A National Agenda for Land Use Modeling

June 6, 1997

John Landis: What is a national agenda? This is a big question, it involves issues like where we need to be, who are the clients, what are the tools, what are the data? Three broad national issues stand out:

1. what is the long term impact of urbanization on environment, habitat, and biodiversity? For this question to be answered, the models need to be valid and reliable.
2. What is the market for urban and central city renewal - what can we do there?
3. To accommodate growth we need infrastructure development. This is both a local and a national issue. How we plan this is important.
4. There are other very important issues for land use modeling, but they are more local. Tool issues:
5. Smart urban data scanner - a technology to rapidly convert urban aerial photographs to intelligent land use maps. We have a real need for basic urban land use data.
6. Cumulative impact modeling system - this would be specifically for urban areas to examine the impact on biodiversity of urbanization.

Data issues:

1. A key data need is to identify what basic land use distributions are at the census level.
2. A national inventory of urban land - what's vacant, what's mixed use....

Brian Pijanowski: I see three main approaches for a national approach to land use modeling issues:

1. Basic Research - Many basic questions remain. A critical one is, how do we deal with nonstationarity? We need to identify key directions in model evaluation, description, and development.
2. Applied - research must be more engaged with different levels of government and planning organizations. Knowledge transfer and education on modeling are very important. We've done a poor job of working with the model user community and the public, particularly in explaining the weaknesses of models.
3. Role of industry in research. Many firms are very concerned about the impact of their land use decisions. Modelers need to tap this funding source. This is particularly true for evaluating how economic decisions affect the environment.

John Morgan Grove: Of real concern here is the knowledge transfer that occurs between land use modelers and decision makers. Models don't tend to be challenged by clients. Differences between the model outcome and reality affect the entire view of modeling by the client community. Can the model simply indicate what is likely to happen, and policy makers then use that output to do "what-if" scenarios and arrive at the most preferable outcome? An integrated research approach might facilitate greater understanding:

1. Incorporate socioeconomic/demographic data in models
2. How to handle different scales (of analysis)
3. Modeling different types of change
4. Developing a set of "approved" or generally used models for particular objectives

Daniel Sui: Each agency has its own agendas and priorities, so an issue here is, "who coordinates?" A national research agenda must be customized for each of the relevant federal groups. Three major issues arise:

1. Impact of technology - how technology manifests itself upon the land. Telecommuting, to name one example, will change land use patterns. The Technological Reshaping of America report indicates the magnitude of impact this may have. Even in the pop literature, Bill Gates brings up "frictionless industries" in his book. As academics and land use modelers, we need to involve ourselves in this.
2. How to link with Census2000? The impending majority-minority in many areas, and possibly for the country as a whole, will affect socioeconomic dynamics in profound way. These in turn will produce changes in land use patterns.
3. Tying land use modeling to hazard mapping. This is an area of substantial possibility. For example, FEMA maps are very controversial; land use modelers may have a role to play in developing them. We might do a better job of developing ties to this and exploring new markets for our work, such as insurers. A possibility is becoming more proactive in identifying hazards - so much hazard work is reactive, and thus much more expensive to deal with.

Thomas Maxwell: Two major development issues:

1. Land use modeling workbench - this would combine all major model environments with a GUI on desktop computers (with links to larger processors). Libraries of methods, spatial models, links to GIS and statistics, would be incorporated. The model structure should be modular to allow facilitation of model sharing, comparison, etc. The GUI should have a three level, hierarchical interface for the model user, the model developer, and the code writer. Links to GIS and other modeling environments would keep it flexible.
2. Linking urban land use "statistical" models with ecological land cover process-based models. The process models can indicate important properties like viability, land cover health, L/C change, and not just at the developed/nondeveloped dichotomy of L/U models. These combined models can incorporate both human-induced change as well as succession and natural development, the latter of which are important to species models.

Brad Parks: 5 topics for a national agenda:

1. There are a shortage of regional applications and modeling approaches. This would seem like a good scale to bring together work, as well as providing a way to interlink across space.
2. Hazard modeling, especially mitigation work, seems promising. With a tight link to ecology, our models could reduce the impact of hazards, and this would be of interest to the federal agencies.
3. Rapid assessment of LU/LC would be helpful, given the data climate and the decision-making process. An agreed-upon assortment of models and data would facilitate this.
4. How do we work together? A working group - loosely affiliated but cohesive, perhaps with a joint Web-based workspace, would be a way to facilitate this.
5. "Summarizing the State of the Science" of LU modeling might be a useful step now. This is an important time for modeling - hardly the beginning, but a good moment to develop a summary.

Reactions to the Panel:

• In terms of data issues, linking modeling to monitoring is important
• the LU, economics-based models are not the same as LC, process-based models. They are different conceptually.
• ...But they need to be brought together. They ("human" & "natural" spatial patterns) affect one another.
• Model assumptions for LU/LC models are critical. Models can use L/U as a proxy for biophysical landcover variables.
• One might use an economic model (like we've seen here) to predict urban land use. Is there a way to compare such a derived LU map with a land cover map - is there a 1:1 match?
• No... Areas may have the same cover but different uses. [A common example used throughout the discussion was two areas with a forested land cover. One, however, was devoted to residential land use, while the other was park or reserve land.] Different land uses for identical cover types can result in different biophysical variables. Runoff may differ between residential and reserve land uses, even if the land cover is the same (forest).
• Data management and sampling strategies are particularly important to LU mapping. It takes fieldwork, as opposed to remote sensing. It's been done, but the datasets are hard to obtain, fragmented, and generally protected by the collecting agency.
• Classification of remotely sensed land cover has been employed to try to obtain LU. The modified use approach seems to work (for example, classifying an area as both forest and residential), but it has not been widely adopted.
• One important question is, can we identify larger numbers of classes, particularly urban classes?
• The multicoded approach, which uses a nested classification system resulting in more and finer classes as resolution increases, wasn't easy to use, but did allow for more urban classes. It can be used for coarser resolution work as well, since the classes aggregated. MRLC (metropolitan regional?) used several urban classes - it is a land use map as well as LC.
• In urban areas, the high-resolution data is land use, which can be used to generate cover, while in rural areas, the higher resolution is land cover.
• The USGS is moving away from a single classification approach. Flexibility is the concept now, since uses for any given product range so widely. The information for classification comes both from remote sensing and from other sources.
• What about the science agenda? The panel has focused mainly on policy. If the models are not good from a science perspective, they won't work. We need good science. There are critical questions regarding model validation. Aggregation and scale issues raise very important science issues that we need to address.
• Models start as 99% basic science research, and then they become progressively more applied. But for now, the state of the art is very much on the basic science side.
• Models often are only usable by their authors! They aren't extendible, they're not documented well, they're too complex for someone else to pick up.
• We should look at what models have been more generally successful. Some, like EPIC, have made this jump.
• The urban modeling tradition is statistical - using real-world data for calibration is the key to making the model run. Validation is conceptually difficult here, since the models are locally driven with no external global characteristics. The model is designed to model pattern in the data.