), who are analyzing the effect of the mosaic of land uses surrounding a property on its economic value and the probability of its future development. Migration raises one of the thorniest issues related to the use of contextual data to study social processes: individuals and households can change their contexts through migration. People move for many reasons, including better economic opportunities, better schools, and a preferred biophysical environment. When they move and change the context in which they live, that context needs to be modeled as an endogenous variable, rather than a simple external influence on behavior. Even without migration, individuals can act to change their contexts- a possibility that may be more easily uncovered when contexts are measured in interviews than when they are measured by remote sensing. Thus, theoretical care is needed when using remotely sensed data to supply contextual data for models of individual or household behavior. Measuring Social Phenomena and Their Effects Remote sensing can provide measures for a number of dependent variables associated with human activity particularly regarding the environmental conse- quences of various social, economic, and demographic processes. For example, remote observations of land cover may show the footprints of agricultural exten- sification, urbanization, and road development;5 observations of vegetation den- sity may be related to the effects of fertilization, irrigation, and other agricultural practices; and observations of new building construction may be linked to the effects of local policies on land use and property taxation. Remote sensing has sometimes proven to be the best method for identifying archaeological sites and relating them to key features of their geographical settings (see Chapter 7~. Models that combine remote observation with ground-based social data have the potential to improve understanding of the determinants of various land-use changes. Geoghegan and colleagues in Chapter 3 and Cowen and Jensen in
RONALD R. RINDFUSS AND PAUL C. STERN 7 Chapter 8 give examples of such modeling in which residential development is the variable being predicted. It may also be possible to study the effects of changes in agricultural commodity prices on cropping patterns and tillage prac- tices by combining price data with remotely sensed data, and to improve under- standing further by incorporating additional data on land-tenure systems or agri- cultural policies. Providing Additional Measures for Social Science Social scientists frequently use aggregated units of analysis: cities or towns, counties or districts, states or provinces, or countries. The substantive questions vary, but investigators typically use multiple indicators for these aggregated units. Remote sensing can provide a variety of additional indicators for these studies, including land cover, moisture measures, locations of major roads and hydro- graphic features, and indicators of crop fertility. Gathering such measures from the ground might be possible, but often is prohibitively expensive because of the need to collect large amounts of small-scale data for aggregation. Remote sens- ing can sometimes provide highly aggregated data at less cost. Indicators from remote sensing can complement indicators from ground- based sources. For example, agricultural intensification can be measured by using data from surveys of farmers' behavior, sales figures on agricultural chemi- cals and farm equipment, or remotely sensed data on crop density and color. Combinations of social and remote data can yield a deeper understanding of the types of intensification possible, as shown in the analysis of Amazonian agro- forestry in Chapter 5. Urbanization can be measured by counting building per- mits, sampling and observing city blocks, or remotely sensing the proportion of land covered by structures (see Chapter 81. Each data source has its imperfec- tions, but combining sources with different limitations might provide a better picture of the entire phenomenon. In this way, remote sensing even with its imperfections can make a contribution to social scientific measurement by im- proving on some measures and cross-checking others. Because remotely sensed data are available with greater spatial and temporal resolution than data from other sources, there has been discussion of the potential for using the former data to conduct finer-grained studies than are possible with typical social science data. These possibilities are expanding as higher-resolution data become available from new satellites and satellite data collected by military and intelligence organizations in the United States and the former Soviet Union are declassified. For example, census data are remarkably accurate in most countries, but they are collected infrequently, typically every 10 years. And there are some countries for which census data are not available, and some in which the data are reported inaccurately for a variety of cultural and geopolitical reasons. Some have expressed the hope that remotely sensed data could be used during intercensal periods to update the census reports. Cowen and Jensen (Chapter 8)
8 HNKING REMOTE SENSING AND SOCIA:L SCIENCE: THE NEED AND THE CHALLENGES report on correlations between remotely sensed indicators of dwelling units and actual census counts in data from South Carolina. The validity of their indicators in other geographical and social contexts or over time has not yet been demon- strated, however. There have been some suggestions that remotely sensed data of fine spatial resolution might be used in statistical models to generate estimates of population counts. If that were possible, there would be numerous uses of such estimates. It remains to be seen whether efforts along these lines will yield accurate estimates. Before this becomes possible, however, a number of method- ological studies are necessary because of certain inherent limitations in the use of remote sensing for population estimates, such as the inability to sense the number of people per housing unit or housing units per building, or to discriminate between residential buildings and some others. Thus, ground-based studies are necessary to determine how the number of people per dwelling unit varies with the socioeconomic and physical characteristics of neighborhoods. If the variance is sufficiently systematic, remote sensing might help improve intercensal popula- tion estimates. Remote sensing might also help with the census undercount problem. One source of a census undercount is the failure to recognize a physical structure that is a dwelling unit. For relatively remote rural areas, finding dwelling units is a difficult undertaking, and missing a dwelling unit can contribute to the under- count.6 The use of satellite images with high spatial resolution might improve this process a possibility the U.S. Bureau of the Census has investigated using aerial photographs (Carls, 1947~. Remotely sensed data have been used for measuring other socially signifi- cant variables, especially in urban and suburban contexts. Cowen and Jensen (Chapter 8) describe the use of remote observation to classify land use and land cover into categories established by the U.S. Geological Survey; to measure the area, height, and volume of buildings; to measure traffic patterns and road condi- tions; to estimate residential energy demand; and to build predictive models of residential expansion. Some of these measurement methods are in the early stages of development, so more experience is necessary to determine how well they work across a variety of social and geographic conditions and over longer periods of time. Nevertheless, these measures may provide important advantages in cost or temporal resolution over conventional measures of the same variables, and may make it possible to improve the quality of modeling used for planning urban infrastructure needs and forecasting the need for utilities or other public services. It may be argued that remote sensing can support comparative social re- search studies that attempt to draw conclusions by systematically comparing the same phenomena in different countries or different regions of the same coun- try by providing comparable measures of variables these studies need to inves- tigate. Clearly, remote sensing is well suited to providing comparable data for
RONALD R. RINDFUSS AND PAUL C. STERN 9 different geographic regions or at different times. The question is whether the parallel social data are available in forms that are comparable. Making Connections Across Levels of Analysis Social science disciplines and subdisciplines have their preferred levels of analysis and often do not communicate across those levels. For instance, psy- chologists and sociocultural anthropologists tend to work with individuals and small groups; political scientists and geographers tend to work at higher levels defined by political units or geophysical features; while sociologists tend to specialize in one level of analysis or another, from individuals to small groups to communities to the world system. Remotely sensed data are essentially global in coverage,7 composed of individual pixels that can be combined to allow work at any scale or level of analysis more coarse than the pixel size. Thus remotely sensed data offer some potential for encouraging social scientists to think across levels of analysis and to develop theories that link these levels. An example is in the work of Moran and Brondizio documented in Chapter 5 which, starting from an anthropological and highly localized perspective, developed ways of examin- ing land use in geographically disparate areas of Amazonia and thereby address- ing regional-level questions. Similarly, Entwisle and colleagues (Chapter 6) collected data on villages in a way that, with the help of comparable data across villages from remote sensing and social surveys, has the potential to place the behavior of individuals in the context of their villages and interpret the character- istics of the village in the context of the region. The issue of linking levels of analysis is explored in more detail in Chapter 3, which suggests some interesting possibilities for combining remotely sensed and ground-based data to study the effects of global economic forces on the behavior of individual land users. Chap- ter 3 also considers a cousin of the linkage issue in the temporal dimension: the property of path dependence in dynamic systems, which may be affected by their histories as well as their current conditions. Providing Time-Series Data on Socially Relevant Phenomena Time-series data can be helpful when social scientists attempt to trace rela- tionships of cause and effect but cannot use experimental methods. Remote platforms sometimes provide time-series data of good comparability (i.e., the same variables measured in the same way across time) on variables of interest to social scientists concerned with the effects of context on behavior or with pro- cesses of human-environment interaction. Examples in this volume include data on thinning of forests by human action (Chapter 3), forest regrowth after clear- cutting (Chapter 5), and development of algal blooms that harbor pathogens (Chapter 10~. In addition, remotely sensed time-series data can be essential for modeling human-environment interactions. Examples in this volume include the
10 HNKING REMOTE SENSING AD SOCKS SCIENCE: THE NEED AD THE CHALLENGES use of remotely sensed data to model the effects of access to forests on out- migration (Chapter 6) and processes of land conversion to urban uses (for ex- ample, in Chapter 3~. WHAT CAN SOCIAL SCIENCE DO FOR REMOTE SENSING? As noted earlier, to the extent that remote observations provide uniquely useful information for social research, these social science applications of remote sensing can be used to provide additional justification for the money spent on observational platforms and data management systems. In addition to this poten- tial practical value of social science to remote sensing, there are several kinds of scientific contributions to remote sensing that might come from its interaction with social science. Validation and Interpretation of Remote Observations Remote sensing specialists are well aware of the need for "ground trothing," that is, for validating remote observations against data collected on the ground. A standard example is the problem of measuring land cover by old-growth forests (Lucas et al., 1993; Moran et al., 1994; Skole et al., 1994; Moran and Brondizio, in this volume). It is necessary among other things to distinguish the spectral signatures of old growth from those of forests regrown after deforestation. Doing so requires comparsion of the remotely observed spectral properties of plots known from ground observation to fall in these categories in order to develop an algorithm that accurately discriminates between the two. Further ground obser- vation is required, of course, to validate the algorithm on plots of land not used to develop it. Although classifying types of land cover requires observations on the ground, it is not usually considered a social science activity. There are, however, some kinds of ground truthing that involve classifying remote observations into more obviously social categories, and thus depend on social science input. An impor- tant example is classification of land uses, which are socially defined in ways that do not correspond exactly to categories of land cover. Thus, some tree cover is socially classified as forest land, some as park land, some as suburban landscap- ing, some as orchard, and some as productive agroforestry land. It is frequently necessary to rely on human informants to make these distinctions. Similarly, different kinds of land tenure, such as family ownership, village commons, and sharecropping arrangements, may all be used in the same kinds of productive activity and may therefore fall within a single land-cover, or even land-use, classification. It may be possible to associate different management practices that can be distinguished spectrally by remote observation with differ- ences in tenure. Discovering such differences would likely require collaboration between remote sensing specialists who can distinguish spectral patterns and
RONALD R. RINDFUSS AND PAUL C. STERN 11 social scientists who can classify land-tenure types and land-management prac- tices. Data Confidentiality and Public Use As noted earlier, remotely sensed data are becoming available for public use in ever finer spatial resolution, increasing the ability to discern the footprints of socially important activities. Moreover, as high-resolution military observations are declassified and made available, various organizations will gain access to information about the landscape heretofore not considered by those responsible for the landscape. As improved technical capabilities, collaboration with social scientists, and especially the linking of remotely sensed data with social data make remotely sensed data increasingly useful, new problems and conflicts may arise over the use of the data. Although there are legal precedents that limit privacy rights with respect to high-resolution aerial photography, the courts have not yet directly addressed questions of privacy and Fourth Amendment rights in the context of space-based remote observation (Uhlir, 1990~. As such observation provides improved reso- lution, new claims of infringements of privacy may surface. There may be new calls for the restriction of access to remotely sensed data, stricter regulations or legislation governing what can be collected remotely, or curtailment of public resources invested in remote sensing technology. There are also unresolved issues of international law (Hosenball, 1990~. By way of illustration, when one of the authors (Rindfuss) first showed his Thai collaborators the satellite images for our study site during a research seminar in Thailand, their first question was: "Where did you get these?" Their tone and facial expressions suggested surprise and perhaps a bit of concern that one could simply buy such images. Informal discussions with others using remotely sensed data suggest that the reactions of our Thai collaborators are not unique. Already, some landowners are concerned that remote platforms will reveal secrets about their land-use practices (perhaps revealing to government officials that those practices are violating land-use regu- lations). We expect that notwithstanding legal precedents in the United States dating from the early days of aerial photography, the increasingly finer resolution and widespread availability of satellite images and their linking to ground-based data sources will fuel public concerns about invasion of privacy and reopen some previously closed issues. Social scientists have experience in dealing with issues of confidentiality in data collection and dissemination that may be of use to remote sensing special- ists. Most social science data collection techniques, from face-to-face interviews to participant observation to the analysis of administrative record forms, require that those providing the information be motivated to remain open and honest, which in turn requires that they trust the researchers to use the information responsibly. There are many examples of censuses, surveys, and other studies in
12 HNKING REMOTE SENSING AND SOCKS SCIENCE: THE NEED AND THE CHALLENGES which respondents suspected that the data might be misused and in which, as a result, response rates or data quality was unacceptably low, or studies had to be canceled. Thus, social scientists have come to recognize their collective self- interest in maintaining respondents' confidentiality, and in vigorously dispelling any rumors that confidentiality may not be maintained or that the data will be used inappropriately. As a result, social scientists plan their research carefully so as to assure respondents that their identities will remain confidential, and their information will be used only for statistical analysis. In addition, universities and research institutes use institutional review boards to review research designs and ensure adequate protection for respondents. Important issues of trust and confidentiality arise with public-use social science data sets, which are increasingly being created in response to the high and escalating cost of collecting large-scale social data. Research sponsors, particu- larly within the federal government, frequently require that researchers make their data publicly available, and this requirement creates a potential conflict with the need to preserve confidentiality. The conflict has tended to be resolved in two different ways. In the first, which has tended to be used with censuses, individual and household data are aggregated to a sufficient level (counties, for example) so that it is no longer possible to identify the individuals or households that provided the data. In the second approach, which is often used with survey data, all identifiers of the respondent (including geographic ones) are stripped from the data set, and data on the individual respondents are then released. What are the implications of these practices for remote sensing? If data are aggregated to large geographic units, social data can be linked with remotely sensed data, although the procedures are technically challenging. However, this solution to the confidentiality problem constrains analyses to larger spatial and social scales. There are many research questions for which analysis at the county level and above is appropriate, but there are also many important research ques- tions that require analysis of individual- or household-level data. If geographic identifiers are removed to allow confidential analysis of individual-level data, the public-use data set cannot be linked to remotely sensed observations or other georeferenced data. Social scientists recognize this prob- lem and have tried various approaches to solving it. Usually, these approaches involve explicit agreements between the data collector and the user that may include legal contracts and the posting of a bond. To date, these agreements have prevented any disastrous breaches of confidentiality, but they are administra- tively cumbersome and can restrict the research process. To the best of our knowledge, there are as yet no examples linking remotely sensed data to publicly available individual- or household-level social science data sets. The linkages that have occurred, including examples in this volume, involve social science data sets that are not yet in the public domain. Remotely sensed data that are not linked to social data are less likely to pose problems of confidentiality. However, making remote sensing more useful may
RONALD R. RINDFUSS AND PAUL C. STERN 13 raise the incidence and severity of conflicts over confidentiality and access to information. Social science, which has more experience with these conflicts, may be able to offer useful insights and institutional responses to the remote sensing community. HOW CAN REMOTE SENSING AND SOCIAL SCIENCE IMPROVE UNDERSTANDING OF HUMAN-ENVIRONMENT INTERACTIONS? An additional argument for better collaboration between remote sensing spe- cialists and social scientists is that such collaboration has been necessitated by a new and important set of intellectual and practical problems: those related to understanding and controlling human impacts on the biophysical environment, as well as anticipating and responding to environmental impacts on humanity. En- vironmental quality has been a major concern of citizens and policy makers for over a quarter-century, and there is a compelling need to understand human- environment interactions. Such understanding depends on better knowledge of biophysical systems, of human activity, and above all, of the relations between the two. Linking remote sensing and social science is a necessary part of devel- oping this knowledge. Interpreting, Modeling, and Predicting the Dynamics of Natural Resources Many projected regional and global environmental problems may be the consequence of human activities that alter land use and land cover (e.g., tropical deforestation and wetland conversion) or affect agricultural productivity (e.g., practices that increase soil erosion). Analyses involving both remotely sensed and social data are critical for understanding these dynamics. Several contributions to this volume illustrate how remotely sensed and social data can be combined to help us understand human-environment linkages. Wood and Skole (Chapter 4) report their initial effort to predict deforestation, habitat fragmentation, and secondary growth in Amazonia from census data on population dynamics, economic activity, and other social indicators at the re- gional level. Entwisle and colleagues (Chapter 6) illustrate some of the dynamics of mutual causation between land cover and human migration, with implications for resource demands in both rural and urban areas of Thailand. And Sever (Chapter 7) graphically illustrates the effects of human settlements and public policy on forest cover in the Peten and adjacent areas of Mexico. Understanding the Human Consequences of Climate Flux Fluctuating weather patterns concern citizens and policy makers because of the tangible effects they may have on human health and well-being. Research
14 HNKING REMOTE SENSING AD SOCKS SCIENCE: THE NEED AD THE CHALLENGES combining social and remotely sensed data can lead to new understandings of the consequences of such fluctuations. For example, Hutchinson (Chapter 9) describes the development of a famine early-warning system for Africa based on remote observations of drought phe- nomena, combined with an understanding of the social processes by which people adapt to drought. Generally, forecasts and status reports on food crop growth in drought-prone regions, derived mainly from remotely sensed data, are combined with ground-based data on patterns of human response to generate famine warn- ings. Such warnings give aid agencies and governments enough advance notice to act to prevent out-migration by threatened human populations. In Chapter 10, Epstein provides several examples of actual and potential uses of remote sensing to link climatic change and variation to human health through changes in the ecology of disease organisms and their vectors. For example, remote sensing of algal blooms in the Indian Ocean can help provide early warn- ing of cholera outbreaks instigated by disease vectors that feed on the algae (Colwell, 1996~. A combination of remotely sensed data and ground observa- tions of rodents helped solve the riddle of a Hantavirus outbreak in the southwest- ern United States and reveal the linkages among climate variation, ecological change, and human health. These examples illustrate how remotely sensed data can be used to build forecasting models that have the potential to serve very important social pur- poses. A significant challenge to social science research is to build understanding of how these forecasting models work and how they can be made more socially useful. MAJOR INTELLECTUAL ISSUES We have already noted that combining social and remotely sensed data pre- sents challenges because of dramatic differences in the intellectual traditions that produce and use the two kinds of data. Once these challenges have been met, as they sometime are, by researchers who are willing to make the effort to under- stand and respect each other's perspectives, several other issues arise to challenge those who would combine the two data sources. Finding the Appropriate Spatial and Temporal Resolution Decisions about the appropriate scale, level of aggregation, and frequency of measurement of various data are driven by considerations of both theory and data availability. On the theoretical level, the appropriate units of analysis depend on the question being asked. For example, the debate over global warming rests mainly on questions about what is happening on a global scale and on a temporal scale of decades to centuries. By contrast, questions about population migration turn on the decisions of individuals and households. Some questions require
RONALD R. RINDFUSS AND PAUL C. STERN 15 analysis at multiple scales. For example, questions about land use and land cover typically require information at the level of individuals and households that may own the land and make many of the decisions on how it is used, local govern- ments (because they often regulate land use and make decisions about the loca- tion of transportation infrastructure), and governments at higher levels. Geoghegan and colleagues (Chapter 3) examine this question of scalar dynamics in more detail. Data availability also influences the choice of scale or aggregation. With remotely sensed data, the characteristics of the sensing instrument determine the finest grain available spatially and the frequency with which measurements can be repeated. On the social science side, the level of resolution is determined by the nature of the data gathering technique and by any aggregation that may have been done before the data were released in a public-use file. For example, data about the actions of legislative bodies are intrinsically gathered at rather highly aggregated levels, whereas data on educational attainment, even if presented in an aggregated format, necessarily begin with information about individuals. Researchers deciding to link social and remotely sensed data must make decisions about the appropriate level of aggregation. On the social side, ignoring temporal issues, the finest grain is an individual. Levels of aggregation above the individual depend on the researcher's theoretical perspective, as well as on data availability. Aggregation by political or geographical units progresses through successively higher levels, such as villages, towns, or cities; counties or districts; states or provinces; countries; and regions or continents. Linking such units to remotely sensed data can be relatively straightforward if the units have clear geographical boundaries. However, when aggregation is based on other kinds of theoretical categories, linking to remotely sensed data is more difficult. For example, individuals can be grouped at successively higher levels of aggregation into families, kin groups, and lineages units that do not always have clear geographic boundaries. Linking such units to remotely sensed or other geocoded data necessarily requires some simplifying assumptions. On the remote sensing side, the issues of linkage are somewhat different. There is no unit that is comparable to the individual. The smallest unit of obser- vation in satellite data is the pixel, and its size is determined by the measuring instrument, rather than by any theoretically meaningful concept. For example, data from the French Systeme Pour ['Observation de la Terre (SPOT) have a much finer spatial resolution than Advanced Very High Resolution Radiometer (AVHRR) data. The size of the pixel, in turn, determines the smallest thing one can "see" with satellite data. A large pond may be quite visible with SPOT data, difficult to see with Multispectral Scanner (MSS) data, and impossible to see with AVHRR data. One can aggregate pixels to produce a coarser resolution as dictated by theoretical or substantive concerns, but it is not possible to go below the pixel for finer resolution.8 (One can, however, combine satellite images with different resolutions or combine satellite images with aerial photographs in an
16 HNKING REMOTE SENSING AD SOCKS SCIENCE: THE NEED AD THE CHALLENGES attempt to merge the properties of various remotely sensed images.) In Chapter 8, Cowen and Jensen summarize the pixel sizes and frequencies of observation necessary for measuring a variety of attributes of urban development. Linking People to Pixels A critical issue in linking people and pixels is the decision on where to georeference individuals or other social units. In some cases, a social unit has a natural georeferent, but frequently this is not the case. Consider individuals, and assume that the substantive questions being examined involve the effect of indi- vidual behavior on some aspect of the land. To which pixel or pixels should the individual be linked? The question can be difficult to answer because, although the land units represented by the pixels do not move, people do. Researchers usually know their respondents' place of residence because this is typically the location of the data collection effort or because data on place of residence are routinely collected, and the natural tendency may be to link the person to the place of residence. In a setting where the primary economic activity is farming and the primary means of transportation is walking, using an area near the person' s residence seems an appropriate way to link people to pixels. This is the assumption used by Entwisle and colleagues in Nang Rong, Thailand (Chap- ter 6), where people live in clustered villages, rice farming is the predominant occupation, and walking is the main means of transport. However, as modes of transportation evolve to enable travel over greater distances and as occupations diversify to include manufacturing and service posi- tions, the validity of linking individuals to the pixels where they reside becomes questionable. For example, in the United States in 1990, only 4 percent of the labor force walked to work, and 3 percent worked at home (calculated from U.S. Bureau of the Census, 1993: Table 18~. The great majority (73 percent) com- muted to work alone in a car, truck, or van. For all those who did not work at home, the mean travel time to work was 22.3 minutes. If their average speed was 48 kilometers (30 miles) per hour, the mean commuting distance was about 18 kilometers (11 miles); if the average speed was 72 kilometers (45 miles) per hour, the mean commute was about 27 kilometers (17 miles). Since people can affect the landscape both at home and at work, one might want to georeference both their residences and their workplaces. Similar considerations apply to shopping patterns, social activities, and religious activities. Routine social science data collection efforts do not provide the capability to georeference these activities; further, tested and accepted methods for collecting such data do not exist. So far, this discussion has considered only residences and workplaces, and assumed that individuals have just one of each. But many workers have more than one job or work in more than one location. Moreover, increasing numbers of Americans own or occupy vacation homes, and certain geographic areas are
RONALD R. RINDFUSS AND PAUL C. STERN 17 being transformed primarily for the purpose of vacationing and tourism by indi- viduals whose primary residences are elsewhere. An additional problem is that of georeferencing the activities of individuals who participate in the global economy and thereby affect the land in places where they may never have traveled. People buy agricultural and other products from widely dispersed places and manufacture products that produce effluents in the widely dispersed locations where they are used, often without even knowing where their activities are having an impact. Social science is a long way from being able to georeference these human activities, but it is obvious that people in modern economies are not easily linked to pixels for the purposes of understand- ing the effects of their economic activities. An analogous problem exists at other levels of analysis. Consider, for ex- ample, linking firms to remotely sensed data in order to understand the influence of different types of firms on land-cover or land-use patterns. Should one use only the point locations of a firm's places of business, or should one also consider the commuting patterns of the firm's employees and the locations of its suppliers of raw materials? One possible approach involves aggregating social data to larger geographi- cal units. This approach assigns individuals to larger areas in which their envi- ronmental effects are more likely to be confined. This approach can answer some scientifically important questions and produce interesting results. Examples can be seen in the work of Wood and Skole in Chapter 4. However, there are limitations to this approach. First, there are numerous processes that are not visible at high levels of aggregation. To exploit fully the potential of integrating social science and remote sensing, one should also examine finer-grained rela- tionships. Second, surveys have become much more important than censuses in most social science disciplines, and the majority of surveys are designed to repre- sent a broad geographic area (a country, region, or state) and not to be represen- tative of smaller geographic divisions within that area. This is true even of very large surveys, such as the Current Population Survey in the United States. It is unusual to have comparable surveys conducted in multiple geographic units, and when they are, the units are usually countries, which are too large for many processes of interest. INSTITUTIONAL ISSUES: MAKING IT HAPPEN As this chapter and the case examples presented in other chapters of this volume make clear, social science and remote sensing are being linked in efforts to address important scientific and public policy questions. The potential of such collaborations is considerable, although there are also significant challenges. The question for the future is not whether this sort of activity should go on it will but how much of it should go on and how it can be facilitated. This section addresses some of the key institutional questions about the future, for example,
18 HNKING REMOTE SENSING AND SOCKS SCIENCE: THE NEED AND THE CHALLENGES how to create a productive community of scholars who combine social science and remote sensing, how to train future scholars for participation in this commu- nity, and how to support the community with needed data. Building a Community of Scholars For the next 5 years, most individuals who work on projects that involve the linking of social science and remotely sensed data will be experts in one of these fields, but probably not both, and will work in collaboration with researchers whose training is complementary. They will acquire some knowledge in the area in which they are not expert, but only the exceptional individual will be expert in both. The volume of research literature in both areas all but precludes an estab- lished researcher's becoming an expert in both. Thus in the near term, research combining social science and remote sensing is likely to be multidisciplinary and interdisciplinary, and will involve all the problems associated with such research. One of these problems is peer review of scientific proposals that incorporate both social science and remote sensing, and that are likely to make contributions to both areas. Ideally, one would want to fund proposals that are of the highest quality and that will make cutting-edge contributions to both the social science and remote sensing components. Social science, however, is a very broad field, and we do not know of anyone who has the audacity to claim expertise in all its subfields. The same is probably true of remote sensing. The peer review process would be best served if it included representatives of all the major subfields of both social science and remote sensing that are represented in the proposals submitted. Given the diversity on both sides, the selection of appropriate review- ers will be a challenging task. Moreover, it may be desirable to fund some projects that break new ground by applying knowledge or techniques that are familiar in either social science or remote sensing in a new and important way, and that are therefore not equally innovative in both fields. Such proposals might be viewed as exciting and innovative by experts in one field, but uninteresting by experts in the other. It will be necessary to find ways of preventing vetoes of such projects by experts in a subfield who do not see the overall value of proposals that go beyond their expertise. Communication of scientific results is another problem for any new interdis- ciplinary scientific field. Communication normally occurs through the auspices of scientific associations that are built around disciplines (e.g., economics, geog- raphy) or problem areas (e.g., population, natural resource management). The communication media range from small workshops to scientific meetings to jour- nals. To date, there is no scientific association or journal for scholars who are integrating social science and remotely sensed data, and this lack of an institu- tional base is likely to impede the development of research at the intersection of the two fields. There are certainly examples of sessions at professional meetings that include papers incorporating both social science and remotely sensed data
RONALD R. RINDFUSS AND PAUL C. STERN 19 (e.g., the 1996 and 1997 meetings of the Population Association of America, the Pecora 12 conference, and the 1997 Open Meeting of the Human Dimensions of Global Change Research Community). But each of these meetings attracts only a small fraction of those bridging the fields of social science and remote sensing, and further, only one member of a research team usually goes to these meetings (e.g., the Pecora conferences tend to attract only remote sensing experts). Scientific journals are perhaps the most important communication mecha- nism. For teams working on projects that use both social and remotely sensed data, the most obvious publication outlets are ones that specialize in only one of the two fields. The peer review process in such journals involves the same problems already noted for the review of proposals: the social science reviewers are generally not competent to review the remote sensing components of the paper, and the remote sensing reviewers are generally not competent to review the social science. The ability to build a community of scholars depends greatly on publications because they are central to the reward system in science. Universities are the primary employers of social scientists and remote sensing researchers, and they tend to be organized along disciplinary lines. Graduate degrees, faculty appoint- ments, and tenure tend to be determined by disciplinary bodies within a vertical structure in which department chairs report to deans, deans to provosts, and so forth. Uniting social science and remote sensing involves horizontal links across departments or even schools. While these horizontal links are crucial for the research, the fact that they are orthogonal to the decision and reward structure of the university means that tensions will inevitably arise. Even a simple question such as where to publish findings will make collaborators consider their self- interests. Further, many department chairs are wary of interdisciplinary research efforts and centers they want their faculty spending time and effort in their departments. Although interdisciplinary research is clearly possible, it takes place against resistance in most universities. The difficulties of peer review, communication, and publication are typical of new interdisciplinary fields, and they are sometimes successfully overcome. Among the ways this has been accomplished are the provision of resources tar- geted to the field during its developmental period and the orientation of that development around a few applied problems for which there are preexisting communities of scholars. The developing scientific interest in the human dimen- sions of global change, and within that field the growing attention to and research support for work on land-use and land-cover change, provides a context that can bring together many of the researchers who are combining social science and remote sensing. Training Future Scholars As noted earlier, most current researchers who work on projects that com
20 HNKING REMOTE SENSING AND SOCKS SCIENCE: THE NEED AND THE CHANGES bine social science and remote sensing do not have solid skills in both; rather, they are expert in one and collaborate with experts in the other. One model of training would replicate this pattern and add special training of students to be effective interdisciplinary collaborators. A drawback to this approach is that collaboration would increase the expense of research. Another is that collabora- tions would often be across institutions, creating an issue of distance that would have to be addressed. A third drawback is that graduate students and junior faculty members who were trained and employed in a disciplinary field would have to worry about whether their interdisciplinary research would hurt their chances of getting a job or tenure. An alternative training model would train young researchers in both a social science and remote sensing. This strategy would address some of the drawbacks of the first model, but has drawbacks of its own. First, it would increase the length of graduate training, which some argue is already too long. Second, such training might have to occur across departments, and most universities are not structured to do such training. Third, young scholars with interdisciplinary train- ing might have difficulty finding employment, especially in universities orga- nized along disciplinary lines. A third model involves gradual expansion of the community through inter- disciplinary research activities that may provide both students and established scholars with what is essentially on-thejob training in remote sensing or social scientific fields that are new to them. This model is an extension of a process that is already occurring in some research institutions. Its chief advantage is that training in the context of ongoing research is likely to be highly effective. How- ever, the process is likely to be slow at first, and it may train people idiosyncrati- cally in narrow segments of a field that are related to a specific research topic. At this juncture we would not want to recommend one model over another. We anticipate that the models will vary across universities, and depend on the strengths and existing institutional arrangements within each. We would suggest, however, that the time has come for funders, both federal and private, to train the upcoming generation of scholars who could bridge the social science and remote sensing fields. Providing Necessary Data Linking remote sensing with social science presents special challenges for data systems. A straightforward but significant problem is to provide georefer- encing for social data so as to link them to remotely sensed data, which are normally geocoded. Preexisting social statistics such as those collected by gov- ernment agencies are typically coded at highly aggregated levels, such as politi- cal units. They can be geocoded to some geographic point within the unit, but cannot be disaggregated below the lowest level at which they were made avail- able. 9 If they are geocoded for general use, it is important to select an appropri
RONALD R. RINDFUSS AND PAUL C. STERN 21 ate geographic point so that researchers can move the point as their scientific purposes require, and to store the geocoded data in an easily accessible place, such as the NASA-supported Socioeconomic Data and Applications Center (SEDAC). New social data could in principle be coded to the location of the individual, firm, farm, or other unit from which they are collected, but as already noted, concerns about privacy and confidentiality often prevent this, and even when it can be done, questions remain about how best to geocode actors in highly interdependent world markets. Again, the geocoded data should be stored, with appropriate documentation, in an easily accessible location. It would be useful to have an organized discussion within the research community about whether it is advisable to develop standard methods of geocoding social data for storage. Researchers also face the problem of finding appropriate social data to match with remotely sensed data, or vice versa. NASA's support for the SEDAC is intended to address this problem, and an indicator of the SEDAC's success will be the extent to which researchers find it useful for locating the matching data sets they need. Another challenge is to match the level of resolution of remotely sensed data with that needed for social data. Different social science questions require differ- ent levels of resolution in space and time, and perhaps also spectrally. The different needs are illustrated by Cowen and Jensen in Chapter 8 in the context of research on urban dynamics. They are also illustrated implicitly in other chap- ters. For example, although Chapters 4 and 5 both report studies of deforestation in Amazonia, they rely on data at very different levels of spatial resolution. Each research area probably has its own diverse needs for resolution, depending the scientific questions being asked. This variety of needs in terms of resolution suggests that any standardized system of data storage and geocoding should be highly flexible. Data maintenance is another important challenge. Many research applica- tions require intact time series of remotely sensed data and benefit increasingly as the time series are extended. For example, improvement of the famine early warning systems described in Chapter 9 depends on continued enhancements to models that account for past data on climate variations, crop production, human response, and famine. The same is true for achieving the promise of public health early warning systems, as described in Chapter 10; for modeling population and land-use dynamics, as described in Chapters 4 and 6; and undoubtedly for many other scientific purposes as well. Thus, maintenance of old data sets is a matter of continuing importance. The cost of remotely sensed data is a matter of considerable concern among researchers. A major issue for the research community is the increasing cost of data maintenance. The sheer volume of remotely sensed data is increasing rap- idly as more platforms are launched and as they provide increasingly finer reso- lution. The costs of transforming the raw data into useful forms, of cataloguing them, of storing and maintaining them, and of making them available increase
22 HNKING REMOTE SENSING AND SOCKS SCIENCE: THE NEED AND THE CHANGES more rapidly than the volume of data because data archives must continually maintain the old data as well as the new. The volume of data increases as new data come in, and new technology often brings the additional problem of translat- ing between different forms of data storage. The cost issue is multiplied further as data from military satellites are declassified and become available. In many cases, military and intelligence organizations have no use for the old data and therefore no incentive to make them useful for social science, regardless of their potential value. Yet the older material from military sources may be especially important for social science because it extends time series further backward with levels of spatial resolution that are not available from any other source for his- toric data. Thus, the cost of data systems is growing in importance just as the data are becoming increasingly useful for social science. Another cost issue concerns the cost of data to potential users. Here, the situation is highly fluid. U.S. government policy has so far kept remotely sensed images in government data systems relatively inexpensive to scientists. How- ever, cost depends on the platform and on whether any government agency has ordered a particular scene. In addition, budgetary pressures and tendencies to- ward commercialization of data systems may alter the situation. We believe it is important for science that the government maintain its policy of keeping remotely sensed data inexpensive or make sufficient research funding available so that scientists have access to the necessary data. PLAN OF THE VOLUME This volume is intended to be useful to researchers and research sponsors who are considering what they can do to foster or participate in collaborations between remote sensing and social science. It is divided into two main sections. The first, consisting of Chapters 1 through 3, provides conceptual and historical background. The second, consisting of Chapters 4 through 10, offers case ex- amples that illustrate the uses and potential applications of remote sensing for social scientific purposes. Each of these case examples describes a research area in which the effort to link remote sensing and social science shows promise for advancing knowledge. The cases also indicate what is involved, both intellectu- ally and in practical terms, in achieving that promise. The most intensive cover- age is given to research on land-use change in rural areas of developing countries (Chapters 4 through 7) because this is currently the area of the most intensive research activity involving collaboration between social scientists and remote sensing specialists. Chapters 8 through 10 present applications to urban land-use issues, famine early warning, and public health that illustrate some promising frontiers for social scientific use of remotely sensed data. The volume does not include other actual or potential social science applications of remote sensing. For example, remote sensing is used in mapping the impacts of and recovery from natural disasters and can be used in research on disaster response. It may
RONALD R. RINDFUSS AND PAUL C. STERN 23 also be possible to link remotely sensed data on atmospheric trace gas concentra- tions to ground-based data on industrial activity in order to improve models that link human activities to their environmental consequences. In Chapter 2, Morain provides a historical perspective by examining past relationships between remote sensing and social science, focusing especially on the long history of the Landsat program. Although there has recently been a proliferation of remote sensing platforms that have great potential usefulness for social science, Landsat and AVHRR have until now been the main space-based platforms used in social science research. The chapter, though perhaps fragmen- tary from a remote sensing perspective, provides a good account of the history of the sources of space-based data most commonly combined with social science data. A good source for more detail on the full range of remote sensing technol- ogy is Ryerson (1996~. In Chapter 3, Geoghegan and colleagues discuss the major issues that emerge from the most extensive current effort to link remote sensing and social science- the Land-Use/Cover Change (LUCC) research program, sponsored by the Inter- national Geosphere-Biosphere Programme and the International Human Dimen- sions Programme on Global Environmental Change. Among these issues are those of spatially explicit modeling in the social sciences, analyses that make links across spatial scales and levels of analysis, and the problem of developing effective concepts and analytical methods for simultaneously analyzing changes in time and in space. Chapters 4 and 5 describe research projects in Amazonia that use remote sensing of land cover to examine such issues as the effects of human population dynamics on deforestation and the effects of deforestation on land-use change. The two studies, though focused on the same geographical region, differ greatly in the levels of analysis at which they examine data and in the variables they select for study. Chapter 6 examines population dynamics and land-use and land- cover change at the village level in one district in Thailand. Of particular interest is the project' s effort to use both social and remotely sensed data in time series to illuminate mutual causation between population dynamics and land-use change. Chapter 7 examines land-use change in the Peten region of Guatemala. As in other areas where deforestation is progressing, remote sensing is used to track the process and its relationships to social driving forces such as road construction and land development or preservation policies. The chapter, which focuses on an important region for archaeological research, shows how remote observation has been used to identify sites not previously discovered from the ground. Chapters 8, 9, and 10 present some promising frontier areas for linking remotely sensed data and social science. These areas are promising because remote sensing has already demonstrated its relevance to socially important phe- nomena; greater integration of social science concepts is likely to yield further practical and scientific advances. In Chapter 8, Cowen and Jensen identify a variety of remotely sensed attributes that could be used for analysis of urban
24 HNKING REMOTE SENSING AND SOCKS SCIENCE: THE NEED AND THE CHANGES dynamics in the United States. They specify the degree of spatial and temporal resolution required to measure change in these attributes and, in an important contribution, compare these data requirements with the capabilities of existing remote sensing platforms. The comparison suggests some areas in which exist- ing remotely sensed data could be used more extensively in social science re- search and others in which such uses are likely to become possible as soon as higher-resolution remote data become available. For example, remote platforms might be able to measure an underinvestigated and potentially important set of social variables represented by the physical characteristics of neighborhoods. Arguably, the behavior of significant outsiders, such as social service providers and mortgage lenders, is shaped more by stereotyping based on a neighborhood's observable physical characteristics than by the actual attributes of its inhabitants. Chapter 8 also provides an illustration of the use of remotely sensed data to model and predict the course of urban development. Hutchinson (Chapter 9) discusses some applications of remote sensing and social science to famine early warning systems in Africa. Remote data are increasingly useful for monitoring and forecasting crop production the supply side of famine and also valuable for measuring roads and other infrastructure important to food distribution. However, famine also depends on the economic demand for food and the economic and political institutions that allocate food or restrict its availability factors much more easily measured by standard social scientific methods. Thus, famine early warning can best be accomplished by collaborative efforts. The same can probably be said for basic understanding of food systems and their implications for the nutritional status of populations. It has often been argued, for example, that famine is more often a result of inad- equate food distribution due to war, economic inequality, or the practices of repressive governments than of supply shortfalls. Nevertheless, famine and nu- tritional deficits are almost certainly the result of an interaction of factors: often, both supply shortages and interferences with distribution are necessary condi- tions. The study of how food production interacts with various political and economic forces is likely to be advanced by combining remotely sensed and ground-based data within the same analytical schemes. In Chapter 10, Epstein reports on uses of remote sensing to monitor and anticipate the emergence of infectious disease outbreaks. Until now, this work has involved contributions from remote sensing experts, ecologists, epidemiolo- gists, and public health specialists without explicit social science involvement. However, there are opportunities for the involvement of researchers in such fields as disease prevention, health promotion, and risk communication, as well as for interdisciplinary research linking the ecolo~v of disease organisms and vectors with human ecology. C7 C7 ~C7 Two appendices are intended as resources for social scientists who are rela- tively unfamiliar with remotely sensed data. Appendix A provides a guide to numerous major sources of remotely sensed data. Because the data sets change
RONALD R. RINDFUSS AND PAUL C. STERN 25 so rapidly, this appendix will be updated periodically on the World Wide Web. Appendix B offers a glossary of technical terms used both in remote sensing and in social science with which experts in either field might need to be familiar when venturing into the other. ACKNOWLEDGMENT We thank John Estes and B.L. Turner II for their very helpful comments on earlier drafts. NOTES 1 Definitions of this and other technical terms can be found in Appendix B. 2 Availability of satellite data changes rapidly. Appendix A identifies a number of current information sources and a Web site that will provide updated information. At the time of this writing, we are told that several improvements in resolution are just over the horizon. The French Systeme Pour ['Observation de la Terre (SPOT) satellite, scheduled for launch in 1998, is to include a 20-meter resolution mid-infrared band in the 1.55-1.75 microns range, 10-meter resolution in the red and green visible bands and the near infrared, and 2.5-meter resolution in the panchromatic band. Updated information on this satellite may be obtained from WWW.spot.com. A private company, EarthWatch Inc., launched the EarlyBird satellite in December 1997 with 3-meter resolution and plans to launch the QuickBird satellite in 1998 with 1-meter resolution, both panchromatic. Updated information on these satellites may be obtained at WWW.digitalglobe.com. India plans to launch a series of satellites with 1-meter resolution, IKONOS 1 and 2, in late 1997 and 1998. In addition, declassified high-resolution data from military and intelligence satellites from the 1960s and early 1970s are beginning to be made available. 3 Partly for economy of exposition and partly to make some of our points more emphatically, we refer to social science and remote sensing as fields. One reviewer questioned whether remote sensing is a "field," a point that raises the question of whether social science is a "field." The latter question is the easier of the two: social science is actually a collection of fields. The status of remote sensing is more ambiguous. As the reviewer correctly noted, remote sensing is in one sense only a technology or a tool. Courses in remote sensing are taught in a variety of departments, including geography, geology, landscape architecture, oceanography, and forestry. However, remote sensing also has many of the characteristics of a field. There is a specialized language that those in the remote sensing area know and use, and outsiders do not understand. There are scientific journals devoted to remote sensing topics, and there are annual meetings for those who speak the remote sensing language. Thus, despite the room for disagreement, we refer to both remote sensing and social science as "fields." 4 With few exceptions (e.g., Chapter 8), the researchers represented in this volume have used satellite data rather than aerial photographs. Among the studies we know that use remotely sensed data and social science data together, the vast majority use satellite data rather than aerial photo- graphs. There are probably a number of reasons for this. While the situation varies from country to country and region to region, in general it is more difficult to find the needed images in the form of aerial photographs than as satellite images. Satellites, because they orbit the earth, collect data for much of the earth's surface, while aerial photographs cover much smaller sections of the earth. Further, the various organizations that have emerged to sell and distribute satellite data tend to have regional or global coverage. On the other hand, those that distribute or sell aerial photographs tend to operate at the national level, or even finer scales. In addition, for many areas of the world, the search costs to determine the availability of aerial photographs are formidable. Satellite data cover
26 LINKING REMOTE SENSING AND SOCIAL SCIENCE: THE NEED AND THE CHALLENGES more of the spectral range, sometimes including the near and thermal infrared, whereas aerial photo- graphs are typically panchromatic (black and white). Finally, satellite data typically come in digital format allowing for easier incorporation into a geographic information system (GIS), but aerial photographs are typically in analog format, thus requiring additional work before they can be incor- porated into a GIS. 5 Roads illustrate some of the limits of remote sensing for measuring visible phenomena. Most sensors "see" the highest level of cover between the satellite and the earth, although there are excep- tions, such as radar data. Thus on a cloudy day, most sensors will see the clouds. If a road is tree lined and well shaded by the trees, on a clear day the typical satellite platform will provide an image of the trees and not the road. 6 The specifics, of course, depend on the constellation of techniques being used to locate all households. 7 Most satellite platforms do not capture the poles. However, given the limited human popula- tion at the poles, we can consider the data from most satellite platforms to be global from a social science perspective. Even though the data from many satellite platforms are essentially global, the uses of the data need not be global. Indeed, the examples in this volume are much more localized. 8 Aerial photographs are not subject to this pixel constraint. Instead, properties of the photog- raphy determine the minimum mapping unit, which can function like a pixel in that it determines what one can "see" in an aerial photograph. 9 Market research firms typically compile social data at much lower levels of aggregation than governments do in the United States, at the level of the postal zip code or even the zip-plus-four, which typically corresponds to a geographic area that encompasses the residences of a few dozen households. Some elements of these privately held data sets, such as data on consumer expenditures, are parallel to data collected by governments but available to clients at finer resolution. These data have not to our knowledge been used much in social science research. REFERENCES Bureau of the Census 1993 1990 Census of Population. Social and Economic Characteristics: United States. Bu reau of the Census, U.S. Department of Commerce. Washington, D.C.: U.S. Govern ment Printing Office. Carts, N. 1947 How to Read Aerial Photographs for Census Work. Washington, D.C.: U.S. Govern ment Printing Office. Colwell, R.R. 1996 Global climate and infectious disease: The cholera paradigm. Science 274:2025-2031. Estes, J.E., J.R. Jensen, and D.S. Simonett 1980 Impacts of Remote Sensing on U.S. Geography. Remote Sensing of Environment 10:43- 80. Faust, K., B. Entwisle, R.R. Rindfuss, S.J. Walsh, and Y. Sawangdee 1997 Spatial Arrangement of Social and Economic Networks among Villages in Nang Rong, Thailand. Paper presented at the annual meeting of the Sunbelt Social Network Confer- ence, San Diego, Calif. Hosenball, S.N. 1990 International and U.S. domestic law governing remote sensing. Pp. 125-140 in Earth Observation Systems: Legal Considerations for the '90s. Bethesda, Md. and Chicago, Ill.: American Society for Photogrammetry and Remote Sensing and American Bar Asso- ciation.
RONALD R. RINDFUSS AND PAUL C. STERN 27 Lucas, R.M., M. Honzak, G.M. Foody, P.J. Curran, and C. Corves 1993 Characterizing tropical secondary forests using multi-temporal Landsat sensor imagery. International Journal of Remote Sensing 14:3061-3067. Moran, E., E. Brondizio, P. Mausel, and Y. Wu 1994 Integrating Amazonian vegetation, land use, and satellite data. BioScience 44(5):329- 338. National Research Council 1997 Rediscovering Geography: New Relevance for Science and Society. Rediscovering Ge- ography Committee. Washington, D.C.: National Academy Press. Rindfuss, R.R., S.J. Walsh, and B. Entwisle 1996 Land Use, Competition, and Migration. Paper presented at the annual meeting of the Population Association of America, New Orleans, La. Ryerson, R.A., ed. 1996 Manual of Remote Sensing. 3rd edition (CD-ROM). Bethesda, Md.: American Society for Photogrammetry and Remote Sensing. Skole, D.L., W.H. Chomentowski, W.A. Salas, and A.D. Nobre 1994 Physical and human dimensions of deforestation in Amazonia. BioScience 44(5):314- 322. Turner II, B.L. in Frontiers of Exploration: Remote Sensing and Social Science Research. In Proceedings press of Pecora 13. Bethesda, Md.: American Society for Photogrammetry and Remote Sens- ing. Uhlir, P.F 1990 Applications of remote sensing information in law: An overview. Pp. 8-22 in Earth Observation Systems: Legal Considerations for the '90s. Bethesda, Md. and Chicago, Ill.: American Society for Photogrammetry and Remote Sensing and American Bar Asso- ciation.