People are constantly changing the land surface through construction, agriculture, energy production, and other activities. Changes both in how land is used by people (land use) and in the vegetation, rock, buildings, and other physical material that cover the Earth’s surface (land cover) can be described and future land change can be projected using land-change models (LCMs). LCMs are a key means for understanding how humans are reshaping the Earth’s surface in the past and present, for forecasting future landscape conditions, and for developing policies to manage our use of resources and the environment at scales ranging from an individual parcel of land in a city to vast expanses of forests around the world.
The U.S. Landsat satellites have provided an invaluable 40-year record of global land cover change, providing input to LCMs, yielding new scientific insights, and informing policy on issues ranging from agriculture to regional planning and disaster relief. A recent explosion in the number and types of new land observations and monitoring data, model approaches, and computational infrastructure has ushered in a new generation of LCMs that are capable of new applications associated with human-environment systems in increasing detail. A wide variety of modeling approaches has been developed, each with different strengths, weaknesses, and applications. However, with increasing recognition of the role of human action in affecting change in the Earth system, greater demand for evaluations and forecasts of these impacts, and a greater variety of data sources available, it is timely to evaluate these approaches and their relative value for particular types of applications.
At the request of the U.S. Geological Survey and the National Aeronautics and Space Administration, the National Research Council established
Statement of Task
A National Research Council committee will review the present status of spatially explicit land-change modeling approaches and describe future data and research needs so that model outputs can better assist the science, policy, and decision-support communities. Future needs for higher resolution and more accurate projections will require improved coupling of land-change models to climate, ecology, biogeochemistry, biogeophysical, and socioeconomic models; improved data inputs; improved validation of land-change models; and improved estimates of uncertainty associated with model outputs. The study will provide guidance on the verification strategies and data, and research requirements needed to enhance the next generation of models. In particular, the study committee will:
1. Assess the analytical capabilities and science and/or policy applications of existing and emerging modeling approaches.
2. Describe the theoretical and empirical basis and the major technical, research, and data development challenges associated with each modeling approach.
3. Describe opportunities for improved integration of observation strategies (including ground-based survey, satellite, and remote sensing data) with land-change modeling to improve land-change model outputs to better fulfill scientific and decision making requirements.
a committee to describe various LCM approaches, suggest guidance for their appropriate application, and describe ways to improve the integration of observation strategies into the models (see Box S.1 for the complete committee charge). To carry out its charge, the committee gathered input from stakeholders in the LCM community at committee meetings, a workshop, and through an online questionnaire. Based on this input, a review of the literature, and their own experience, the committee examined the primary modeling approaches and their most suitable applications and identified several key ways to improve LCMs for decision makers and scientists. This report provides a summary and evaluation of several modeling approaches, and their theoretical and empirical underpinnings, relative to complex land-change dynamics and processes, and identifies several opportunities for further advancing the science, data, and cyberinfrastructure involved in the LCM enterprise. Because of the numerous models available, the committee focused on describing the categories of approaches used along with selected examples, rather than providing a review of specific models. Additionally, because all modeling approaches have relative strengths and weaknesses, the report compares these relative to different purposes.
LAND CHANGE MODELING APPROACHES
A wide variety of LCMs has been developed to examine land change processes and to make land use and land cover projections. The committee grouped these individual models into six categories of modeling approaches:
1. Machine-Learning and Statistical approach uses observations of past land-cover or land-use changes to calibrate parametric or non-parametric relationships between those changes and spatially and temporally specific predictors,
2. Cellular approach integrates maps of suitability for land cover or land use with neighborhood effects and information about the amounts of change expected to project future changes,
3. Sector-Based Economic approach uses partial and general equilibrium structural models to represent supply and demand for land by economic sectors within regions based on overall economic activity and trade,
4. Spatially Disaggregate Economic approach estimates structural or reduced form econometric models to identify the causal relationships influencing the spatial equilibrium in land systems, and
5. Agent-Based approach simulates the decisions and actions of heterogeneous land-change actors that interact with each other and the land surface to make changes in the land system.
6. Hybrid approaches encompass applications that combine different approaches into a single model or modeling framework.
The first five approaches are arranged roughly in order from least to most focused on process. The approaches that rely on data about land-change patterns, including Machine Learning and Statistical, and Cellular, tend to use land-cover information from satellite imagery, and relationships based on observed changes in the past. These approaches are useful for projecting observed land-cover changes over short periods into the future, but often have limited ability to evaluate conditions not observed in the past. The more process-based approaches, such as Sector-Based Economic, Spatially Disaggregate Economic, and Agent-Based, make greater use of social science information about land-change processes. These latter approaches provide more realistic representations of the processes of change that can be used to evaluate a wider range of alternative futures, but they are more challenging to calibrate and validate and may provide only qualitative information about possible future land-change outcomes.
The best modeling approach to use depends on the application. The relative advantages of the approaches for particular purposes can be used in various policy and decision-making contexts and the modeling approaches tend to serve different roles within the context of the four-stage policy cycle: (1) problem identification, (2) intervention design, (3) decision and implementation, or (4) evaluation. Machine Learning and Statistical and Cellular Modeling approaches
are most suitable for problem identification because, though they lack the richer structural detail about process needed to evaluate the effects of changes in policy structure, they are easy to implement and can provide valuable descriptions and projections of patterns and trends. Agent-Based and Structural Economic approaches are useful for intervention design because they provide a means for exploring interactions in the land system and for assessing the possible effects of policies or decisions ex ante. Once policies or decisions have been implemented, the ex post effects of these implementations can be evaluated using reduced-form econometric models that compare observable outcomes either before and after the intervention or in an intervention area and a comparable location. Understanding the underlying structures, assumptions, and data requirements of different modeling approaches is critical for understanding their applicability for various scientific and decision-making purposes.
IMPROVING LAND CHANGE MODELS
New observations, improvements in modeling capability and computer infrastructure, and advances in understanding the theoretical and social context of land change have created opportunities to improve LCMs to support research and decision making on current and future land change. Opportunities are grouped into five categories: (1) advances in LCMs themselves, (2) advances in land observation strategies, (3) advances in cyberinfrastructure, (4) advances in other infrastructure, and (5) developing and using best practices in model evaluation. Within each opportunity category are two to four specific initiatives selected by the committee that represent important near-term (three to six years) approaches to realizing the potential for LCMs to better serve an integrated Earth system science enterprise, our understanding of sustainability in human interactions with the environment, and decision making about land-related management and policy.
Opportunities for Advances in LCMs
Advancement of process-based models
While data- (or pattern-) based models have succeeded in using land-cover data products and contributing to land-change science and applications, process-based models of land change are not as mature. Better process-based models are necessary for understanding interactions and feedbacks between people’s actions and land change and for simulating policy scenarios to evaluate the impacts of a potential policy change on land use. Further developing these approaches to make the best use of available data will advance the goal of evaluating past efforts and possible future implementations of new policies and management strategies to address sustainability challenges.
Cross-scale integration of LCMs
Because land-change processes occur at multiple scales, LCMs need to link patterns and processes across multiple scales. These kinds of links require that models account for connections between distant locations of consumption and production of land-based commodities, and their network interactions, and employ new analytical methods that link models of global, regional, and local processes of land-based decision making. Better understanding of how to include representations of heterogeneous actors linked through social networks is also needed to better represent both the top-down and bottom-up causes of land change.
Integration with other Earth system models
Better dynamic coupling of LCMs with a variety of Earth system models would improve the ability to understand and project the direct and indirect effects of land management decisions and policies on the tradeoffs among various ecosystem services (e.g., food and fiber production, maintenance of biodiversity, and carbon storage). Land cover change model results have typically been used as input to other environmental models. However, coupling LCMs and environmental models would enable feedbacks between environmental and land-change dynamics to be represented and investigated, which is important for long-term forecasting.
Bridging LCM with optimization and design-based approaches
Most LCMs seek to explain and predict changes in land use and land cover using either a process- (structural) or pattern-based approach. In contrast, decisions about policy require the ability to determine, given a choice among a set of possible policies or designs, which policies will generate landscape patterns that are both plausible and acceptable for society. An important challenge is to further integrate LCMs with optimization, which is extremely computationally intensive, and design-based approaches, which require the engagement of human designers and landscape architects, in order to integrate considerations of what could be with what should be.
Opportunities in Land Observation Strategies
The second set of opportunities makes use of the flood of new data to inform the development of the next generation of land change models.
Improved capture and processing of remotely sensed data
A variety of developments in Earth observations have the potential to spur advances in land change modeling. Data collected at finer resolutions, coupled with object-based image analysis tools, offer the opportunity to develop models
that better represent diverse features in the built and natural environments. Data available at finer temporal resolutions and over longer time periods, including the free Landsat archive and historical aerial photo records, present the opportunity to better understand the dynamics and non-stationarity of land-change processes and incorporate that understanding into LCMs. Data on the three-dimensional structure of the landscape from LiDAR and other active sensors permit the development of models that can represent quantitative differences in attributes of land cover (like biomass) and land use (building density). Creative uses of satellite measurements like the nighttime lights products to estimate human settlement densities, energy use, and economic activity provide opportunities to develop spatially, temporally, and thematically richer inputs to LCMs. Hyperspectral sensors, like those on an array of smaller satellites, permit more detailed information about canopy composition that might be useful in parameterizing models that represent land-management (e.g., fertilization and irrigation) behaviors. Maximizing the ability to capture, interpret, and manage these kinds of data and incorporate them into new LCMs represents a significant opportunity for advancing the ability to use observational data to inform new modeled processes and projections.
Integration of heterogeneous data sources
Some land change decisions require information not typically included in LCMs—including land function, land-use density, land tenure, land management, and land value—or information at a variety of spatial and temporal resolutions. Integrating these data with socioeconomic and biogeophysical data would facilitate coupling of LCMs and other types of models such as those of climate change, ecosystem services and biodiversity, energy use, and urbanization.
Data on land-change actors
Land change is the cumulative result of the decisions and interactions of a variety of actors—households, firms, landowners, policymakers at local, regional and global levels. Micro-data on actors are collected by the Bureau of the Census, the Department of Agriculture, and other agencies. Better integration of data on these actors and their beliefs, preferences, and behaviors with Earth Observation data is critical for improving the ability of LCMs to project future land change and to evaluate the consequences of alternative policies.
Making systematic land-use observations
Many observations of natural and human systems must be measured from ground-based systems, which are commonly divided among multiple agencies and geographies. Possible programs like a national land observatory or national survey of land resources could be developed to collect spatially referenced data with linked records on land patches, land parcels, and land users. Such a program would improve the ability of the LCM community to learn more about land change processes, test hypotheses, and improve predictive ability.
Opportunities in Cyberinfrastructure
A number of the opportunities noted above have the potential to find solutions through contemporary advances in cyberinfrastructure.
Crowd sourcing and distributed data mining
The ability to collect and analyze large amounts of data on individual behaviors, much of which is referenced in time and space, has grown tremendously over the past decade. Crowd sourcing and distributed data mining are two primary examples of this kind of development. Combining these data collection approaches with LCMs has the potential to extend the reach of LCM results to a variety of users and could also lead to better construction, calibration, and validation of structural or process-based models. However, privacy and proprietary concerns will have to be resolved.
Cyberinfrastructure is increasingly able to meet the computational demands of some of the modeling approaches outlined above. Advances in computing power are increasingly based on deployment of multiple processing cores and increasing numbers of processors. Taking advantage of this enhanced computing power requires that models be written to take advantage of parallel processing, i.e., partitioning computational tasks among multiple processors running simultaneously. Greater volumes of distributed data storage provide opportunities to incorporate data over larger areas and at finer resolutions into the next generation of models.
Opportunities for Infrastructure to Support Land Change Modeling
Progress in land-change modeling is partially impeded by the continued reinvention of modeling environments, frameworks, and platforms by various research groups. Opportunities to improve the research infrastructure and help to overcome this barrier are summarized below.
Model and software infrastructure
Developing a consistent infrastructure for documenting and sharing models and software would help avoid duplication of effort among various constituents in the LCM community. The challenge for the community is to assemble the existing infrastructure and enhance it to serve two purposes: (1) advancement of the fundamental understanding and representation of land-change processes and (2) integration of a wide range of biophysical and socioeconomic models for evaluating the impacts of land change.
A data infrastructure would provide access to a common set of data resources that are necessary for running and validating models of land change. Infrastructure developments that aim to support compilation, curation, and comparison of the heterogeneous data sources for input to land change models would advance this kind of access directly.
Community modeling and governance
A consistent and widely adopted community modeling and governance infrastructure is important to support developments in LCM. Such an infrastructure would provide mechanisms for making decisions and advancing modeling capabilities within a broad community and toward specific, achievable goals and capabilities. In particular, it would provide a framework for reaching community agreement on specific goals and endpoints to move modeling and data capabilities forward.
There are a variety of practices that can enhance land-change modeling to make it more scientifically rigorous and useful in application. Some of these practices are established but not always followed, while others require more research to test and establish.
Sensitivity analysis is an established procedure whereby the investigator examines the variation in model output due to specific amounts of variation in model input, parameter values, or structure.
Pattern validation requires matching the choice of a metric that compares model output to data with the purpose of the modeling exercise for the particular application; how this is best done requires additional research.
Structural validation, or validating model processes, remains a challenging task in part because the underlying processes that give rise to observed land-use patterns are themselves not fully observable. Continued research on how to validate the maintained assumptions that are necessary in order to even specify a model would benefit model validation and projections.
Multiple communities of science and practice in critical areas associated with environmental sustainability, including food, water, energy, climate, health, and urbanization, are adopting land-change models (LCMs) to help with understanding and improving human-environment interactions at multiple scales. While LCMs have already contributed in all of these areas, an opportunity exists to consolidate the understanding of land system interactions, refine and improve the
best available modeling approaches, and make significant progress towards new analytical and predictive capabilities. The time is ripe to envision, plan for, and invest in the next generation of land-change models for an increasingly interdisciplinary scientific enterprise that takes advantage of the best available knowledge, data and computing resources.
If appropriately planned and executed, the next generation of models can be increasingly process based, link processes in social and natural systems from the parcel scale to regional and global scales and make use of better methods for process validation, in order to enhance both their predictive skill and their utility for policy analyses. New LCMs can also be routinely used, appropriately and with greater confidence, for a wider range of scientific and policy purposes, supporting better understanding of land systems, the effects of economic and social processes on their dynamics, and their effects on important environmental and social outcomes. Taking advantage of a wider range of Earth observation data types to enhance their spatial and temporal detail and the categories of information they represent, future LCMs can integrate these with data on the human attitudes, preferences, and behaviors related to land change, both from traditional and a growing number of novel sources for social data. Highly interconnected data systems, well-documented model and software code, and a well functioning community of land-change modelers can support the scientific enterprise to advance these goals.
Near-term intellectual and resource investments (three to six years) in the science of and infrastructure to support advancements in LCMs could help achieve these goals. This report outlines a number of specific areas that are ripe for advancement. Such investments have the potential to move forward our understanding of, ability to predict, and tools for analyzing policy related to key environmental sustainability challenges.