Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 51
~2
"Socializing the Pixel" and "Pixelizing the
Social" in Land-Use and Land-Cover Change
Jacqueline Geoghegan, Lowell Pritchard, Jr.,
Yelena Ogneva-Himmelberger, Rinku Roy Chow~hury,
Steven Sanderson, and B.~. Turner I!
Remote sensing both data and image processing and analysis through geo-
graphic information systems (GIS) are increasingly affecting the research agendas
on global environmental change, as evidenced by various reports of the Intergov-
ernmental Panel on Climate Change (IPCC) and the International Geosphere-
Biosphere Programme (IGBP), as well as a number of initiatives by agencies and
organizations that fund research on global change.) The impacts of remote sens-
ing and GIS to date have been greatest within the environmental and policy
arenas because space-based and other imagery is used primarily to determine the
physical attributes of the biosphere and the earth's surface, such as forest cover or
size of housing information that is needed in spatially explicit form by various
stakeholders and decision makers. The majority of the social sciences have been
slow to incorporate remote sensing and GIS as basic elements of research and
reluctant to respond to global-change science. The reasons are many and com-
plex, and cannot be addressed within the scope of this chapter (see B.L. Turner,
1991, 1997a). It is sufficient to note here that the core questions of the social
sciences are seen as difficult (even impossible) to address through these imaging
techniques,2 and the understanding that might be gained in those areas from
spatially explicit approaches has not been fully demonstrated or appreciated.
There are now a number of opportunities to pursue some of the core social
science research issues more closely through remote sensing and GIS. Examples
are issues of equity/equality, gender, demography, institutions, democratization,
Under development, and decision making as they relate to resource use and
environmental change. One such opportunity is represented by the core research
project on Land-Use/Cover Change (LUCC) of the IGBP and the International
51
OCR for page 52
52
f`SOCIAL17;1NG THE PIXEL'' AND f`PIXELI7;1NG THE SOCIAL''
Human Dimensions Programme on Global Environmental Change (IHDP) (B.L.
Turner et al., 1995~. This project (or project framework) is designed to improve
understanding of the human and biophysical forces that shape land-use/cover
change through three means of assessment: (1) ground-based studies of use-
cover dynamics, focused on the land manager; (2) space-based observation of the
land-cover consequences; and (3) integrative models of these dynamics at vari-
ous scales of analysis.
The objectives of the LUCC project include making remote sensing in gen-
eral (but especially that involving satellite imagery) more relevant to the social,
political, and economic problems and theories pertinent to land-use and land-
cover change (B.L. Turner, 1997c, in press), which we euphemistically call "so-
cializing the pixel" and "pixelizing the social." This objective involves methods
and tools, such as GIS, that are relevant to analysis of spatial imagery and
""ridded" data in general. Attempts to achieve this objective must invariably
address issues of scalar dynamics interpreting, merging, and analyzing the data
and analysis across spatial, temporal, and hierarchical scales. These methods and
tools and the associated scalar dynamics are the long-standing subjects of study
and an extensive literature (e.g., Ehleringer and Field, 1993; Foody and Curran,
1994; Fox et al., 1995; Michener, 1994; Quattrochi and Goodchild, 1997;
Rosswall et al., 1988; M. Turner, 1990; M. Turner et al., 1995; Woodcock and
Strahler,1987~.3 It is not our purpose to review this work here. It is sufficient to
note that the majority of this work leads up to and has significant implications for the
notion of socializing and pixelizing in land-use and land-cover change research.
Most of it, however, stops short of the work that is the focus of this chapter.
To date, work on the human dimensions of global change has focused largely
on indirect linkages between information embedded within spatial imagery and
the core themes of the social sciences. Such work is exemplified by assessments
of the proximate causes of land-use and land-cover change (e.g., slash-and-burn
cultivation, clear-cutting of timber), environmental constraints/opportunities as-
sociated with human activities (soil sustainability and zones of intensive cultiva-
tion), or assessments of infrastructure in planning (e.g., green spaces, road net-
works) (see, for example, Behrens et al., 1994; Ehrlich et al., 1997; Fischer and
Nijkamp, 1993; Martin, 1996; Massart et al., 1995; Sader, 1995; Sample, 1994;
Walsh et al., 1997~. This work, as important as it is, tends to address factors that
mediate behavioral and social actions or outcomes, and does not focus on under-
lying causes and structures. Exploration of more direct linkages would involve,
for example, extracting information about social standing, wealth, or human
health directly from the imagery (pixels), or conducting tests of a land-use theory
in which such information informs the basic tenets and operation of the underly-
ing human model.
Various efforts are now under way to advance the notions of socializing and
pixelizing, although they may not employ those terms (Entwisle et al., 1997;
Frohn et al., 1996; Guyer and Lambin, 1993; Wear et al., 1996~. The aim of this
OCR for page 53
JACQUELINE GEOGHEGAN ET AL.
53
chapter is to clarify this genre of research on land-use and land-cover change,
illustrate the work through a few select examples, identify some of the scalar
issues confronting research of this kind, and present some major lessons this
research has begun to reveal.
SOCIALIZING THE PIXEL
As suggested earlier, to socialize the pixel is to take remote sensing imagery
beyond its use in the applied sciences and toward its application in addressing the
concerns of the social sciences. Two avenues of research of this kind offer the
potential to shed light on land-use and land-cover change: mining the pixel and
modeling from the pixel.
Mining the pixel involves seeking social meaning in imagery information
and indicators relevant to such concerns as economic well-being or "criticality"
(Kasperson et al., 1995), perhaps signaling the underlying processes that give rise
to land-use and land-cover change. This meaning is often hidden deep within the
analysis of the imagery (Moran et al., 1994), and this very depth may impede
such investigation. Two examples illustrate this concept.
Work in landscape ecology indicates that landscape patterns indicative of
gross operating processes, some social in origin, can be identified with the use of
spatially explicit indicators and fractal analysis (O'Neill et al., 1988~. Might
similar patterns be found through the use of remotely sensed data? Mertens and
Lambin (1997) use Landsat Thematic Mapper (TM) imagery and GIS analysis to
identify at least six spatial patterns of land-use and land-cover change in eastern
Cameroon that are indicative of market, subsistence, policy, and urbanization
processes, although the authors do not follow the pattern-process linkage fully. A
rudimentary attempt to do so using TM imagery for Nepal proved difficult
(Millette et al., 1995~. These landscapes are a composite of rain-fed and irrigated
agriculture, village forests, household trees, small grazing plots, and landslide
features. Changes in the mosaics or composite patterns of these land covers
appeared to signal general trajectories of the socioeconomic health and environ-
mental sustainability of the production system at the village or village cluster
level, although the sample was not sufficient to determine the statistical signifi-
cance of this inference. Nevertheless, work by Moran and colleagues (1994;
Mausel et al., 1993) demonstrates the details of human action that can be found in
pixel analysis and, coupled with the inferences from the Nepal work, suggests
that further explorations of this kind are warranted.
As a second example, use of time series and principal component analyses-
long used in other research with remote sensing and GIS is gaining momentum,
and this has implications for social science research (e.g., Anyamba and Eastman,
1966~. For instance, a time series analysis of Advanced Very High Resolution
Radiometer (AVHRR) imagery of a forest reserve in Malawi provides strong
Normalized Difference Vegetation Index (NDVI)-based evidence of decreasing
OCR for page 54
54
f`SOCIAL17;1NG THE PIXEL'' AND f`PIXELI7;1NG THE SOCIAL''
forest biomass, apparently because residents adjacent to the reserve trim the trees
for fuel (Eastman and Toledano, 1996~. This activity, of course, has significant
implications for energy consumption, human health, policy enforcement, and
economic well-being. The evidence, however, is found in the seventh residual
(component) of the analysis, which is usually ignored in such analyses. Nonethe-
less, in the study of land-use and land-cover change, the reality of the component
is an empirical question because the identified human imprint can actually be
observed (or not).
These two examples are surely not exhaustive of avenues of research that
offer the potential to make remote sensing more relevant to social science inter-
ests. They illustrate, however, that a business-as-usual approach in the remote
sensing community may not be sufficient for serving these interests. More atten-
tion must be paid to the less obvious signals in the imagery, be they complex
arrays in the patterns of land cover or changes found deep within the analysis of
land-use and land-cover change.
In addition to the search for social meaning in the pixel, various approaches
can be used to model from the pixel toward the interests of the social sciences,
although they have been minimally explored. These approaches are largely
empirical and Theoretical in nature, but can be used to model land-use and land-
cover change directly from remotely sensed imagery. Markov chain modeling,
for example, offers a means of addressing land-use and land-cover change when
the data are not spatially explicit enough or are of spatially coarse resolution.
More important, however, it allows an inductive exploration of land-use and
land-cover change that may provide clues about the underlying dynamics in-
volved.
Markovian approaches assume that the immediate past is the best predictor
of the near future, under the condition of stationarity, and uses transition prob-
abilities of past states (e.g., land uses or covers or their signals in the imagery) to
estimate future states. Such approaches have been used successfully for estimat-
ing change in phenomena involving processes that conform to the stationarity
principle, where previous land uses are a proxy for stationary human behavior
(Usher, 1981~. These approaches would seem appropriate for cases involving
low levels of chronic change, as in the case of subsistence-driven cultivation in
forest regions that is associated with natural population growth or decline.
Unfortunately, most cases of land-use and land-cover change do not conform
to the stationarity principle. Rather, changes are the result of multiple actors and
structures combining in complex, synergistic ways. Moreover, critical exog-
enous forces, especially international and national policy decisions, may have
profound effects on land-use and land-cover change. These forces can be seen as
shocks to the existing land management system that fundamentally alter the
pathways and trajectories of change, thus rendering the estimated probabilities
from Markovian and other such analyses invalid. To address some of these
problems, "raw" Markov applications drawing on remotely sensed data, such as
OCR for page 55
JACQUELINE GEOGHEGAN ET AL.
55
the pixel, can be socialized by accounting for various land-use and land-cover
factors that change the estimated probabilities in question: soil quality or slope,
management rules, resource institutions, and so on. Where this understanding
can be registered to the pixel, that is, where these attributes can be recorded along
with the pixel's geographic and spectral characteristics, the estimated probabili-
ties of change from observations of the past can be altered. In this process, of
course, a point is reached at which the Theoretical empirical approach is affected
by the choice of the variables incorporated in the analysis, complete with their
theoretical implications.
An exploration of this kind is under way as part of a Southern Yucatan
Peninsular Region (SYPR) project that will produce, compare, and merge Markov
modeling approaches based on remote sensing with field- and statistics-based
models (see below) of the various land managers that are producing the signals
registered in the imagery.4 SYPR focuses on the southern portions of the Mexi-
can states of Quintana Roo and Campeche, extending along Route 186 north of
the border of Peten, Guatemala. The dominant semideciduous tropical forests of
the region came under assault after the highway was built, becoming the pathway
for various new land users: first were slash-and-burn farmers on communally
designated lands, followed by private ranchers, rice projects sponsored by non-
governmental organizations (NGOs), and, more recently, various smallholder
market operations and the Calakmul biosphere reserve. Virtually the entire pe-
riod of major land-use and land-cover change is captured by the Landsat Multi-
spectral Scanner (MSS) and TM.
A pretest exploring the socialization of a Markov approach has been under-
taken with MSS imagery alone as a precursor to the larger SYPR effort, and is
now is in its final stages. To model from the pixel as argued here, the Markovian
transition probabilities of land-use and land-cover change must be made spatially
explicit, and expanded through the insertion of biophysical and socioeconomic
factors into the probability analysis. For the pretest, three Landsat MSS images
spanning the period 1975 to 1990 were used, focusing on the southern Yucatan
peninsular region and more or less centered on Lago Silvituc, Campeche. Six
land-cover classes were derived in three land-cover maps (1975,1986, and 1990)
produced by supervised classification of MSS imagery: forest, scrub vegetation
(bush and early secondary growth), natural savanna, land in crops, bare soil/
roads, and water. These land-cover maps were overlaid in a GIS to create transi-
tion maps for the two periods 1975-1986 and 1986-1990. By calculating the
transition probability of each cell in the land-cover maps as a function of existing
land covers in the neighborhood of that cell, a spatial component was added to the
transition probabilities. Multinomial logistic regressions were used to link the
spatially explicit actual land-use transitions to biophysical, distance, and socio-
economic variables. A suite of models involving different combinations of ex-
planatory variables, some of which begin to socialize each cell (pixel), were
estimated for the transition types from the map for the first period (1975-1986),
OCR for page 56
56
f`SOCIAL17;1NG THE PIXEL'' AND f`PIXELI7;1NG THE SOCIAL''
and the estimated coefficients from each model were used to predict land-use
changes in the following time period. These predicted probabilities of transition
were then compared with actual transitions for the second period (1986-1990~.
Because the basic transition during the study period was deforestation for
cropland (with a minor amount for pasture), we focus here on two transitions:
persistence in forest cover and conversion of forest cover to cropland. Using the
spatially explicit Markov approach, the suite of models correctly predicts 94.5 to
96.5 percent of the observed persistence in forest cover for the transition period
1986-1990. This finding is not surprising given that the overwhelming land
cover was forest sufficiently distant from human activity to be protected from
conversion. The same suite of models was less accurate in predicting transitions
from forest to cropland. In this case, the raw Markov model predicted only 16.2
percent of the observed transitions (1986-1990) correctly, but up to 20.0 percent
predictive accuracy was achieved by including distance variables (to roads, vil-
lages, and markets).
These results may not seem so promising, but the conditions of the trial must
be understood. First, no assessment of the classification accuracy of the identi-
fied land covers was undertaken. Second, only three social variables were used
because of inadequate availability and consistency, and the surrogates used for
village affluence (number of cattle and trucks) may have been inappropriate.
However, most of the transition probabilities were generated for a period during
which large-scale, state-sponsored agricultural projects were undertaken along
the new highway, biasing transition probabilities toward high rates of forest
conversion to cropland. The predicted period (1986-1990), however, witnessed a
collapse of this dynamic. In short, the assumed principle of stationarity was
violated.
Given this result, one might conclude that Markovian approaches of this kind
may prove problematic and not useful. Yet examination of the ability of the
expanded Markov models to explain the transitions found during period one
(1975-1986) indicates that improvement is gained by socializing the pixel. Using
the same suite of models (increasing in complexity with the addition of biophysi-
cal, distance, and socioeconomic variables), with a measure of the fit of the
model increasing from 7.0 to 34.2 percent (pseudo Rib. This result suggests that
further exploration of the socialized Markovian approach may be useful, espe-
cially with superior data and techniques accounting for time lags and shocks to
land-conversion dynamics.
PIXELIZING THE SOCIAL
A paucity of spatially explicit data has constrained spatial modeling of hu-
man behavior and social structures, especially beyond the field of geography, and
fostered modeling approaches that abstract from the essential spatial nature of the
problem. As a result, either aggregate relationships are specified, or the spatial
OCR for page 57
JACQUELINE GEOGHEGAN ET AL.
57
components in a model are reduced to unidimensional variables, such as the
distance between economic activities in a location model, the wage differential in
a migration model, or the cost of access in a transportation model. The increased
availability of spatially explicit data, both remotely sensed and other data, and
GIS has begun to change this situation, especially with regard to broadly inter-
preted land-use and land-cover change. Advances are being made in linking on-
the-ground human actions and consequences to imagery (pixels) through models,
or modeling to the pixel. Such efforts require, for our interests here, that each
pixel (or "ridded datum) be modeled to have an empirically estimated probability
of change from one land use or land cover to another. In contrast to modeling
from the pixel as discussed earlier, such estimates are derived by linking theoreti-
cal models directly to the imagery (Lambin, 1994), as in modeling the determi-
nants of the decisions of individual land managers on the basis of utility maximi-
zation, satisficing, or other theories of human behavior.
Related to these approaches are questions of empirical estimation and em-
pirical tests of hypotheses of human behavior or social structures using remotely
sensed data. There appears to be a general belief among some social scientists
that the uses of spatial data are restricted to enhancing the measurement and
definition of explanatory variables. If this were true, the value of geographical
(spatially explicit) data for the social sciences would certainly be an empirical
issue and would be sensitive to the characteristics of the particular application.
Better data would indeed yield higher payoffs. There are, however, additional
potential gains to be realized from using spatial data. Analyzing a problem that is
essentially location based without geographically coded data is analogous to
analyzing a time-series problem without knowing the chronological order of the
observations.5 The further development of statistical techniques for estimating
spatially explicit models using remotely sensed data is essential, as articulated for
spatial econometrics by Bockstael (1996~. Taking account of the spatial nature of
the problem will improve estimation and shed light on interactions and interde-
pendencies in the system that may be interesting in their own right.
Additionally, what we euphemistically refer to as mining and modeling the
pixel can be brought together in land-use and land-cover change studies. Cre-
ative explanatory variables can be constructed from remotely sensed data through
the use of GIS, as with landscape patterns or land-use mosaics. If these patterns
and mosaics (e.g., fragmentation, urbanization) significantly change land-
management options, one could hypothesize a relationship between changing
patterns in an area over time and explore the effect of that relationship on an
individual land owner's current management schemes (e.g., Frohn et al., 1996~.
For example, using an index of the pattern as an additional explanatory variable
in a model, the relationship between past and current patterns and current land-
use decisions can be estimated (Geoghegan et al., 1997~. However, to include
such variables in an empirical specification, a model must start with a theoretical
understanding of the human behavior of valuing different types of land uses and
OCR for page 58
58
f`SOCIAL17;1NG THE PIXEL'' AND f`PIXELI7;1NG THE SOCIAL''
the distribution of those land uses across the landscape (Geoghegan and Bockstael,
1997).
Human-induced land-use and land-cover change is currently being modeled
for the data-rich Patuxent Watershed of the Chesapeake Bay,6 revealing the spa-
tial configuration and dynamic evolution of a landscape by capturing ecological
functions, human behavior, and their interaction. The effort links remotely sensed
data on land use and land cover with a variety of spatially explicit socioeconomic
and physical data, as well as with separate ecological and economic models that
are constructed so that the outputs of each can be easily used as inputs to the
other. Each model employs a landscape perspective that captures the spatial and
temporal distributions of the services and functions of the natural system and
human-related phenomena, such as surrounding land-use patterns and population
distributions. Configuration and reconfiguration of the landscape follows from
the intertwining of these phenomena, and the Patuxent work offers the potential
for a richer model of land use and its change by accounting for spatial heteroge-
neity and linking land-use conversion to features of the landscape. The aim is to
predict the probability that a given pixel of a given description and in a given
location will remain in its current use or be converted to an alternative use. While
the conversion process is affected by inertia and other disequilibrium consider-
ations and constrained by zoning and other land-use controls, the changes in land-
use probabilities are functions of the value of each parcel in alternative uses.
Consequently, the analysis must be able to explain what factors affect land values
in alternative uses (Bockstael, 1996~.
The land within the 7,000 km2 of seven counties of the Patuxent Watershed
located in Maryland ranges from suburbs of Washington, D.C. to rural and agri-
cultural areas of southern Maryland. The conversion of agricultural and forested
land (open use) to residential uses constitutes 78 percent of the total land-use
change in these seven counties during the past 10 years. As a consequence, the
economic modeling effort focuses on prediction of the conversion of open land
use to residential use through a four-part process: (1) analysis of residential value
as function of a variety of spatially related economic and ecological variables that
are hypothesized to affect residential land values,7 estimated on actual transac-
tions of residential parcels; (2) use of the estimated coefficients of the explana-
tory variables from step 1 to predict values for open land of a given description
were it to be converted to residential use; (3) use of these predictions with other
explanatory variables, such as zoning, soil type, and costs of conversion, to
estimate the spatial distribution of the relative probability that any such land will
be developed; and (4) linking of these relative probabilities with a macroeconomic
model of the state of the local economy to predict annual housing starts and thereby
how much land (how many of the pixels) will change in a given year.
Many of the themes noted earlier are included in this ongoing modeling
endeavor. The model is a utility-theoretic econometric model of human behavior
affecting land-use decisions, not driven by GIS determinism. Using spatial data
OCR for page 59
JACQUELINE GEOGHEGAN ET AL.
59
leads to interesting complications, such as spatial autocorrelation, temporal dy-
namics, and spatial structural change. Therefore, applying standard econometric
techniques to either aggregate or disaggregate spatial data generates nonspherical
disturbances, misspecification, and measurement error. New estimation tech-
niques in spatial econometrics have been developed to take some of these issues
into account in the Patuxent modeling effort (e.g., Bell and Bockstael, 1997~.
Whether the information gained by using spatial econometric techniques vastly
improves the estimation is still an empirical issue. The initial spatial econometric
modeling work with the Patuxent model demonstrates, however, the potential
improvements in explaining and predicting land values (Geoghegan and
Bockstael, 1995~. Further improvements and refinements of both theoretical and
applied econometric modeling techniques for use with the Patuxent model are
presently under way.
Another theme introduced above is not just linking the pixel to people, but
trying to use the remotely sensed data more creatively. For example, to better
capture the spatial externalities that often characterize land use and also influence
land values, indices based on the diversity and fragmentation of the surrounding
landscape around each pixel have been included in the Patuxent land-value model
to further explain residential land values. The intuition behind including these
variables is that increasing land-use and landscape diversity may adversely affect
aesthetics, but may also have convenience value signifying the proximity of
important work, shopping, recreation, and institutional destinations. Which ef-
fect dominates is an empirical question. Fragmentation might be considered
more obviously undesirable. Holding diversity constant, increasing fragmenta-
tion signals a hodgepodge of land uses. A high fragmentation index is synony-
mous with a checkered landscape, and implies the potential for large negative
locational externalities. Confusion over the sign of expected effects may be very
much tied to the issue of scale (see below). Preliminary estimation demonstrates
that these additional GIS-created variables, measured at different scales, can add
explanatory power to the Patuxent model of housing values (Geoghegan et al.,
1997~. Not only do these variables add explanatory power, but, depending on the
scale at which they are measured, the spatial index variables of land use can be
either amenities (adding to value) or disamenities (reducing value). For example,
the estimated coefficient on a small-scale measure of diversity implied that indi-
viduals valued a homogenous pattern of land use in their immediate neighbor-
hood, but the estimated coefficient of a larger-scale diversity measure implied
that a higher degree of heterogeneous land uses was valued at this higher scale.
The nature and pattern of the land uses surrounding a parcel have an influence on
the price, implying that people care very much about the patterns of landscape
around them, and supporting the belief that severe externalities exist in land use
and land-use patterns.
An illustrative application of this model (Bockstael and Bell, 1997) involved
steps 1-3 above (see Figure 3-1~. This map shows the outcome of a model
OCR for page 60
60
~ low
; medium
'~SOCIALIZING THE PIXEL', AND ``PIXELIZING THE SOCIALS
FIGURE 3-1 Predicted probability of development: Anne Arundel, Calvert7 Charles,
and Prince George's counties, Maryland.
OCR for page 61
JACQUELINE GEOGHEGAN ET AL.
61
focusing on the four southern counties of the Patuxent Watershed.8 Areas more
darkly shaded have a higher probability of development; all nonshaded areas are
land parcels that are either currently developed or precluded from development.
Some of the areas of higher predicted probability of development are closer to
Washington, D.C., as would be predicted by a traditional model of residential
choice, whereby commuting distance is an important cost component in indi-
viduals' choices of residential location. Other areas with high probability of
development are waterfront properties, as would be predicted by a simple spatial
amenity model. However, it is interesting to note from this map that even after
controlling for these two effects, there are still many areas dispersed throughout
the region that have a high probability of development.
Given the spatially disaggregate data and GIS capabilities, hypotheses were
developed to test how the distribution of land uses around a location can affect
human behavior specifically, how individuals value this distribution of land uses
and how these values can affect probabilities of land-use change. The estimated
econometric results on which Figure 3-1 is based suggest that individuals do value
different types of land-use patterns and seek to reside in locations that have a specific
distribution of local land uses, which show up as the darker areas on Figure 3-1 in
regions that are not waterfront property or relatively close to Washington, D.C.
Through this modeling exercise and earlier work on including spatial land-use indi-
ces in econometric models (Geoghegan et al., 1997), it was found that including
dissaggregate location-specific data to control for different amenities and
disamenities in land use greatly enhanced the models' explained variance. Adjust-
ing for spatial statistical issues also enhanced prediction (Bell, 1997~.
SCALAR DYNAMICS AND PATH DEPENDENCE
In the Patuxent, SYPR, and other analyses, land covers are modeled as a
function of biophysical and socioeconomic variables and their interactions. The
critical variables change in incidence and importance, however, through time and
across scales of analysis (Sanderson and Pritchard, 1993~. A primary challenge
in the remote sensing and GIS initiative to model these variables is to escape the
tendency toward a GIS-driven determinism, which tries to account for land-cover
heterogeneity through elaborate map algebra involving multiple layers of land-
scape features. The landscape is commonly taken to be in some kind of dynamic
equilibrium: driving forces, human or not, may change, creating a kind of distur-
bance, but endogenous processes restore the equilibrium. Even within this frame-
work, land use and land cover are often not a simple function of these endog-
enous processes; there can be time lags and spatial-diffusion processes, and the
processes themselves can be buffered, amplified, inverted, or otherwise trans-
formed before the resulting change in the landscape can be seen.
As an example, consider the case of the proposition that international agri-
cultural prices determine a significant share of agricultural land use. To what
OCR for page 62
62
f`SOCIAL17;1NG THE PIXEL'' AND f`PIXELI7;1NG THE SOCIAL''
extent (and through what mechanisms) do these prices pass through to the micro
level? Is it true that all hierarchical systems conduct price signals with identical
"resistance"? If not, what are the implications for using international prices as a
driving force at the unit of production (land-use manager) scale? Most important,
what makes different land-use systems more or less permeable to such macroscale
signals? How do these land-use system vulnerabilities vary through time?
As another example, many researchers have found strong links between the
external sector (international commodity prices and exchange rate dynamics, or
E1 Nino/Southern Oscillation [ENSO] phenomena) and changes in land use or
land cover, such as forest biomass or cropping schedules. These results are
difficult if not impossible to generalize across regions and nations, however, and
simple correlations tend to fall apart (for a review see B.L. Turner et al., 1995~.
Similarly, some research postulates a straightforward link between population
levels (or rates of change) and deforested area (or deforestation rates), but such
relationships typically explain no more than 50 percent of the variance in forest
cover across diverse regions (Masher, 1996) and commonly disappear in place-
specific analysis (Kasperson et al., 1995~. When these supposed macro-
mechanisms are not understood and set in context, even statistically significant
correlations may be spurious. This simple but sometimes overlooked observation
is one of the major scaling issues that confronts modeling efforts.
Even when these lags and transformations are taken into account, land-use
and land-cover systems do not always respond in predictable ways to predicted
driving forces because land cover is a function not only of socioeconomic and
biophysical variables, but also of itself. A mismatch between driving forces and
the state of land cover is not necessarily an indication that the scale of analysis is
wrong. Endogenous, contagious processes (e.g., fire, disease outbreak, techno-
logical change and diffusion, or frontier clearing) may well explain the break-
down in predictive capacity between scale levels. It may be important to consider
the extent to which a linked land-use and land-cover system exhibits its own
dynamic (even apart from driving forces). If the system is path dependent, its
current state and trajectory of change depend on its history (as in Markov ap-
proaches noted above), not on current values of driving forces alone.
A path-dependent system may exhibit several properties that must be consid-
ered in land-use and land-cover change assessments (Arthur, 1989~: varying
predictability (unpredictability is followed by high predictability as the system is
"locked ink; nonergodicity (historical events are not averaged away, and small
perturbations may significantly influence long-run development); progressive
inflexibility (the system is ultimately insensitive to perturbation); and potential
inefficiency (the outcome is not optimal for society).
Path dependency may arise from several sources, but two primary sets of
dynamics are discussed in the literature. The first, self-reinforcement, is a pro-
cess of increasing returns to scale (David, 1985) or network externalities, a form
of increasing returns to agglomeration (Arthur, 1995; Krugman, 1994~. The
OCR for page 63
JACQUELINE GEOGHEGAN ET AL.
63
geographic concentration of industries, the ascendancy of particular agricultural
technologies, and the dynamics of commodity booms and busts are all explicable
in these terms. Historical accident may explain as much as driving forces do.
The other set of processes leading to path dependency is investment rigidities-
sunk costs, infrastructure development, landesque capital such as field drains and
terraces, institutional evolution that constrain and shape future development
possibilities. The two are not mutually exclusive: self-reinforcing processes
build their own infrastructure, leading to irreversibility and inflexibility (at least
in the short run). Path dependencies and disequilibria in land-use change are also
currently being explored as part of the larger Patuxent (Irwin, 1997) and SYPR
modeling projects described above.
Recognition that historical accident is critical to land-use and land-cover
analysis does not preclude modeling, projecting, or other such scientific efforts.
As important as path dependence may be, it does not subsume all land dynamics.
The task is to identify when and where it operates and thus the spatial, temporal,
and hierarchical scales in which general dynamics operate.
Why not just choose a different scale for modeling one that spans these
historically contingent processes, or one that encompasses and subsumes loca-
tion-specific heterogeneity? This strategy has, in fact, been recommended for
modeling based on traditional hierarchy theory. Fine-scale unpredictable pro-
cesses are seen as being filtered out at higher levels of organization, so that they
appear as averages or statistical distributions, with details smoothed out or aggre-
gated. On the other hand, broad-scale processes are so slowly changing and
infrequent that they appear as constraints in a model. Such "vertical decoupling"
of time dynamics allows models to be built at a small range of scales, without
having to be concerned about cross-scale interactions. Scaling up and scaling
down thus require nesting models together and specifying the weak linkages
among them (Pattee, 1973~.
However, land-use and land-cover change is not a single hierarchy of pro-
cesses along a continuous space-time graph, so capturing the necessary effects
and dynamics is not just a question of finding the right scale of analysis. Parallel
hierarchies of geological/edaphic conditions, human land-use processes, vegeta-
tive processes, and atmospheric dynamics exist (Gallop~n, 1991~. Scaling up and
down within each hierarchy is one matter, but linking across hierarchies may
occur for processes with similar spatial scales and very different time dynamics,
which makes the neat predictions from the vertically decoupled world collapse.
Even within a hierarchy, levels are not static, and mechanisms for transmit
ting cross-scale signals are not stable. When fine-scale processes are linked
together, they may give rise to sudden changes, to radical flips between alternate
stable states in systems. For example, long periods of stability in relative prices
may synchronize previously heterogeneous agricultural systems across a land-
scape. During this period, there may be a relatively strong relationship that
explains incremental change in land-use systems as a function of small changes
OCR for page 64
64
f`SOCIAL17;1NG THE PIXEL'' AND f`PIXELI7;1NG THE SOCIAL''
in relative commodity prices. But the very process of synchronization may create
vulnerability to a previously unimportant variable lightning strikes, an agricul-
tural pest, or rainfall variability which may have existed previously, but now
has a much broader-scale effect than in the heterogeneous system. Synchroniza-
tion leads to a progressive loss of resilience, defined by the size of the perturba-
tion a system can tolerate and still recover (Holling, 1986~.9
Such surprises occur over multiple scale ranges, so predicting when the
models will break down is as necessary as predicting change at particular scales.
By exploring the interactions between slow variables (e.g., the gradual synchro-
nization of agricultural systems in a market) and fast variables (e.g., pest out-
breaks, fires), understanding can be directed toward the limits to predictability
and the generation of surprise (Sanderson and Holling, 1996~.
CONCLUSION
The evolution of global environmental change to global sustainability (B.L.
Turner, 1997a) has enlarged the human dimensions of the research agenda, in-
creasingly necessitating cooperation and collaboration among the natural, social,
and remote sensing/GIS sciences. The LUCC project and initiatives within the
project that involve socializing the pixel and pixelizing the social offer the poten-
tial to achieve such integration. They do so because they do not regard the social
sciences as an appendage to the natural and remote sensing/GIS sciences, but
hold the promise of providing information and understanding that speak to the
core issue of social science understanding.
While largely in its infancy, work of this kind provides some rudimentary
lessons that warrant special attention for the various collaborative research initia-
tives under way:
.
Indicators of social or human-environment conditions in remotely sensed
data, especially satellite imagery, are likely to be found in complex and compos-
ite patterns, requiring analytical techniques and tools to register. Since these
patterns are generated by the unfolding of many processes in place, as in the
Nepalese case, they are likely to be applicable only at the regional level.
.
Depending on the signal in question, as the Malawi case suggests, the role
of seasons and climatic flux may mask the human imprint. Analytical means
must be employed to filter through the layers of information in the signal to find
the human imprint of spatial processes.
· Probability approaches from the pixel per se can be made spatially ex-
plicit to meet the needs of the land-use and land-cover change research commu-
nity.
· Markov chain and other such probability approaches have not been suffi-
ciently explored in the sense of socializing the pixel, and the conditions under
which they may provide robust modeling outcomes (or not) remain unclear.
OCR for page 65
JACQUELINE GEOGHEGAN ET AL.
.
65
As social scientists use spatial data, their theoretical models of human
behavior need to be made more spatially explicit to aid understanding of how
humans use and value the landscape at different scales.
· As empirical specifications of models are developed using remote sensing
and GIS, development of the necessary statistical techniques for estimation is
critical.
.
Ultimately, various issues of scalar dynamics must be addressed and re-
solved if improvement in understanding and modeling of land-use and land-cover
change is to be achieved. These issues include buffering, amplification, and
inversion of transforming processes; the path dependence and historicity of
process-system relationships; and coupling of synchronous processes with paral-
lel hierarchical structures.
ACKNOWLEDGMENTS
This paper explicates various themes under development by the International
Geo sphere-Bio sphere Programme (IGBP) -International Human Dimensions
Programme on Global Environmental Change (IHDP) core project on Land-Use/
Cover Change (LUCC), although it is not a formal LUCC document. Sanderson
and Turner serve on the Science Steering Committee (SSC) LUCC, and Sanderson
and Pritchard are, respectively, the chair and science officer of the LUCC Focus
1 Research Activity. The authors thank the SSC LUCC and various colleagues at
the University of Florida and University of Maryland and in the George Perkins
Marsh Institute, Clark University, for their insights. Parts of this paper were
supported by the Carnegie Mellon University Center for Integrated Studies (NSF-
SBR 95-21914), the U.S. Man and Biosphere Program (Tropical Ecosystems
Directorate, #TEDFY94-003), the U.S. Environmental Protection Agency
(#CR8219525010 and #R825309-010), and the Maryland Agricultural Experi-
ment Station (AREC-96-62~.
NOTES
1 See, for example, IGBP Report 35/IHDP Report 7 (Turner et al., 1995) or the 1995 IPCC
report (Houghton et al., 1995).
2 Variously articulated, these questions can concern humanity's relationship with the mystical
and religious, with itself, and with nature (B.L. Turner, 1997b). With the exception of the last
question, immediate links to remote sensing are not necessarily apparent. More important, however,
understanding in the social sciences is embedded in human behavior and social structures, the es-
sence of which is not readily linked to remote sensing and, until recently, not commonly conceived
in terms of spatial relations (see National Research Council, 1997).
3 It is interesting to note that many advances made in spatial geography during the 1960s and
1970s, which gave way to interdisciplinary spatial statistics and analysis more generally, are likewise
applicable to the stated objective (National Research Council, 1997). Unfortunately, many of the
research communities now engaged in examining the human dimensions of global change are largely
OCR for page 66
66
f`SOCIAL17;1NG THE PIXEL'' AND f`PIXELI7;1NG THE SOCIAL''
unaware of lessons learned from these efforts, and the community of these spatial researchers has not
synthesized these lessons in ways that make them tractable for other communities.
4 SYPR is funded by the National Aeronautics and Space Administration's Land-Cover and
Land-Use Change initiative (NASA-NAG 56046) and involves a collaboration of the George Perkins
Marsh Institute (Clark University), Harvard Forest (Harvard University), and E1 Colegio de la
Frontera Sur Unidad Chetumal with the assistance of Carnegie Mellon University Center for Inte-
grated Studies (NSF-SBR95-21914).
5 While geographers have long championed the significance of place and spatial relations for
understanding (National Research Council, 1997), this significance is increasingly embedded within
critical social postmodern approaches to understanding, and often takes the form of "context" or
"contextualization. "
6 This project is funded by the U.S. Environmental Protection Agency (Cooperative Agree-
ment #CR82 19525010), the Maryland Agricultural Experiment Station (AREG-96-62), and the
EPA/NSF Decision Making and Valuation for Environmental Policy Research Initiative (EPA Grant
#R825309-010), and involves a collaboration between Clark University and the University of Mary-
land.
7 These values are predicated on such attributes as location, distances to features in the land-
scape, view, and surrounding landscape amenities and neighboring land uses, where the land use is
residential or another developed use. In the case of residential land values, individuals are modeled
to trade off reduced commuting distance to major employment centers for lot size, as well as neigh-
borhood and environmental amenities.
8 Anne Arundel, Prince George's, Calvert, and Charles counties.
9 Such flips or collapses in land-use and land-cover systems are the subject of research for the
emerging Resilience Network of the Beijer Institute for Ecological Economics, Stockholm.
REFERENCES
Anyamba, A., and J. R. Eastman
1996 Interannual variability of NDVI over Africa and its relation to E1 Nino/southern oscilla-
tion. International Journal of Remote Sensing 17:2533-2548.
Arthur, W.B.
1989 Competing technologies, increasing returns, and lock-in by historical small events. The
Economic Journal 99:116-131.
1995 Urban systems and evolution. Pp. 1-14 in Cooperation and Conflict in General Evolu-
tionary Processes, J.L. Casti and A. Karlqvist, eds. New York: John Wiley.
Behrens, C.A., M. G. Baksh, and M. Mothes
1994 A regional analysis of Bari land use intensification and its impact on landscape heteroge-
neity. Human Ecology 22: 279-316.
Bell, K.P.
1997 A Spatial Analysis of the Transportation-Land Use Linkage. Ph.D. dissertation. Depart-
ment of Economics, University of Maryland.
Bell, K.P., and N. E. Bockstael
1997 Applying the Generalized Method of Moments Approach to Spatial Problems Involving
Micro-Level Data. Department of Agricultural and Resource Economics Working Paper
#97-03, University of Maryland.
Bockstael, N. and K. Bell
1997 Land use patterns and water quality: The effect of differential land management controls.
In International Water and Resource Economics Consortium, Conflict and Cooperation
on Trans-Boundary Water Resources, Richard Just and Sinaia Netanyahu, eds. Kluwer
Publishing.
OCR for page 67
JACQUELINE GEOGHEGAN ET AL.
Bockstael, N.E.
67
1996 Modeling economics and ecology: The importance of a spatial perspective. American
Journal of Agricultural Economics 78: 1168- 1180.
David, P.A.
1985 Clio and the economics of QWERTY. American Economic Review 75:332-337.
Eastman, J.R., and J. Toledano
1996 Forest monitoring in Malawi using long time series vegetation index data. Earth Obser-
vation Magazine. 5(9):28-31.
Ehleringer, J.R., and C.B. Field, eds.
1993 Scaling Physiological Processes: Leaf to Globe. San Diego, Calif.: Academic Press.
Ehrlich, D., E.F. Lambin, and J-P. Malingreau
1997 Biomass burning and broad-scale land-cover changes in western Africa. Remote Sensing
of Environment 61:201-209.
Entwisle, B., R. Rindfuss, S. Walsh, T. Evans, and S. Curran
1997 Geographic information systems, spatial network analysis and contraceptive choice. De-
mography 34: 171 - 187.
Fischer, M.M. and P. Nijkamp, eds.
1993 Geographic Information Systems, Spatial Modelling, and Policy Evaluation. Berlin, Ger-
many: Springer-Verlag.
Foody, G.M., and P.J. Curran, eds.
1994 Environmental Remote Sensing from Regional to Global Scales. New York: John Wiley &
Sons.
Fox, J., J. Jrummel, S. Yarnasarn, M. Ekasingh, and N. Podger
1995 Land use and landscape dynamics in northern Thailand: Assessing change in three up-
land watersheds. Ambio 24:328-334.
Frohn, R.C., K.C. McGwire, V.H. Dale, and J. E. Estes
1996 Using satellite remote sensing analysis to evaluate a socio-economic and ecological model
of deforestion in Rondonia, Brazil. International Journal of Remote Sensing 17:3233
3255.
GallopIn, G.C.
1991 Human dimensions of global change: Linking the global and local processes. Interna-
tional Social Sciences Journal 130:707-718.
Geoghegan, J. and N. Bockstael
1995 Economic Analysis of Spatially Disaggregated Data: Explaining Land Values in a Re-
gional Landscape. Paper presented to the Association of Environmental and Resource
Economists at the Allied Social Sciences Association meeting, Washington, D.C., Janu-
ary 1995.
1997 The Value of Open Spaces for Residential Land Prices and Land Use Change. Working
paper, Department of Economics, Clark University.
Geoghegan, J., L. Wainger, and N. Bockstael
1997 Spatial landscape indices in a hedonic framework: an ecological economics analysis us-
ing GIS. Ecological Economics 23:251 -264.
Guyer, J., and E. Lambin
1993 Land use in the urban hinterland: Ethnology and remote sensing in the study of African
intensification. American Anthropologist 95:839-859.
Holling, C.S.
1986 The resilience of terrestrial ecosystems: local surprise and global change. Pp. 292-317 in
Sustainable Development of the Biosphere,W.C. Clark and R.E. Munn, eds. Cambridge,
England: Cambridge University Press.
OCR for page 68
68
f`SOCIAL17;1NG THE PIXEL'' AND f`PIXELI7;1NG THE SOCIAL''
Houghton, J.T., L.G. Meira Filho, B.A. Callender, N. Harris, A. Kattenberg, and K. Maskell, eds.
1995 Climate Change 1995: The Science of Climate Change. Cambridge, England: Cam
bridge University Press.
Irwin, E.
1997 Local Interactions and Aggregate Effects: The Role of Spatial Externalities in the Evolu
tion of Land Use Pattern. Dissertation proposal, Department of Agricultural and Re
source Economics, University of Maryland.
Kasperson, J.X., R.E. Kasperson, and B.L. Turner II, eds.
1995 Regions at Risk: Comparisons of Threatened Environments. Tokyo: United Nations
University.
Krugman, P.R.
1994 Complex landscapes in economic geography. American Economic Review 84:412-416.
Lambin, E.
1994 Modelling deforestation processes: A review. Trees Series Bill. Luxembourg: Euro-
pean Commission DGXIII.
Martin, D.
1996 Geographic Information Systems: Socioeconomic Applications, 2nd ed. New York:
Routledge.
Massart, M., M. Petillon, and E. Wolff
1995 The impact of an agricultural development project on a tropical forest environment: The
case of Shaba (Zaire). Photogrammetric Engineering and Remote Sensing 61:1153-1158.
Mather, A.
1996 The Human Drivers of Land-Cover Change: The Case of Forests. Paper presented at the
Open IGBP/BACH-LUCC Joint Inter-Core Projects Symposium on Interactions between
the Hydrological Cycle and Land Use/Cover. Kyoto, Japan, Nov. 4-7.
Mausel, P., Y. Wu, Y. Li, E. Moran, and E. Brondizio
1993 Sectral identification and successful stages following deforestation in the Amazon.
Geocarta International 8: 1-11.
Mertens, B., and E. F. Lambin
1997 Spatial modelling of deforestation in southern Cameroon. Applied Geography 17(2): 143-
162.
Michener, W.K., ed.
1994 Environmental Information Management and Analysis: Ecosystem to Global Scales. Lon-
don, England: Taylor & Francis.
Millette, T.L., A.R. Tula&ar, R.E. Kasperson, and B.L. Turner II
1995 The use and limits of remote sensing for analyzing environmental and social change in
the Himalayan middle mountains of Nepal. Global Environmental Change 5:367-380.
Moran, E., E. Brondizio, P. Mausell, and Y. Wu
1994 Integrating Amazonia, land use and satellite data. BioScience 44:329-338.
National Research Council
1997 Rediscovering Geography: New Relevance for Science and Society. Rediscovering Ge-
ography Committee. Washington, D.C.: National Academy Press.
O'Neill, R.V., J.R. Krummel, R.H. Gardner, G. Sugihara, B. Jackson, D.L. DeAngelis, B.T. Milne,
M.G. Turner, B. Zygmunt, S.W. Christensen, V.H. Dale, and R.L. Graham
1988 Indices of landscape pattern: Landscape Ecology 1:153-162.
Pattee, H.H.
1973 The physical basis and origin of hierarchical control. Pp. 73-108 in Hierarchy Theory:
The Challenge of Complex Systems, H.H. Pattee, ed. New York: George Braziller.
Quattrochi, D.A. and M.F. Goodchild, eds.
1997 Scale in Remote Sensing and GIS. New York: Lewis Publishers.
OCR for page 69
JACQUELINE GEOGHEGAN ET AL.
69
Rosswall, T., G. Woodmansee, and P.G. Risser, eds.
1988 Scale and Global Change. New York: John Wiley and Sons.
Sader, S.A.
1995 Spatial characteristics of forest clearing and vegetation regrowth as detected by Landsat
Thematic Mapper imagery. Photogrammetric Engineering & Remote Sensing 61:1145-
1151.
Sample, A., ed.
1994 Remote Sensing and GIS in Ecosystem Management. Washington, D.C.: Island Press.
Sanderson, S.E., and C.S. Holling
1996 The dynamics of (dis)harmony in human and ecological systems. In Rights to Nature:
Ecological, Economic, Cultural and Political Principles of Institutions for the Environ-
ment, S. Hanna, C. Folke, Karl-G. Mater, and A. Jansson, eds. Washington, D.C.: Island
Press.
Sanderson, S.E., and L. Pritchard, Jr.
1993 The Human Dynamics of Land-Use and Land-Cover Change in Comparative Perspective.
Paper prepared for the IGBP/HDP Core Project Planning Committee Meeting of the
Global Land-Use and Cover Change Project (LUCC), New York City, July 29-August 1,
1993.
Turner II, B.L.
1991 Thoughts on linking the physical and human sciences in the study of global environmen-
tal change. Research and Exploration 7:133-135.
1997a The sustainability principle in global agendas: implications for understanding land-use/
cover change. The Geographic Journal 163: 133 - 140.
1997b Spirals, bridges, and tunnels: Engaging human-environment perspectives in geography.
Ecumene 4:196-217.
1997c Socializing the pixel in LUCC. LUCC Newsletter No. l(Feb,):10-11.
in Frontier of Exploration: Remote Sensing and Social Science Research. Proceedings of
press Pecora 13. Bethesda, Md.: American Society for Photogrammetry and Remote Sensing.
Turner II, B.L., D. Skole, S. Sanderson, G. Fischer, L. Fresco, and R. Leemans, and others
1995 Land-use and land-cover change: science/research plan. IGBP Report No. 35/HDP Re-
port No. 7. Stockholm and Geneva: International Congress of Scientific Unions and
International Social Science Council.
Turner, M.
1990 Spatial and temporal analysis of landscape patterns. Landscape Ecology 4:21-30.
Turner, M., G.J. Arthaud, R.T. Engstrom, S.J. Hejl, J. Liu, S. Loeb, and K. McKelvey
1995 Usefulness of spatially explicit population models in land management. Ecological Appli-
cations 5: 12-16.
Usher, M.B.
1981 Modelling ecological succession, with particular reference to Markovian models.
Vegetatio 46:11- 18.
Walsh, S.J., P.H. Page, and W.M. Gesler
1997 Normative models and healthcare planning: Network-based simulations within geo-
graphic information system environment. Health Services Research 32:243-260.
Wear, D.N., M.G. Turner, and R.O. Flamm
1996 Ecosystem management in a multi-ownership setting. Ecological Applications 6:1173-
1188.
Woodcock, C.E., and A.H. Strahler
1987 The factor of scale in remote sensing. Remote Sensing of Environment 21 :311-332.
Representative terms from entire chapter:
remotely sensed
