Probability of Development, Northeast U.S. is one of a suite of products from the Nature’s Network project (naturesnetwork.org). Nature’s Network is a collaborative effort to identify shared priorities for conservation in the Northeast, considering the value of fish and wildlife species and the natural areas they inhabit. This index represents the integrated probability of development occurring sometime between 2010 and 2030 at the 30 m cell level. It was based on models of historical patterns of urban growth in the Northeast, including the type (low intensity, medium intensity and high intensity), amount and spatial pattern of development, and incorporates the influence of factors such as geophysical conditions (e.g., slope, proximity to open water), existing secured lands, and proximity to roads and urban centers. The projected amount of new development is downscaled from county level forecasts based on a U.S. Forest Service 2010 Resources Planning Act (RPA) assessment. A complementary product, Probability of Development, 2080, Northeast U.S., estimates the probability of development over a longer time-scale.
Note: based on revisions of the sprawl model, this version was revised in July 2017 to better reflect relatively higher probabilities of development in close vicinity to roads, which is most evident in rural areas.
A number of additional datasets that augment or complement the Probability of Development product are also available in the Nature’s Network gallery: https://nalcc.databasin.org/galleries/8f4dfe780c444634a45ee4acc930a055. You may detailed refer to the detailed technical guide to Probability of Development for more information about this product and the urban growth model.
Combined with any of the other landscape conservation design (LCD) products that reveal places of high ecological value, this product can be used to indicate places of ecological value that are at risk of development and thus may warrant land protection. This product also can be used to identify places at risk of future development independent of designated core areas and any formal landscape conservation design.
You might explore this product in combination with:
Aquatic Core Networks and Terrestrial Core-connector Networks to identify intact habitat cores and important connectors that should be maintained in the face of development in order to sustain diverse and resilient ecosystems.
Core Habitats for Imperiled Species to focus on at-risk habitats which are critical for the preservation of Species of Greatest Concern.
Probability of Development, 2080, to compare to urban growth projected forward over a longer interval
The Secured Lands, Eastern U.S. layer to better understand which lands are protected from development.
Description and Derivation
The derivation of the integrated probability of development layer was complex. Please consult the detailed technical documentation for a full description of the background data used, the computation of integrated probabilities from a stochastic model, and information about the related urban growth model. The following is a summary of the five major steps of the derivation:
1) Determining historical patterns of growth
To understand how past patterns of development have occurred, historical data from NOAA (for Maine and Massachusetts) and the Chesapeake Bay Watershed Landcover Data Series were obtained for the years 1984 (Chesapeake Bay only), 1996, and 2006. The data were used to model the occurrence of six different development transition types:
undeveloped to low-intensity (20-49% impervious surface; e.g., single-family homes)
undeveloped to medium-intensity (50-79% impervious surface; e.g., small-lot single-family homes)
undeveloped to high-intensity (80-100% impervious surface; e.g., apartment complexes and commercial/industrial development)
low- to medium-intensity
low- to high-intensity
medium- to high-intensity
Separate models were developed to represent development patterns at model points representing landscapes differing along two dimensions: intensity of development and amount of open water. Predictor variables in the models account for the intensity of existing development and landscape context (e.g. intensity and distance of nearest roads, amount of open water).
Analysis of the historical data was based on dividing the landscape into “training windows,” 15km on a side, to determine the historical distribution of transition types and the total amount of historical development.
2) Application to current landscapes
Future patterns of development were projected based on the observed historical patterns. As the first step in this process, the entire Northeast was subdivided into 5km “application panes,” each of which was the center pane of a (3 x 3) “application window”, 15 km on a side. Each of these overlapping application windows was then matched to the three most similar training windows on the basis of intensity of development from the UMass integrated landcover layer, (derived in turn from the 2011 National Landcover Database and other sources), as well as geographic proximity, amount of open water, and density of roads. .
For each application window, according to how it mapped on to the dimensions of development and open water modelled above, the relative probability of each of the six development transition types was determined on a scale of 30m cells.
3) Predictions for changing land-use
Future urban acreage by county was predicted as part of an assessment for the U.S. Forest Service 2010 Resources Planning Act. The derivation of this product, the new growth forecasted for the 20 years between 2010 and 2030 was transformed into demand in units of 30m cells. Demand for each county (or census Core Based statistical Area, where relevant) was allocated to the corresponding application windows based on the average of the total amount of historical development in the three matched training windows.
4) Combining models of past and predictions for the future
The relative probability of a transition type occurring in each cell in a window was used to distribute the allocated demand of new growth throughout the window. The result was an actual probability of development for the transition occurring sometime between 2010 and 2030 at the 30 m cell level. Already existing urban land-use was intensified (i.e., transitions 4-6) in proportion to historic patterns determined from the matched training windows, and distributed according to the probability of those transition types across the cells in the window. The combining of probabilities and demand to distribute development to cells was done for each transition type in turn; thus, each cell received a separate probability of being developed through each of the six transition types. Through the application of this process in every application window, an actual probability of development was determined for each cell with reference to nine slightly different contexts corresponding to each of the overlapping windows in which the pane was situated.
5) Smoothing and integration
An additional step was used to create a smooth and continuous probability of development surface, not subject to abrupt differences along arbitrary boundaries. Cell by cell, actual probabilities of development from each of the overlapping windows were combined such that the closer to a window’s center a cell was located, the more weight the probability derived from it was given. Consequently, each cell had one weighted average probability that was part of a continuous probability of development surface for each transition type.
Finally, the probability of development by each of six transition types was integrated for each cell. More weight was given to new growth, such that the probability of undeveloped land becoming urban had more impact than the probability of an intensification of development. The final product is a single layer of the integrated probability of development by 2030, extending across the entire Northeast on the scale of 30 m cells.
Known Issues and Uncertainties
As with any project carried out across such a large area, the Probability of Development dataset is subject to limitations. The results by themselves are not a prescription for on-the-ground action; users are encouraged to verify, with field visits and site-specific knowledge, the value of any areas identified in the project. Known issues and uncertainties include the following:
Although this index is a true probability, it is best used in a relative manner to compare values from one location to another
The GIS data upon which this product was based, especially the National Land Cover Dataset (NLCD), are imperfect. Errors of both omission and commission affect the mapping of current development and in turn, models of the probability of future development. Likewise, the forecasts in the 2010 Resources Planning Act assessment, the basis of the projected demand for new growth, contains uncertainties. While the model is anticipated to generally correctly indicate where development is likely to occur, predictions at the cell level are not expected to be highly reliable.
Users are cautioned against using the data on too small an area (for example, a small parcel of land), as the data may not be sufficiently accurate at that level of resolution.
This model is built on the assumption that future patterns of development will match patterns in the past.
It is important to recognize that the integrated probability of development is highest near existing roads, largely because the urban growth model does not attempt to predict the building of new roads and the development associated with them, nor does it incorporate county or town level planning for infrastructure. Because proximity to roads is an important and dominant predictor of development at the 30- m cell level in the model, the integrated probability of development surface is heavily weighted towards existing roads. It is not specifically designed to predict where a subdivision might be developed in the future.
Click on title to download individual files attached to this item.
Original FGDC Metadata
Potential Metadata Source