Modeling Land Cover Change

Predictions of future land cover are important for a number of conservation and restoration goals, including targeting areas for restoration, assessing the impacts of possible restoration and mitigation scenarios, and determining the vulnerabilities of various resource lands to future land conversion.  Because the conversion of natural resource lands to developed land cover poses a significant threat to Bay, we have focused efforts on simulating and predicting urban and suburban land use change.

When simulating and forecasting spatial patterns of urban development, it is a challenge to capture both the rates and locations of change.  Microeconomic models offer perhaps the best option for process-based modeling, but require highly detailed parcel-level spatial economic data in order to model the economic aspects of the development decision.  Because of these considerable data requirements, economic models currently are not applicable over large areas.  Cellular automaton (CA) models are pattern-based, mechanistic models, but offer some insight into the constraints (e.g topography) and "drivers" (e.g. road building, location of amenities) of the development process.

Our long term goals entail an integration of economic and CA modeling approaches to simulate growth patterns across the entire Bay watershed.  Our modeling activities to date consist of exploring and testing various non-economic models, including a CA-based model, and comparing and contrasting these approaches with a microeconomic model to assess the potential for integrated modeling.

A Comparison of Approaches to Model Land Use Change

Regional efforts to restore the water quality of the Chesapeake Bay incorporate land use planning goals to control the conversion of natural resource lands to impervious surfaces.  Land use modeling and scenario development play a key role in the development of management plans, but creating a predictive modeling system for the entire Chesapeake Bay watershed presents a challenge.  As part of our efforts to help develop a Bay-wide model of urban land use change, we have worked with our collaborators to rigorously test different models, evaluating them in terms of their ability to capture land use change patterns and processes.

The supply-demand-allocation approach, such as the Western Futures model, is conceptually simple, easy to implement, and requires the least amount of data, but is the most unsophisticated in terms of how land use change processes are modeled.  CA modeling, such as SLEUTH, has limited data requirements and is computationally intensive, yet provides some insights into growth processes through the calibration of process-related growth rules.  The view of development processes gained by the CA approach is superficial, however, when compared to the econometric approach, which attempts to understand, describe and simulate the economic behavior of individual land owners.  The amount of data required to develop econometric models of land use change is considerable and limits the applicability of this approach to large areas.

Below, we provide a comparison of three different modeling approaches, supply-demand-allocation (SDA), cellular automata (CA), and economic (EC), in terms of their units of observation, nature of approaches, analytical methods, development drivers and constraints, land use change processes, source of growth pressure information, and data requirements.

Unit of observation

• SDA: arbitrary areal unit (e.g. Census block)
• CA: cell in landscape
• EC: privately owned parcel of land

Nature of approach

• SDA: build out and spill over
• CA: pattern-based
• EC: process-based

Analytical method

• SDA: supply/demand/allocation model that allocates new housing units to areal units, such as Census blocks; spread of development occurs as "spill over" as the density within an areal unit reaches that of its neighbors.
• CA: cellular automaton model that simulates cell changes using development rules, which are calibrated using observed, historic changes.
• EC: discrete choice or hazard model analysis to test hypotheses and calibrate parameter estimates for forecasting.

Development drivers and constraints

• SDA: existing patterns of housing unit density, user-defined developable land per areal unit.
• CA: generation of growth by calibrated rules, state of current land cover/land use, physical features of the landscape, user-defined areas that are protected from development
• EC: value of land in undeveloped use, value of land in developed uses and conversion costs; all are functions of current land cover/land use, physical and locational features, public goods provisions, and relevant regulations

Nature of land use change processes

• SDA: new housing units are allocated to an areal unit until that unit's density equals that of its neighbors; growth then spreads to neighbors.
• CA: stochastic processes regulated by conceptually simple transition rules.
• EC: stochastic model of behavior of land owners, who choose optimal timing of development and optimal density of development to maximize their profits.

Source of growth pressure information

• SDA: external population forecasts, existing patterns of housing unit density
• CA: historic rates and/or patterns of development; calibrated growth rules
• EC: model of housing starts as a function of regional economic projections

Data requirements

• SDA: housing unit data from U.S. Census, population forecasts
• CA: urban extent for at least 4 points in time, road networks for 2 points in time, excluded layers for calibration and predictive scenarios
• EC: parcel level data including locations of parcels, GIS data on physical features, regulations, public goods, land cover

We would like to acknowledge the collaboration of our colleagues Dr. Nancy Bockstael at the University of Maryland and Peter Claggett at the Chesapeake Bay Program for their work with the economic modeling (Bockstael) and the Western Futures model (Claggett).

©Woods Hole Research Center, 2007