CICEET Progress Report for the period 3/01/08 Through 8/31/08

Project Title: Integrating Geospatial and Web-Based Technologies to Improve Land Use Planning in Coastal New Hampshire
Principal Investigator(s): Fay Rubin
Project Start Date: September 1, 2007
Report compiled by: Fay Rubin

Figures

Figure 1

Figure 2

Figure 3

Tables

Table 1

Our specific project objectives for this period were as follows:

1. Complete the detailed land use data set for Rockingham County
2. Initiate development of the impervious surface data set for Rockingham County
3. Continue to refine and implement the SUBNET model
4. Prioritize GIS tools to adapt to ModelBuilder format
5. Continued development of web tools to display and query data

Objectives
1) Land use data is one of the core requirements for the modeling component of the project, and thus completing the detailed land use data set for Rockingham County constituted the first objective for the current reporting period.

2) A large-scale impervious surface (IS) data set is being generated to support the prediction of “degree of imperviousness” associated with future development scenarios. The objective for this quarter was to acquire the training necessary to initiate development of the IS data.

3) Continued refinement of the SUBNET model involved the following:
i. To complete the accessibility surface (number of jobs accessible from any location in the study area) for the SUBNET model, a regional roads dataset is required to calculate drive-time.
ii. The SUBNET model is based on the economic concept of bid-rent, a function describing the hypothetical value of land based on accessibility to markets. To test the validity of this concept, we planned to compare cost-per-acre land sales to our accessibility surface and several other geographic factors.
iii. The index of lands likely to be developed is based on an artificial neural network (ANN) classification of similarity to built lands. Numerous input datasets and Octave code are required to run the ANN.
iv. The demand component of SUBNET (the number of total workers in each census tract at each model time-step) will be based by regressing built area metrics (e.g. urban or commercial acreage) on total workers.

4) Our objective was to begin to identify the existing resource models that will most benefit the coastal community, and adapt those models to the ModelBuilder environment.

5) The continued development of web tools for displaying maps of the model output (development pressures), threats to water quality and land use information, constituted the 5th major objective during this project period.

Tasks to meet objectives
1) To complete the land use data set, staff from the Complex Systems Research Center (CSRC) planned for collaboration with Rockingham Planning Commission (RPC) staff. Commission staff performed initial data automation for the remaining towns in the county, and provided the data to CSRC for review, refinement, and incorporation into the regional data set.

2) The sole task required to meet the objective was participation in software training, as described below.

3) The tasks to meet the objective included:
i. Develop roads data and accessibility surface
ii. Complete the statistical analysis of cost-per-acre of land sales data and accessibility data and other geographic factors
iii. Create input datasets and code for ANN
iv. Acquire employment forecasts and current employment estimates.

4) Tasks for this reporting period including meeting with partners from the Town of Exeter and Rockingham Planning Commission (RPC) to evaluate needs for GIS tool development, and hosting a forum on land use planning tools for the coastal planning community.

5) Tasks to meet the objective included the construction of web tools to display and query data, and the development of three-dimensional data visualization techniques.

Progress on Tasks
1) Through continued coordination between the CSRC and RPC staff, the land use data set (based on 2005 color, 1-foot resolution orthophotography) was completed for the entirety of Rockingham County in southeast New Hampshire. This included the mapping, mosaicking and comprehensive review of the land use data set to produce a countywide, seamless data layer. Figure 1 displays the agencies that participated in the development of the regional land use data set. Figure 2 shows a detailed view of the final data set. Additionally, Table 1 displays the land uses mapped in Rockingham County along with the acreage for each.

Previous CICEET-funded activities supported the development of land use data for the same geography for three time periods ­ 1962, 1974, and 1998. Because the data sources for the historic data were small-scale, black and white photographs, the resulting land use data included only generalized use categories. A crosswalk schema has been generated to allow for trend analysis using the historic land use data in concert with the 2005 data set.

2) A member of CSRC participated in eCognition training in Morristown, NJ, in August of 2008. This formal training served as the initiation of the development of detailed impervious surface data for Rockingham County.

3) i. We have completed the development and refinement of a regional roads dataset for the SUBNET model. Based on the ESRI / TeleAtlas roads dataset, we have extracted road centerlines for 1962, 1974, and 2000 with speed limit and drive-time impedance attributes. Using the 2000 roads dataset and US Census tracts data we were able to complete the accessibility surface (number of accessible jobs by drive-time for each tract) for the study area.

ii. To demonstrate the theoretical framework of the SUBNET model (the bid-rent function) we completed a statistical review of the real-estate sales versus other geographic phenomenon. For the SUBNET I model, we geocoded real-estate data and retained only sales of undeveloped land for analysis. However, with the multi-centric approach being used in SUBNET II, a different statistical method was used to demonstrate the bid-rent concept:

Step 1, Hot-spot Analysis: As an initial coarse analysis of land values within the county, we used a Getis-Ord metric. This analysis uses an inverse-distance weighted algorithm to assess cost-per-acre over 30km (about half the width of the study area) and in relation to nearest neighbors. The model output indicates how far (standard deviations) from the mean cost-per-acre value within the specified distance the particular sale falls. General results from this analysis indicate that the cost of land is relatively higher along the coast and I95 corridor (east) and slightly less so in the towns in the southern Merrimack River valley (southwest) as can be seen in Figure 3 ­ panels a and b.

Step 2, Principal Components Analysis and Spatial Regression: To demonstrate the effect of accessibility to employment on land values, we compared land sale cost-per-acre with the number of available jobs by drive time (in 10-minute increments). To account for other factors influencing real estate costs, we also included distance-to-coast, distance-to-Great-Bay, per-pupil-spending (by town), and whether or not the sale was waterfront property.

As a first pass, we used a Principal Components Analysis (PCA) to reduce the number of input variables and to identify the most important factors. The four variables which influenced the analysis (and were uncorrelated) most were the distance-to-coast, the number of jobs available from 0-10 minutes, 40-50 minutes, and 80-90 minutes.

To determine actual statistical significance, we carried out an ordinary least squares (OLS) spatial regression. While the strength of the model is not particularly high ( R² = 0.2), we found that distance-to-coast, the number of jobs available from 0-10 minutes, and the number of jobs available from 40-50 minutes were statistically significant variables in predicting cost-per-acre. The regression coefficient for the number of jobs available from 40-50 minutes was negative indicating that as the number of jobs in this drive-time increment increases, cost-per-acre decreases. Likewise, the regression coefficient for distance-to-coast was negative, indicating that further from the coast, cost-per-acre is generally lower. However, in locales with good accessibility to employment (the number of jobs within a 0-10 minute drive) cost per acre increases (positive regression coefficient).

iii. SUBNET I ANN code has been translated to Octave (an open source software package similar to Matlab) for use in this model run. Input datasets (accessibility surface, 1962, 74, 98, and 2005 landuse, soils, wetlands, and slopes) have been assembled and processed for model input.

iv. No significant progress has been made towards the demand side of the model.

4) UNH Cooperative Extension (UNHCE) staff convened a meeting in June, 2008 with the Exeter, NH town planner and planners from the RPC. Through this discussion, it became clear that most planners in this region do not have access to ArcGIS, and instead rely largely upon regional planning commissions and/or consulting firms to provide packaged GIS products or mapping tools. The planning commissions, though, seem a likely target for ArcGIS compatible tools (ModelBuilder based). To the extent possible, these ModelBuilder based tools should also be made available through web interfaces, such as sites developed with ArcGIS Server.

Of the suite of potential mapping tools discussed at this meeting, one concept was clearly of most interest to both the Exeter and RPC planners. The proposed tool would allow the user to sketch a parcel of interest onto a map in a web interface, which would initiate a series of geoprocessing steps to a display the presence of important data layers at three scales: within the parcel, close to the parcel (~500 feet) and within the same general area of the parcel (~2500 feet). Ideally, these size ranges would be adjustable in the final tool. The types of layers processed by this tool could include important wildlife habitat, steep slopes, soils suitable (or unsuitable) for development, conservation focus areas, etc. We have begun development of this tool in ModelBuilder, and plan to deploy as much of the tool as possible through ArcGIS Server.

5) The web mapping application continues to progress. We have added and tested a feature that allows users with permissions to draw vectors over the web and store them in the database. These vectors will be useful for performing simple analysis on the SUBNET data. Additionally, testing has begun on the "administration module" that allows a subset of users to control access to data via the web, rather than through a configuration file.

Difficulties
1) We found that there were significant time expenditures while edgematching the town-based land use data to produce a regional, countywide data set. This was a result of editing difficulties realized while using ArcGIS to adjust the road features that participated in a geodatabase topology. These features comprised large sinuous polygons covering a broad geographic region and when any edge of these “shared features” (i.e. the road polygons) needed adjustment, it required minutes instead of seconds (observed with other features) to complete the edit. We estimate that this “slowness” added several days to the overall editing task. This behavior was reported to and acknowledged by ESRI although the issue was never satisfactorily resolved. We highlight this difficulty as a warning to others who may plan similar editing activities on large data sets containing regional road polygons.

2) There were no unexpected difficulties in accomplishing the scheduled tasks.

3) The difficulties associated with the four tasks included:
i. Region-wide traffic impedance attributes in a form compatible with the SUBNET model data simply are not available at this time. However, we are incorporating impedance from speed-limits and turns.
ii. None (aside from lacking traffic related impedance in drive-time calculations as noted above).
iii. There were no unanticipated difficulties associated with this objective.
iv. Unlike the first SUBNET model, REMI employment forecasts for Rockingham County are not available. We are relying on forecasts made by the Rockingham Planning Commission for their Air Quality Forecast Report (August, 2008). NH Department of Economic Security employment data are being summarized by traffic analysis zone (by RPC) and should be available in September, 2008.

4) UNHCE attempted to schedule stakeholder forums during the summer by working with the Natural Resource Outreach Coalition, but were unable to arrive at dates of common availability for a variety of reasons. The stakeholder forums will instead take place in the fall and will allow feedback on the tool described above, as well as provide insight into additional tools useful for coastal planning.

5) There have been no unexpected difficulties in accomplishing the scheduled tasks.

Data Generated to date
As reported previously, several geospatial data sets are under development. The land use data set is finalized, and the associated metadata is now being prepared for posting to the GRANIT web site (www.granit.unh.edu) for public access. As the project progresses, additional data sets that are not restricted will also be posted to the site.

Project Objectives for Next Reporting Period

Objectives
Our project objectives for the next 6 months include:
1) Develop impervious surface data for Rockingham County
2) Run and refine ANN
3) Complete demand side of model (predictions of built-area by census tract)
4) Select existing resource identification techniques to convert to ModelBuilder format
5) Release first iteration of the web site

Work Plan to Meet Objectives
Objective 1:

• Based on recent training, finalize our technical approach to the data development task.
• Collect any required field data.
• Process the imagery.
• Conduct post-processing accuracy assessment.

Objective 2:

• Adjust code to accommodate input multi-centric datasets

Objective 3:

• Acquire and process the employment estimates and predictions being distributed by Rockingham Planning Commission
• Develop regression equations to estimate built area based on current employment data
• Develop predictive model based on equations and economic forecasts

Objective 4:

• Utilize the Northeast Arc User Group meeting in late September as a forum for continued conversations with relevant organizations and a chance to identify the best GIS techniques used in the coastal areas of New England.
• Include some of these resultant ModelBuilder models in the November workshop and in future workshops offered through our training program.
• Where appropriate, investigate the use of these models through a web interface through ArcGIS Server

Objective 5:

• Continue web tool development, and release first version to targeted set of users for feedback.

Expenditures
Expenditures have been in the range anticipated during the first project period.

What Else?
In addition to the activities outlined above, UNHCE offered a workshop titled “Getting to Know ModelBuilder“ in Concord, NH, in April 2008 with materials developed in partnership with the University of Connecticut Geospatial Technology Program. The intent of the workshop was to provide expertise in the use of ModelBuilder for coastal community planners. Another ModelBuilder workshop is scheduled for November 2008. This workshop will become a staple of the UNH Cooperative Extension training program, and some of the materials will be incorporated into other training courses.

Several of the tools previously identified as useful by GIS-savvy planners have already been incorporated into the UNHCE workshop rotation. In addition to the ModelBuilder workshops previously described, we developed a Co-occurrence & Habitat Mapping workshop using the Habitat Priority Planner (NOAA Coastal Services Center) and Landscape Fragmentation Tool (University of Connecticut). We offered the workshop in May and June of 2008, both of which were full subscribed, and have it scheduled again in October and December of this year. This workshop represents some of the only, if not the first, hands-on training courses in the country on either of these tools. We are also developing a workshop in conjunction with NH GRANIT on the use of Community Viz for community planning. The first offering of this workshop is scheduled in December 2008. These training opportunities have and will provide a solid basis for the eventual development of targeted trainings for ArcGIS-savvy coastal planning organizations in 2009, especially once the appropriate currently existing resource identification techniques have been coded in the ModelBuilder format.