HomeMy WebLinkAboutScenario Planning - Trend Scenario & Land Use Allocation Technical Report - Placeways (2016)
Trend Scenario and Land Use Allocation Technical Report 1
PROJECT DOCUMENT
County of Hawaiʻi General Plan Comprehensive Review Trend Scenario and Land Use Allocation Technical Report
Date: April 18, 2016
Authors: Amy DeBay, Ian Varley, Doug Walker
Introduction
Task S2 (Land Use Allocation) sets up a framework for estimating future development patterns
(amounts and location) based on a set of rules. Task E (Trend Scenario) uses an initial, calibrated run of
the allocation model to estimate future development patterns based on historical trends. These tasks
go hand-in-hand and are combined in this technical report.
Allocation Concepts and Approach
Future development patterns (amounts and location) are estimated using an algorithm-driven process
called allocation. Allocation models the interplay between market demand for development in certain
locations (“desirability”) and amount of development allowed according to current regulations or by
future land use patterns suggested by alternate scenarios (“capacity”). Given a pre-determined amount
of growth expected in the given time frame (here, new growth between 2015 and 2040), the allocation
process estimates where each incremental unit of new development will go, following the basic
presumption that the most desirable areas will be developed first, capacity allowing. Thus highly
desirable areas are assigned growth first, and then slightly less desirable areas get developed next, etc.,
until all the estimated growth amount has been accommodated. Numerous refinements to the basic
principle are used to produce the final estimates. For example, parcels aren’t always filled all the way to
capacity, a certain amount of controlled “randomness” is often applied to the growth pattern, etc. For
mixed use areas, both residential and non-residential growth can be assigned.
In this study, a CommunityViz tool called Allocator 5 is used. The methods combined with the
algorithms in Allocator 5 provide a well-reasoned analysis that will be helpful for this and myriad other
planning studies, but it is recognized that the results have limitations in terms of modelling precision
and confidence. The CommunityViz allocation method is sometimes described as “light-weight” to
“medium-weight,” differentiating it from the “heavy-weight” algorithms such as UrbanSim or PECAS
that are considerably more sophisticated but are more difficult and expensive to implement. In
contrast, the Allocator 5 method is easier to use and lends itself well to “what-if” type scenario
planning.
At the highest level, the allocation algorithm takes 3 inputs—growth amount, capacity, and
desirability—and generates 1 output—a pattern of future development. Our methods for each of these
are described next.
Growth Amount
As a 2015 baseline for housing unit data, Placeways used data from the County’s Real Property Tax
(RPT) office to identify the number of housing units and non-residential square feet. The procedure
began with a database file from RPT that, unfortunately, lacked metadata, so the fields were
interpreted manually. For each TMK, buildings were converted to housing units where appropriate
Trend Scenario and Land Use Allocation Technical Report 2
(buildings and dwelling units were tracked separately). Single family homes and ʻohana units were
readily identified in the RPT data. Multifamily housing required some additional steps. In cases where
there were multiple records per TMK (as with condos), the records required consolidation to identify
the total number of dwelling units per TMK. Separate analyses were performed for single family,
multifamily, and commercial properties. In RPT data, housing units that are treated as commercial
property (e.g., apartments and timeshares) were counted as residential only when the RPT data
showed them as such. Once this basic processing was complete, Placeways used the RPT online tool,
Google Street View, digital air photos and other tools to verify the number of existing units with the
goal of establishing an accurate baseline and using the RPT database to its maximum extent.
For the trend scenario, growth projections are provided by SMS, a Hawaiʻi-based research and
consulting company (see their report “General Plan Comprehensive Review Trends and Forecast
Analysis Final Report (2015)”). These projections are broken out by 13 geographic areas called “forecast
analysis zones” or FAZs (see Figure 1) and by use type (residential dwelling units and non-residential
square feet). In order to add additional land use information to the allocation, the SMS forecasts were
further broken down into four categories: single family dwelling units, multifamily dwelling units,
commercial square feet, and industrial square feet. The ratio of single family to multifamily was found
using the mean of the ratio from three dates in the recent past (2000, 2010, 2015) for which the ratio
was known. This ratio was then applied to the combined residential allocation amounts to produce the
single family/multifamily splits seen in Tables 1, 3, and 6. Similarly, growth amounts for non-residential
development were developed as a single forecast and had to be split into amounts for commercial and
industrial uses. The 2013 ACS Employment by Occupation Type data were used to identify the ratio of
industrial employment to commercial employment and to produce the splits seen in Tables 2, 3, and 7.
While this method assumes no change in the ratio of single family to multifamily units and commercial
to industrial space, it reflects the recent development patterns in the Trend Scenario and can easily be
adjusted for use in alternate scenarios.
Trend Scenario and Land Use Allocation Technical Report 3
Figure 1. Map of Forecast Analysis Zone (FAZ) Areas (source: SMS)
In contrast to the RPT-based method for establishing a baseline, the housing forecast data produced by
SMS rely on US Census and Hawaiʻi DBEDT as sources. The methods used to collect Census and
DBEDT data are quite different from RPT, resulting in differing 2015 baseline quantities. In addition,
SMS did not suggest a 2015 baseline, instead using decadal increments for forecasting. In order to
establish an SMS 2015 baseline, Placeways used the average of SMS’s 2010 and the first forecast year
of 2020. This results in a 2015 baseline difference of 11,558 housing units (RPT 2015: 75,100;
SMS/Census: 86,658). The reasons that the RPT and Census derived baseline amounts are different are
due to the sources’ two distinct methods, and no attempt was made to reconcile them.
To calculate the amount of new residential growth, Placeways used a method to find the relative
amount of net new amount of growth per FAZ. This method finds the percent change, per SMS,
between 2015 and 2040 and applies that to the RPT 2015 baseline. This forecast results in fewer net
new units (35,750) than the SMS forecast (40,160 new units), but its rates of change match SMS.
The SMS non-residential forecasts were already reconciled with the RPT data, and their forecast was
calibrated to closely match the 2015 RPT baseline square footage. Therefore, in the case of non-
residential growth, there was no need to rectify the forecast numbers as was the case with residential
growth. Non-residential square feet were rounded and translated from square feet to 1,000 square feet
Trend Scenario and Land Use Allocation Technical Report 4
for the purposes of allocation. This ensured that allocation amounts are in whole increments and not in
very small portions of square feet.
Table 1. Residential Forecast by FAZ
FAZ 2015 2040 Change 2040 Total
Single
Family
Multi-
family
Total Single
Family
Multi-
family
Single
Family
Multi-
family
Total
Hilo 14,713 1,138 15,851 2,953 833 17,666 1,971 19,636
North Hilo - Hāmākua
Coast Villages
2,822 12 2,834 721 71 3,543 83 3,626
Honokaʻa-Paʻauilo 2,399 14 2,413 871 56 3,270 70 3,340
Waimea 3,212 98 3,310 1,420 158 4,632 256 4,887
North Kohala 2,499 17 2,516 785 59 3,284 76 3,360
Kawaihae-Puakō-
Waikoloa-Waikoloa
Resorts
2,610 3,390 6,000 1,337 1,337 3,947 4,727 8,675
North Kona 11,181 5,989 17,170 4,418 2,708 15,599 8,697 24,295
South Kona Villages 3,437 73 3,510 1,129 125 4,566 198 4,765
Kaʻū 3,397 76 3,473 2,135 112 5,532 188 5,720
Keaʻau -Kurtistown 1,640 10 1,650 834 35 2,474 45 2,518
Upper Puna 4,884 0 4,884 3,373 104 8,257 104 8,361
HPP-Orchidland 6,654 0 6,654 7,431 152 14,085 152 14,237
Lower Puna 4,835 0 4,835 2,515 78 7,350 78 7,428
Total 64,283 10,817 75,100 29,922 5,828 94,205 16,645 110,850
Trend Scenario and Land Use Allocation Technical Report 5
Table 2. Non-Residential Forecast by FAZ (in 1,000 square feet)
FAZ 2015 2040 Change 2040 Total Commercial Industrial Total Commercial Industrial Commercial Industrial Total Hilo 9,187 3,762 12,949 3,979 346 13,166 4,108 17,274
North Hilo - Hāmākua
Coast Villages
347 30 377 98 8 445 38 483
Honokaʻa-Paʻauilo 438 38 476 130 7 568 45 613
Waimea 1,303 75 1,379 421 37 1,724 112 1,837
North Kohala 290 255 545 181 12 471 267 738
Kawaihae-Puakō-Waikoloa-Waikoloa
Resorts
5,406 63 5,470 1,608 85 7,014 148 7,163
North Kona 6,512 5,135 11,648 3,400 296 9,912 5,431 15,344
South Kona Villages 868 16 884 235 18 1,103 34 1,137
Kaʻū 303 0 303 110 11 413 11 424
Keaʻau-Kurtistown 760 902 1,662 454 51 1,214 953 2,167
Upper Puna 201 2 203 54 3 255 5 260
HPP-Orchidland 120 0 120 57 4 177 4 181
Lower Puna 413 0 413 114 14 527 14 541
Total 26,150 10,279 36,428 10,841 892 36,991 11,171 48,161
Capacity
Capacity values used in this report are based primarily on the results of Task S1, covered in a separate
report. The focus is on net capacity, which is gross (or total) capacity minus existing development.
Numeric capacity is assigned to every parcel on the island for residential dwelling units, and a separate
numeric capacity for non-residential square feet. The majority of visitor units, existing and projected,
are within 3 of the 13 FAZ areas. More information on visitor units and how they are addressed can be
found in the Indicator Modeling technical report (Task G).
During the initial calibration of the allocation model, it was decided to place caps on the capacity of
individual parcels that limited the amount of development that could occur on any single parcel. Caps
for both residential and non-residential development were identified by looking at the County’s
subdivision records and the 2004-2015 building permit data. This information revealed what the recent
historical maximums were for each kind of development: 600 units for residential development and
306,000 square feet for non-residential development. (The historical maximums and their place in the
model are also discussed below in the section on additional calibration factors.) The final numbers
below represent capped net capacity. Later in the project, these capacity values may be adjusted to
model different potential planning and policy decisions.
Trend Scenario and Land Use Allocation Technical Report 6
Table 3. Net Capacity (Capped) by FAZ
FAZ Name
Residential
Single Family
Capacity (DU)
Residential
Multifamily
Capacity (DU)
Commercial
Capacity
(sq ft)
Industrial Capacity
(sq ft)
Hilo 22,299 1,183 7,037,091 6,766,170
North Hilo -
Hāmākua Coast
Villages
8,721
50
1,173,273 894,155
Honokaʻa-Paʻauilo 9,691 39 579,692 133,073
Waimea 6,807 827 1,242,247 577,334
North Kohala 9,950 231 1,660,320 1,499,805
Kawaihae-Puakō-
Waikoloa-Waikoloa
Resorts
15,237
3,815
1,944,304 310,342
North Kona 21,855 5,376 9,831,956 12,956,188
South Kona Villages 21,445 190 539,714 0
Kaʻū 25,088 1,971 768,304 620,075
Keaʻau-Kurtistown 5,518 75 1,481,939 4,484,680
Upper Puna 25,390 31 135,975 0
HPP-Orchidland 9,996 0 0 0
Lower Puna 25,928 151 656,727 74,451
Total 207,925 13,939 27,051,542 28,316,273
Desirability
Desirability is a complex topic and represented a large proportion of the effort for this task.
The desirability of a given parcel for a given use (residential or commercial) is represented by a score
from 0 (least desirable) to 100 (most desirable). A parcel that is not eligible for a given use is assigned a
desirability score of -1. On a map, the pattern of desirability scores is sometimes called a “desirability
surface” because one can picture a lumpy blanket covering the island with high points in areas of high
desirability and low points in areas of low desirability.
Desirability was calculated all at once for the entire island. Given more time and resources, it would be
possible to perform separate calculations for subareas such as east/west or individual FAZs. However,
the additional effort would make little difference because growth amounts are constrained to meet
FAZ-specific totals and thus growth is allocated based on relative desirability scores within a FAZ, not
between FAZs.
Trend Scenario and Land Use Allocation Technical Report 7
The detailed procedure for creating the initial desirability surface, used to create the Trend Scenario,
follows:
1. Spatial and non-spatial factors affecting location desirability were hypothesized. These factors,
such as proximity to infrastructure or coastline, likely affected development desirability in the
past and could be quantified for all parcels on the island given available information. While a
complete list of factors tested is included at the end of this report, all hypothesized factors
could fit into the following categories:
• Distance to infrastructure, geographical features, town and commercial centers
• Parcel shape/size
• Terrain/climate characteristics
• Neighbor (10 ft) and neighborhood (1000 ft) context: the number of nearby parcels and
development proximity (the number of nearby parcels that are built)
• Current land use and build status
• Geographic location (District, FAZ, etc.)
• Property and building valuation.
Using CommunityViz, these factors were calculated for each parcel on the island and exported
to a table for use in SPSS statistical analysis software.
2. Statistical regression analysis in SPSS was used to calculate how well each factor correlates
with new development in a given time frame. Three timeframes were initially considered:
• All development since the beginning of RPT (County of Hawaiʻi Real Property Tax
Office) recordkeeping (1880s)
• All development since 1995
• All development between 1975 and 1995.
However, because the goal of this statistical regression was to capture the principal factors
influencing urban growth in the recent past, a cutoff date of 1995 was established. This year
was selected to encompass a full cycle of real estate development and to capture the “highs
and lows” in development activity. Development during this period followed this approximate
pattern:
• Trough: 1996
• Peak: 2005-2006
• Trough: 2009-2011
• Recent uptick: 2015.
The SMS CoH 2016 General Plan Final Report (Figure 1, Resident Population) illustrates this
pattern.
The statistical analysis models the relationship between dependent variables – Commercial
Square Feet (COM_SQFT) and Number of Residential Buildings (NumberOfResidences) – and
the hypothesized list of independent variables (see Appendix 1, Table 12. Factors Tested for
Historical Growth Trends). As a result, a step-wise Multiple Linear Regression model was
chosen to create coefficients associated with each of the variables in order to represent the
independent contributions of each independent variable to the prediction of the dependent
variable after controlling for all other independent variables.
Trend Scenario and Land Use Allocation Technical Report 8
The initial analysis included an overall view of development, an earlier era of development
(1975-95), and post-1995 recent development patterns for both of the dependent variables (see
Appendix 1: Comparing Post 1995 Regression Factors with the 1975-1995 Regression Factors for a
discussion of these results). The results show the top 10 variables—that is, the top 10 of the
hypothesized desirability factors—that influence each of the analyses, along with the absolute
value of each of the standardized coefficients. The coefficient values allow ranking the
variables from most to least influential. Detailed analysis information is included at the end of
this report.
The standardized regression coefficients with the 10 highest absolute beta values for the post-
1995 period were converted into CommunityViz weighting factors normalized to the scale 0 –
10, where 0 is no correlation and 10 is the highest correlation of any factor (though less than 1).
Candidate factors with lower beta coefficients, below the top ten, were ignored for the
desirability score.
A cutoff of 10 factors was chosen for a few reasons. One was to keep the most significant
factors in the mix. The top 10 account for the majority of the causal influence of all factors
tested. Additionally, there was a benefit to limiting the number of movable parts for testing.
Even with 10, it is challenging to understand the interplay of all inputs and the individual effect
on the overall score. The goal here is to incorporate both sophistication and manageable
interactive parts, and 10 is a reasonable number for that.
Some factors are negatively correlated, and some factors are inversely correlated. For
instance, distance values that correlate to growth are often inverse: nearer, smaller distance
values are more desirable and further, larger distance values are less desirable. In the table
below, negative and inverse correlations are indicated by a negative beta value score.
Understanding this, many factors below make intuitive sense. The slope factor for example
suggests that as land gets steeper (slope increases), the likelihood of development decreases.
Some results are not always intuitive, however. Statistically, for example, it is found that
parcels that are closer to old lava flows are more desirable for non-residential development than
those far away (i.e. Distance2LavaFlow1790). Some positively correlated cases also benefit
from some explanation. For example, the strongest factor for residential development is
Distance2VolcanoHazard. This is a positively correlated factor meaning that as distance
increases away from volcano hazards, the likelihood of development also increases.
Trend Scenario and Land Use Allocation Technical Report 9
Table 4. Residential Factors (Top 10) for Growth Desirability Model (Post 1995 Development)
Factor Description Beta
Value
Absolute
Beta Value
Normalized
Value
(Weighting)
Distance2VolcanoHazard Distance to high volcano risk area,
defined as areas classified as
category 1 or 2 on the risk layer
0.0808 0.0808 10
Distance2LavaFlow1250 Distance to older lava flow,
deposited between the year 1250
and the present
-0.0575 0.0575 7.1
Slope Mean slope of the parcel in percent
rise
-0.0494 0.0494 6.1
Intersections1miDensity Road intersection density: the
number of intersections on major
roads within ½ mi of the parcel
divided by area of the parcel
-0.0439 0.0439 5.4
Rainfall Annual average rainfall, in inches -0.0429 0.0429 5.3
Distance2CenterAdjusted Network distance to commercial
center. Commercial center is
defined by RPT land use
"commercial." This adjusted value
uses straight line distance for those
features not picked up by the
patchy network.
-0.0411 0.0411 5.1
RESpost1995ProximitySum1000ft Total number of dwelling units on
parcels within 1000 ft of a parcel
built after 1995
-0.0394 0.0394 4.9
RESpost1995ProximitySum10ft Total number of dwelling units on
parcels within 10 ft of a parcel built
after 1995
-0.0321 0.0321 4
ParcelPerimeter2DivArea Lot shape: perimeter squared divided by area 0.0305 0.0305 3.8
Distance2Coastline Distance to the island’s coastline -0.0301 -0.0301 3.73
Trend Scenario and Land Use Allocation Technical Report 10
Table 5. Non-Residential Factors (Top 10) for Growth Desirability Model (Post 1995 Development)
Factor Description Beta
Value
Absolute
Beta
Value
Normalized
Value
Distance2LavaFlow1790 Distance to recent lava flow,
deposited since 1790
-0.3410 0.3410 10
Intersections1miDensity Road intersection density: the
number of intersections on major
roads within ½ mi of the parcel
divided by area of the parcel
-0.2560 0.2560 7.5
Distance2Airports Distance to nearest major airport 0.2246 0.2246 6.6
Proximity1000ftParcelDensity The number of parcels within 1000ft
divided by the area of the parcel
0.1723 0.1723 5
Distance2ExistingMF1 Distance to nearest multifamily
residential development
-0.1380 0.1380 4
Distance2SewerService Distance to nearest wastewater
service line
-0.1207 0.1207 3.6
Distance2WaterService Distance to nearest potable water
service line
0.1089 0.1089 3.2
Proximity10ftParcelDensity The number of parcels within 10 ft
divided by the area of the parcel
-0.1006 0.1006 3
ParcelPerimeter2DivArea Lot shape: perimeter squared
divided by area
-0.0935 0.0935 2.7
Distance2VolcanoHazard Distance to high volcano risk area, defined as areas classified as
category 1 or 2 on the risk layer
0.0885 0.0885 2.6
3. The CommunityViz Suitability Wizard was then used to create a suitability analysis using the
parcels’ normalized factor values (such as distance to roads) as inputs. Since these values were
pre-calculated for the statistical analysis, Suitability Wizard was pointed directly at the numeric
values, improving processing performance. The wizard default is set to 5 (on a scale of 0 to 10)
for each factor weighting. After the suitability analysis is run with defaults, the weighting
assumption defaults are set for each factor according to the values calculated in the previous
step to calibrate the score to historical trends.
4. The suitability (desirability) score for each parcel is displayed using the parcels layer symbolized
by the suitability scores. The factor weightings are adjustable, so each factor that contributes
to the overall score can be given a level of priority appropriate to the goals of the scenario.
Trend Scenario and Land Use Allocation Technical Report 11
Trend Scenario Allocation
The allocation of forecasted development in the “Trend Scenario” is based on the desirability surface
representing historic development trends. Later in the project, it will be possible to develop alternate
desirability surfaces for scenario planning in which planners test planning strategies that could
encourage growth to evolve in particular ways.
Allocation is performed using Allocator 5 with the inputs above. Again, allocation distributes
predetermined growth amounts for each FAZ to parcels within that FAZ, developing the most desirable
parcels first and proceeding until all growth has been accommodated. No parcel receives more
development than it has capacity for, and many parcels receive little or no development even though
they have capacity available. For the initial run, used to develop the Trend Scenario, randomness is set
at 1 out of 10. Results for each FAZ are as follows:
Table 6. Allocated Residential Growth by FAZ
FAZ Name
Residential
Single Family
Dwelling Units
Remaining
Single Family
Capacity
Residential
Multifamily
Dwelling Units
Remaining
Multifamily
Capacity
Hilo 2,952 19,347 833 350
North Hilo - Hāmākua Coast Villages 721 8,000 50 0
Honokaʻa-Paʻauilo 871 8,820 39 0
Waimea 1,419 5,388 158 669
North Kohala 785 9,165 59 172
Kawaihae-Puakō-Waikoloa-
Waikoloa Resorts
1,338 13,899 1,338 2,477
North Kona 4,418 17,437 2,708 2,668
South Kona Villages 1,130 20,315 126 64
Kaʻū 2,135 22,953 112 1,859
Keaʻau-Kurtistown 833 4,685 35 40
Upper Puna 3,373 22,017 31 0
HPP-Orchidland 7,431 2,565 0 0
Lower Puna 2,515 23,413 78 73
Total 29,921 178,004 5,567 8,372
Trend Scenario and Land Use Allocation Technical Report 12
Table 7. Allocated Non-Residential Growth by FAZ in 1,000 square feet
FAZ Name Commercial
Allocated
Remaining
Commercial
Capacity
Industrial
Allocated
Remaining
Industrial
Capacity
Hilo 3,979 3,039 346 6,424
North Hilo - Hāmākua Coast Villages 98 1,074 8 887
Honokaʻa-Paʻauilo 130 450 7 125
Waimea 421 814 37 541
North Kohala 181 1,479 12 1,487
Kawaihae-Puakō-Waikoloa-Waikoloa Resorts 1,608 336 85 224
North Kona 3,400 6,411 296 12,656
South Kona Villages 235 303 0 0
Kaʻū 110 655 11 609
Keaʻau-Kurtistown 454 1,023 51 4,434
Upper Puna 54 80 0 0
HPP-Orchidland 0 0 0 0
Lower Puna 114 542 14 60
Total 10,784 16,206 867 27,447
Trend Calibration Factors
In order to better replicate the patterns of the previous 20 years, additional calibration factors were
developed to supplement the factors used in the statistical analysis. The objective of the Trend
Scenario is to reflect a continuation of recent patterns, and these factors help reflect the recent
development characteristics. While the statistical analysis provides an impartial and “data-driven”
perspective on the patterns of recent development, its results cannot provide a complete picture
because of a number of limitations:
• Data limitations. Data were not available on all factors and across all historic time scales. In
particular, the model lacked in-depth real estate market data, including market conditions and
consumer preferences for both residential and non-residential development. In addition,
historic building data were not available in appropriate forms.
• Geographic biases that resulted from using parcel data as the unit of analysis. For example,
large parcels tended to score very well in the statistical analysis. These features tend to show
advantages that small parcels do not: they have more capacity, they have more neighboring
parcels, and they can be closer to more desirable locations simply because of their size. A grid-
based analysis would reduce these issues, but it would introduce many other issues (e.g.,
mismatch between the grids and the available parcel data) in their place.
• Difficulty reflecting all patterns and preferences for all the island at a local scale (e.g., what
drives growth in North Kona is somewhat different from what drives growth in Lower Puna). It
was outside the scope of this project to run individual analyses for smaller areas (e.g., FAZs).
Trend Scenario and Land Use Allocation Technical Report 13
• No forecast model is perfect, and it is expected that any algorithmic results will require some
degree of adjustment to match observed results.
Because of these limitations on factors derived solely from statistical analysis, additional calibration
factors based on expert human knowledge were included in the analysis as well. The combination
resulted in a hybrid system that carefully combines both statistical modeling and expert judgement.
Each calibration factor works in a different way to help fine-tune the model by what is believed as
common knowledge by County of Hawaiʻi planners and by comparing values observed in the recent
development data to the results of the draft versions of the Trend Scenario allocation.
Some of these factors (Redevelopment Friction, Residential Subdivision Friction and Non-residential
Size Friction) help steer growth towards areas that have received consistent development pressure in
the previous 20 years. The Pipeline Projects Factor prioritizes parcels that currently have development
proposals pending. While these factors help calibrate the Trend Scenario to reflect recent patterns, it is
difficult—and not necessarily desirable—to exactly replicate rates and patterns of the recent past with
what occurs in the Trend Scenario’s modeled future. The future is unknown and many external factors
and unforeseeable conditions will affect how growth actually occurs. The goal of the Trend Scenario is
to provide a plausible and useful reference for comparing alternative scenarios to help inform policy
and the contents of the General Plan; it is not intended as a detailed forecast.
Recent Lava Factor
While a significant amount of development in the County of Hawaiʻi occurs on geologically recent lava
flow (often on lava less than a few hundred years old), lava that has been deposited very recently is a
significant obstacle to development. Lava flows that are more recent than 1990 were identified, and
desirability of parcels that overlapped post-1990 lava flows was reduced.
Pipeline Projects Factor
This factor adds a bonus to the desirability of parcels that are currently under development or that
County planning staff assume are likely to develop in the near future. The pipeline projects are divided
into two groups. Group one consists of two projects, Kamakana Villages and Kealakehe Homesteads,
which are very likely to develop or are in the process of development. The second group includes five
projects that are less advanced in the planning process but have greater than average chances of
developing. The project pipeline projects are mostly located in North Kona and primarily entail
residential development. Table 8 identifies the pipeline projects and the number of units allocated to
them.
Trend Scenario and Land Use Allocation Technical Report 14
Table 8. Pipeline projects
Project Name Approximate Location Allocated
Residential Units
Kamakana Villages Ane Keohokālole Highway, North of
Kailua Kona 600
Kealakehe Homesteads North of Kailua-Kona, Corner of
Keanalehu & Manawalea 184
UH Pālamanui UH West Hawaiʻi campus, just east of
airport 300
Pualani South of Kailua-Kona along Queen
Ka‘ahumanu highway 178
Keahuolū Queen Liliʻuokalani
Trust
Various parcels, north of Kailua-Kona
downtown 48
Redevelopment Friction
The uncalibrated allocation model orients development towards areas of greater desirability and
available capacity. Many parcels already have existing residential or non-residential structures on them
but according to the capacity analysis have some additional capacity (these parcels are considered
“underbuilt” for purposes of the model). The Redevelopment Friction factor considers redevelopment
in the sense of any new development on a parcel that has an existing residential or non-residential
structure (according to RPT). The uncalibrated model does not distinguish the nature of this new
development: whether it is a physical addition to a structure, a new structure on an undeveloped
portion of the parcel, or the wholesale redevelopment of an existing structure. Redevelopment areas
can sometimes be more difficult and costly to develop but they may be highly desirable because of
their location or other amenities. The 2004-2015 building permit data reveal that 34% of residential
development occurred on parcels that already had a residential or non-residential structure. This figure
is lower for commercial development, where 15% of growth occurred on parcels with existing
structures. The Redevelopment Friction factor applies a penalty to the desirability score for both
residential and non-residential development to parcels that already have some development.
Residential Subdivision Friction
An analysis of residential growth since 1995 revealed that over 75% of new dwelling units were built on
parcels that were subdivided to sizes of less than 5 acres. Subdivisions yielding parcels larger than 25
acres were not as common and did not contain large amounts of residential development in the recent
past. The County’s subdivision data revealed that in the last fifteen years, the largest subdivision to be
approved was 590 parcels. During initial runs of the allocation model, it was observed that the tool
tended to select larger parcels over smaller parcels, a pattern that was not consistent with the recent
development patterns. The County’s subdivisions layer depicts all of the island’s major subdivisions and
using this data as a guide, subdivided parcels were flagged and received higher desirability scores than
Trend Scenario and Land Use Allocation Technical Report 15
non-subdivided parcels. This factor actually consists of two factors, one factor that penalizes large
parcels by size and another that caps residential development at 600 units, similar to the historic
maximum.
Non-residential Footprint Friction
Similar to residential subdivision friction, recent non-residential development had certain
characteristics related to building footprint size (i.e., the square footage of the non-residential
development) that were not well captured by the regression analysis. Early runs of the model revealed
some extremely large non-residential developments that seemed uncharacteristic with the recent
development patterns. According to RPT records, the largest commercial development since 1995 was
306,000 square feet and the median size 6,510 square feet. For the Trend Scenario, non-residential
capacity was capped at 300,000 and desirability was boosted for smaller capacity parcels.
Comparing Recent Development Trends and the Trend Scenario
After calibration, the Trend Scenario matches recent development patterns fairly well, as shown in the
table below.
Table 9. Comparing recent development with the trend scenario
Recent Development* 2004-
2015 Trend Scenario
Residential Redevelopment 11% of recent growth occurred
on built parcels
4% of growth occurs on built
parcels
Commercial Redevelopment 15% of recent growth occurred
on built parcels
16% of growth occurs on built
parcels
Residential Subdivision
Development
75% of recent growth occurred
in existing subdivisions; median
parcel size is 0.7 acres
73% of recent growth occurs in
existing subdivisions; median
parcel size is 0.9 acres
Non-residential Footprint
Friction
Median development size was
6,510 square feet
Median development size is
5,000 square feet
* Recent redevelopment is based on RPT and building permit data; subdivision and parcel size factors
use RPT data.
Trend Scenario and Land Use Allocation Technical Report 16
Appendix 1. Comparing Post 1995 Regression Factors with the 1975-1995 Regression
Factors
The regression analysis conducted for all development and development post-1995 was also run for
development that occurred between 1975 and 1995. Development in this era appeared to manifest
some similar patterns (Distance to Volcano Hazards, Lava Flow, and Intersection Density) as the post
1995 development. Distance to centralized services such as water and sewer systems was more
strongly correlated in the 1975-95 development, perhaps reflecting the closer proximity of
development in that era to those services. In a more extreme case, the correlation for Distance2Airport
reverses: it is negative for 1975-95 development and positive for post 1995 development. In other
words, as distance to airports diminishes the likelihood of development increases in the 1975-95 era.
The opposite is true for development in the post 1995 era.
Valuation factors (building, land values) were poorly correlated in the 1975-95 era for residential
development. Valuation factors were more strongly correlated with non-residential square footage.
However, the valuation factors used reflect present-day conditions. In general, care should be taken
when comparing the results of the 1975-95 era beta values with the post 1995 beta values because this
analysis relies on data that represents conditions as they are now, not as conditions were during the
1975-95 phase of the island’s development. Some factors rely on data that has remained relatively
consistent through time; for example the location of the island’s coastline, slope, rainfall patterns, etc.
are very similar now to how they were 40 years ago. Other factors have changed considerably since the
1975-95 development occurred (distance to commercial development, development proximity, land
and building valuations, etc.). The 1975-95 development patterns would be better compared to the
post 1995 development by factoring in the conditions as they existed in that 1975-95 era, not conditions
as they exist in the present day. Unfortunately that level of analysis was not feasible within the scope of
the current project. Below are two tables comparing the 1975-95 and the post 1995 beta values for
residential and nonresidential development.
Table 10. Comparing post 1995 and 1975-95 non-residential beta values
Factor ComPost95Beta Com75-95Beta Difference
Distance2LavaFlow1790 -.341 .006 -0.3474
Intersections1miDensity -.256 -.100 -0.1564
Distance2Airports .225 -.145 0.3700
Proximity1000ftParcelDensity .172 -.042 0.2140
Distance2ExistingMF1 -.138 -.126 -0.0120
Distance2SewerService -.121 .161 -0.2813
Distance2WaterService .109 .013 0.0958
Proximity10ftParcelDensity -.101 .019 -0.1196
ParcelPerimeter2DivArea -.093 .077 -0.1703
Distance2VolcanoHazard .088 .090 -0.0016
Trend Scenario and Land Use Allocation Technical Report 17
Table 11. Comparing post 1995 and 1975-95 residential beta values
Factor
ResPost95Beta ResPost75-95Beta Difference
Distance2VolcanoHazard .081 .079 0.0017
Distance2LavaFlow1250 -.057 -.173 0.1151
Slope -.049 -.049 0.0000
ValueBuilding .049 .010 0.0389
Intersections1miDensity -.044 -.078 0.0338
Rainfall -.043 -.007 -0.0359
Distance24CenterAdjusted -.041 -.028 -0.0132
RESpost1995ProximitySum1000ft* -.039 #N/A -0.0419
RESpost1995ProximitySum10ft** -.032 #N/A -0.0273
ParcelPerimeter2DivArea .031 .037 -0.0067
* The 1975-95 results used the factor, RESpost7595ProximitySum1000ft instead of the post 1995
proximity sums, yielding the beta value of 0.002.
** The 1975-95 results used the factor, RESpost7595ProximitySum10ft instead of the post 1995
proximity sums, yielding the beta value of -0.005.
Table 12. Factors Tested for Historical Growth Trends
Factor Name Description Unit Source
Acres Size of the parcel in acres acres
Distance2Airports Distance to nearest major airport feet County GIS
Distance2Belt10 Distance to the island's belt road feet County GIS
Distance2Coastline Distance to coastline feet County GIS
Distance24CenterAdj
Network distance to commercial
center. Commercial center is defined
by RPT land use "commercial,” this
adjusted value used straight line
distance for those features not
picked up by the patchy network. feet RPT and county GIS
Distance2ElectricService Distance to nearest electric utility
pole feet County GIS
Distance2ExistingMF1 Distance to nearest multifamily
residential development feet RPT and county GIS
Distance2ExistingSF1 Distance to nearest single family
residential development feet RPT and county GIS
Distance2Hospitals Distance to hospitals feet County GIS
Distance2LavaFlow1250 Distance to older lava flow,
deposited since 1250 feet
USGS Hawaiʻi
Geologic Map
Distance2LavaFlow1790 Distance to recent lava flow,
deposited since 1790 feet
USGS Hawaiʻi
Geologic Map
Distance2Major10 Distance to nearest major, arterial-
type road feet County GIS
Trend Scenario and Land Use Allocation Technical Report 18
Factor Name Description Unit Source
Distance2Schools Distance to nearest public school feet County GIS
Distance2SewerService Distance to nearest wastewater
service line feet County GIS
Distance2Towns
Distance to nearest major town,
towns defined by county provided
"towns" layer feet County GIS
Distance2VolcanoHazard
Distance to high volcano risk area,
defined as areas classified as
category 1 or 2 on the risk layer feet
USGS, via county
GIS
Distance2WaterService Distance to nearest potable water
service line feet County GIS
Slope Mean slope of the parcel in percent
rise percent rise USGS DEM
ValueLand Land value in dollars per the
County's RPT records dollars RPT
Rainfall Annual average rainfall, in inches inches
University of
Hawaiʻi
District Planning district County GIS
FAZ Forecast analysis zone SMS
MaukaMakaiNum
Uses 1 or makai (outside the belt)
and 2 for mauka (inside the belt) Placeways
COM_YRBLT_YN If built 1, if not 0. RPT
RES_YRBLT_YN If built 1, if not 0. RPT
isCom If commercial, 1; if not, 0 RPT
isSF If single family residential, 1; if not, 0 RPT
isMF If multifamily residential, 1; if not, 0 RPT
COMAllYearsProximitySum10ft
Total commercial square feet on
parcels within 10ft of any parcel with
commercial square footage RPT
COMAllYearsProximitySum1000ft
Total commercial square feet on
parcels within 1000ft of any parcel
with commercial square footage RPT
COMpost1995PS1000ft
Total commercial square feet on
parcels within 1000ft of any parcel
with a structure built after 1995 RPT
COMpost1995PS10ft
Total commercial square feet on
parcels within 10ft of any parcel with
a structure built after 1995 RPT
RESAllYearsProximitySum10ft
Total number of dwelling units on
parcels within 10ft of any parcel with
a residential structure RPT
RESAllYearsProximitySum1000ft
Total number of dwelling units on
parcels within 1000ft of any parcel
with residential structure RPT
DUpost1995PS10ft
Total number of dwelling units on
parcels within 10 ft of any parcel
with a structure built after 1995 RPT
DUpost1995PS1000ft Total number of dwelling units on RPT
Trend Scenario and Land Use Allocation Technical Report 19
Factor Name Description Unit Source
parcels within 1000 ft of any parcel
with a structure built after 1995
ParcelPerimeter2DivArea
Lot shape: perimeter squared
divided by area
square
feet/feet County GIS
Intersections1/2miDensity
Road intersection density: the
number of intersections on major roads within ½ mi of the parcel
divided by area of the parcel
intersection
per acre County GIS
Proximity10ftParcelDensity
The number of parcels within 10ft
divided by the area of the parcel
Parcels per
acre County GIS
Proximity1000ftParcelDensity
The number of parcels within 1000ft
divided by the area of the parcel
Parcels per
acre County GIS
Trend Scenario and Land Use Allocation Technical Report 20
Appendix 2. Detailed Statistical Methods
Commercial All Data
The first multiple linear regression was calculated to predict COM_SQFT for all possible parcels based
on ParcelPerimeter2DivArea, Intersections1miDensity, Slope, Rainfall, Distance2Major10,
Distance2WaterService, Distance2SewerService, Distance2Schools, Distance2Coastline,
Distance2VolcanoHazard, Distance2Airports, Distance2LavaFlow1250, Distance2LavaFlow1790,
Distance2Towns, Distance2ElectricService, Distance2Hospitals, Distance24CenterAdjusted,
Distance2ExistingSF1, Distance2ExistingMF1, Distance2Belt10, COMAllYearsProximitySum10ft,
COMAllYearsProximitySum1000ft, ValueLand, Proximity10ftParcelDensity, Proximity1000ftParcelDensity
with n= 132,842. A significant regression equation was found (F(28,132814) = 161.888, p <0.001 ) with
an R2 of 0.033.
After the analysis the 12 coefficients that influence Commercial square feet are listed below.
Table 13. Twelve most influential variables for commercial square feet, all years
Variables Beta Value Coefficient
COMAllYearsProximitySum10ft .125
COMAllYearsProximitySum1000ft .056
Distance2VolcanoHazard .054
Intersections1miDensity -.051
Distance2Hospitals .037
Distance2LavaFlow1790 .032
Distance2ExistingSF1 .025
Distance2ElectricService -.023
Distance2ExistingMF1 -.020
Slope -.019
Distance2SewerService -.018
Distance2LavaFlow1250 -.018
Commercial Post 1995
The first multiple linear regression was calculated to predict COM_SQFT commercial buildings built on
or after 1995 based on ParcelPerimeter2DivArea, Intersections1miDensity, Slope, Rainfall,
Distance2Major10, Distance2WaterService, Distance2SewerService, Distance2Schools,
Distance2Coastline, Distance2VolcanoHazard, Distance2Airports, Distance2LavaFlow1250,
Distance2LavaFlow1790, Distance2Towns, Distance2ElectricService, Distance2Hospitals,
Distance24CenterAdjusted, Distance2ExistingSF1, Distance2ExistingMF1, Distance2Belt10,
COMpost1995ProximitySum1000ft, COMpost1995ProximitySum10ft, ValueLand,
Proximity10ftParcelDensity, Proximity1000ftParcelDensity with n= 451. A significant regression
equation was found (F(27,424) = 2.259, p <0.001 ) with an R2 of 0.126.
After the analysis the 12 coefficients that influence Commercial square feet after 1995 are listed below.
Trend Scenario and Land Use Allocation Technical Report 21
Table 14. Twelve most influential variables for commercial square feet, post 1995
Variables Absolute Value Coefficient
Distance2LavaFlow1790 -.341
Intersections1miDensity -.256
Distance2Airports .225
Proximity1000ftParcelDensity .172
Distance2ExistingMF1 -.138
Distance2SewerService -.121
Distance2WaterService .109
Proximity10ftParcelDensity -.101
ParcelPerimeter2DivArea -.093
Distance2VolcanoHazard .088
ValueLand .082
Distance2Towns -.067
Commercial Post 1975-95
The first multiple linear regression was calculated to predict COM_SQFT commercial buildings built on
or after 1975 and before 1995 based on ParcelPerimeter2DivArea, Intersections1miDensity, Slope,
Rainfall, Distance2Major10, Distance2WaterService, Distance2SewerService, Distance2Schools,
Distance2Coastline, Distance2VolcanoHazard, Distance2Airports, Distance2LavaFlow1250,
Distance2LavaFlow1790, Distance2Towns, Distance2ElectricService, Distance2Hospitals,
Distance24CenterAdjusted, Distance2ExistingSF1, Distance2ExistingMF1, Distance2Belt10,
COMpost1995ProximitySum1000ft, COMpost1995ProximitySum10ft, ValueLand,
Proximity10ftParcelDensity, Proximity1000ftParcelDensity with n= 687. A significant regression
equation was found (F(27,660) = 3.634, p <0.001 ) with an R2 of 0.129.
After the analysis the 12 coefficients that influence Commercial square feet between 1975-95 are listed
below.
Table 15. Twelve most influential variables for Commercial Square Feet > 1975 and < 1995
Variables Beta Value Coefficient
Distance2SewerService .161
Distance2Airports -.145
ValueLand -.144
Distance2ExistingMF1 -.126
Distance2ExistingSF1 -.126
Distance2Schools .109
Intersections1miDensity -.100
Distance2VolcanoHazard .090
ParcelPerimeter2DivArea .077
Distance24CenterAdjusted -.072
Distance2Coastline -.070
Distance2Belt10 -.059
Trend Scenario and Land Use Allocation Technical Report 22
Residential All Data
The first multiple linear regression was calculated to predict NumberOfResidences for residential
buildings built after 1995 based on ParcelPerimeter2DivArea, Intersections1miDensity, Slope, Rainfall,
Distance2Major10, Distance2WaterService, Distance2SewerService, Distance2Schools,
Distance2Coastline, Distance2VolcanoHazard, Distance2Airports, Distance2LavaFlow1250,
Distance2LavaFlow1790, Distance2Towns, Distance2ElectricService, Distance2Hospitals,
Distance24CenterAdjusted, Distance2ExistingSF1, Distance2ExistingMF1, Distance2Belt10,
RESAllYearsProximitySum10ft, RESAllYearsProximitySum1000ft, ValueLand, Proximity10ftParcelDensity,
Proximity1000ftParcelDensity with n=132,842. A significant regression equation was found
(F(27,132,815) = 73.875, p < 0.001 ) with an R2 of 0.015.
After the analysis the 12 coefficients that influence Number of Residences the most are listed in the
following table.
Table 16. Twelve most influential variables for residential, all years
Variables Beta Value Coefficient
Distance2LavaFlow1250 -.062
Distance2VolcanoHazard .060
Distance2ExistingMF1 -.050
Distance2SewerService -.048
Intersections1miDensity -.043
Distance2WaterService .035
Distance2Coastline -.028
Distance2Airports .028
Rainfall -.025
Distance2Towns -.024
RESAllYearsProximitySum1000ft .022
Slope -.021
Residential Post 1995
The first multiple linear regression was calculated to predict NumberOfResidences for residential
buildings built on or after 1995 based on ParcelPerimeter2DivArea, Intersections1miDensity, Slope,
Rainfall, Distance2Major10, Distance2WaterService, Distance2SewerService, Distance2Schools,
Distance2Coastline, Distance2VolcanoHazard, Distance2Airports, Distance2LavaFlow1250,
Distance2LavaFlow1790, Distance2Towns, Distance2ElectricService, Distance2Hospitals,
Distance24CenterAdjusted, Distance2ExistingSF1, Distance2ExistingMF1, Distance2Belt10,
DUpost1995ProximitySum10ft, DUpost1995ProximitySum1000ft, ValueLand,
Proximity10ftParcelDensity, Proximity1000ftParcelDensity with n=18,819. A significant regression
equation was found (F(27,18,792) = 11.549, p < 0.001 ) with an R2 of 0.016.
After the analysis the 12 coefficients that influence Number of Residences built after 1995 the most are
found in the table below.
Trend Scenario and Land Use Allocation Technical Report 23
Table 17. Twelve most influential variables for residential, post 1995
Variables Beta Value Coefficient
Distance2VolcanoHazard .081
Distance2LavaFlow1250 -.057
Slope -.049
Intersections1miDensity -.044
Rainfall -.043
Distance24CenterAdjusted -.041
RESpost1995ProximitySum1000ft -.039
RESpost1995ProximitySum10ft -.032
ParcelPerimeter2DivArea .031
Distance2Coastline -.030
Distance2Schools -.029
Distance2Hospitals 0.23
Residential 1975-95
The first multiple linear regression was calculated to predict NumberOfResidences for residential
buildings built on or after 1975 and before 1995 based on ParcelPerimeter2DivArea,
Intersections1miDensity, Slope, Rainfall, Distance2Major10, Distance2WaterService,
Distance2SewerService, Distance2Schools, Distance2Coastline, Distance2VolcanoHazard,
Distance2Airports, Distance2LavaFlow1250, Distance2LavaFlow1790, Distance2Towns,
Distance2ElectricService, Distance2Hospitals, Distance24CenterAdjusted, Distance2ExistingSF1,
Distance2ExistingMF1, Distance2Belt10, DUpost1995ProximitySum10ft,
DUpost1995ProximitySum1000ft, ValueLand, Proximity10ftParcelDensity, Proximity1000ftParcelDensity
with n=23,822. A significant regression equation was found (F(27,23,795) = 17.956, p < 0.001 ) with an R2
of 0.020.
After the analysis the 12 coefficients that influence Number of Residences built between 1975-95 the
most are found in the table below.
Table 18. Twelve most influential variables for Residential > 1975 and < 1995
Variables Absolute Value Coefficient
Distance2LavaFlow1250 -.173
Distance2ExistingMF1 -.137
Distance2Airports .123
Distance2SewerService -.106
Distance2VolcanoHazard .079
Intersections1miDensity -.078
Distance2Schools .073
Distance2WaterService .073
Proximity10ftParcelDensity .064
Proximity1000ftParcelDensity -.059
Distance2LavaFlow1790 .051
Slope -.049