Econometric Analysis

Introduction

As mentioned in earlier sections (e.g., CVA), land provides a variety of benefits such as water quality and biodiversity. Although it is true that these are substantial benefits, these are typically not significant in land development decisions. Instead, this section engages in analyses that take into account the market-oriented factors that do impact decision patterns. As a means of identifying areas of likely future change in the landscape, an econometric model was employed. Understanding the patterns of likely development (or highest market value) becomes a powerful tool for comparing with areas of highest conservation value. This comparison is provided later in the report as shown in Table 2-35 and Figure 2-61. This section describes the theoretical basis for the econometric modeling, it details the procedures followed, and describes the results of the model that are incorporated into other parts of this assessment.

The econometric analysis is fairly different than the previously discussed build-out analysis (in Section 2-3. above) both in its nature and its application. The build-out analysis was conducted under a set of temporal assumptions looking only at the present State of land policy and conditions. While the econometric analysis is based primarily on historic patterns which then get projected forward.

The general land allocation problem is that of a land user wishing to maximize net benefits obtained from an area by choosing the appropriate land uses. The problem faced by a user of L hectares is to allocate land between developed and rural uses. The land user’s allocation decisions depend on the land’s ability to provide benefits, the prices of outputs and inputs, and preferences for outputs. The area is divided into classes based on its ability to provide benefits (e.g., timber, recreation, crops). For discussion purposes, treat relevant land attributes (e.g., soil fertility, physiographic characteristics) as a composite commodity measured by a scalar, q, called land quality that is defined so that higher quality land provides more benefits than lower quality land. Quality ranges from worst, q^-, to best, q⁺, with L(q) acres in each class. Thus the total area L is divided into tracts so that L = L(q^-) + ... + L(q⁺).

Maximizing net benefits from the land input requires the user to allocate uses to land classes. Land benefits consist of the present value of net benefits from developed and rural uses, PVNB^*D(p,q) and PVNB^*R(p,q), respectively. The user determines these potential benefits by determining optimum developed or rural uses on land quality q when faced with price vector p for land benefits and nonland inputs. Given PVNB^*D(p,q) and PVNB^*R(p,q), the user can obtain maximum benefits from the entire ownership by selecting f(q), the proportion of land quality class q that should be devoted to developed uses. If we allow the function f(q) to describe the distribution of quality on the ownership, then the user’s total net benefit from the entire ownership can be written

PVNB = S_q=q-,q+{ f(q)PVNB^*D(p,q) + [1- f(q)]PVNB^*R(p,q)}f(q)L.

The benefits that the user can obtain are subject to the available land distribution, the feasible land use choices, and the production possibilities implicit in PVNB^*D(p,q) and PVNB^*R(p,q).

To simulate the effects of land policies, specific policy tools (e.g., subsidies, taxes) must be included in the land benefit functions. Other policies such as zoning could be introduced as constraints on the land quality distribution, or restrictions on potential uses for specific types of land.

Since PVNB is linear in f(q), the user will maximize benefits by solving dPVNB/df(q) = 0, which requires allocating land to developed uses until

PVNB^*D(p,q^*) - PVNB^*R(p,q^*) = 0,

then selecting rural use. For qualities below the land quality margin q^*, rural benefits exceed developed benefits and the user maximizes total net benefit by allocating the land to rural use (i.e., choosing f(q)=1). For hectares with quality above q^*, developed benefits exceed rural benefits and the user maximizes total net benefit by allocating the land to developed use (i.e., choosing f(q)=0). In the event benefits from one use exceed the other over the entire quality range, the user will obtain maximum benefits by selecting a single land use.

The optimal amount of land for the user to devote to forest use is F(q^*)L, where F is the cumulative distribution function corresponding to f. Because the attributes relevant to the user to characterize land quality may be spatial (e.g., distance to where land products are used) or nonspatial (e.g., soil fertility), the quality margin between land use alternatives may or may not be associated with a contiguous location.

Methods

In this type of land model, the landscape at time t is treated as a collection of N individual tracts, each of which has its own land quality. A specific tract of land i will be developed at time t if land quality is above the critical threshold implicitly defined by PVNB^*D(p_it,q^*_t) = PVNB^*R(p_it,q^*_t) for that tract and time. Let q_it = g(x_it'b_t) be an unobservable index of quality for tract i. The index may be unobservable because (i) some components of the composite quality index q may not be observable, or (ii) land benefits cannot be precisely calculated from the available data, or (iii) some components of benefits cannot be observed.

Observable data include x_it, a vector of attributes of the tract at time t (e.g., components of net benefit from competing uses, observable land attributes), and the outcome of the land allocation decision for the tract. For each tract, if land quality is below the level q^*_it, the land in the tract is optimally maintained in rural use; otherwise, it is developed.

The index g(x_it'b_t) is defined so that the probability that a tract drawn at random at time t is developed is P_it = Probability{q^*_t £ g(x_it'b_t)}. Since this probability is bounded by zero and one, and quality classes in this model are monotonically arranged from worst to best, the relationship between q_it and P_it can take the form of a cumulative distribution function (cdf), for example, the logistic cdf:

P_it = Pr{q^*_t £ g(x_it'b_t)} = 1/ {1+exp[-g(x_it'b_t)]}.

Since the land use on the tract is observable, we can define a variable y_it that takes the value 1 when the land use developed and zero otherwise. Defining this variable allows the probability of observing the land uses for the N tracts to be written as a single equation that depends on x_it , y_it and the parameters of the distribution function for land quality.

The probability of observing the land uses that are in fact present on the landscape is

Õ_i=1,N (1-P_it)^(1-yit) (P_it)^(1-yit)

and is referred to as the likelihood of observing the sample data. The likelihood function depends on the parameters b_t in the logistic distribution, and estimates for b_t are obtained by finding those values of b_t that maximize the likelihood of observing the data.

Data

Land use data for 1995 and 2000 are constructed as a stratified random sample of 5750 points that were in rural use (agriculture or forest cover) in 1995. The land use data are from remotely-sensed data for the New York–New Jersey Highlands maintained by the Center for Remote Sensing and Data Analysis at Rutgers University. Publicly-owned lands and open water bodies are excluded. The sample of points is stratified so that the proportions of land that changed use between 1995 and 2000 in the sample match the proportions of land that changed use between 1995 and 2000 in the Highlands region. To allow for the possibility that the processes driving land use change may differ within the region, the Highlands are divided into upper New York, lower New York, upper New Jersey, and southern New Jersey subregions (Figure 2-38).

The available data for x_it include measures of land quality, block-level Census information, and policy variables influencing land use. Measures of land quality include slope, and an indicator variable indicating whether the land is classified as prime farmland. Census information includes 1990 population and household densities, and the 1990 housing value. Policy variables include the maximum housing density permitted by zoning, and the proportion of nearby lands in Forest Stewardship programs.

Spatial variables include linear distances to the nearest train stations, developed lands, and water bodies. Commuting costs to the nearest employment center and to New York City are approximated for each sample point based on its location. For this purpose, 22 employment centers are identified (Vernon township, Waterbury town, Flemington borough, Flemington, Oakland, Parsippany-Troy Hills, Mount Olive, Morristown, Wayne, Bridgewater, Newton, Phillipsburg, Newburgh, Mount Kisco, Poughkeepsie, Beacon, Middletown, Monroe, Peekskill, West Haverstraw, Ossining, and Suffern). By assigning different weights to interstate highways, Federal highways, State highways and local roads, the commuting cost variable measures the cumulative cost-distance for traveling from each sample point to an employment center that involves the least costly route. Commuting cost to New York is measured as the cumulative cost-distance (or cost-weighted distance) along the least costly route from the sample point to New York City. For a more detailed description of the 28 variables used in the econometric analysis, refer to Table 2-26. Descriptive statistics for the selected variables are listed in Table 2-27.

Table 2-26. Description of the 28 variables included in the initial explorations for the econometric analyses.

1. PLOT – unique ID number given to each sample points.

2. UPPERNY – binary variable identifying whether a sample point is in the upper New York region (one of the four Highlands scenarios).

3. LOWERNY – binary variable identifying whether a sample point is in the lower New York region (one of the four Highlands scenarios).

4. UPPERNJ – binary variable identifying whether a sample point is in the upper New Jersey region (one of the four Highlands scenarios).

5. LOWERNJ – binary variable identifying whether a sample point is in the lower New Jersey region (one of the four Highlands scenarios).

6. CHANGE – binary variable indicating whether there was a conversion of 1995 undeveloped land to 2000 developed land.

7. Slope – derived from DEM; measured in percentage.

8. DST_TRN – linear distance to nearest train station.

9. ZONE_C – binary variable describing whether zoning type is commercial.

10. ZONE_I – binary variable describing whether zoning type is industrial.

11. ZONE_M – binary variable describing whether zoning type is commercial and residential mixed.

12. ZONE_ONR – binary variable describing whether zoning type is other nonresidential.

13. ZONE_OS – binary variable describing whether zoning type is open space.

14. ZONE_R – binary variable describing whether zoning type is residential.

15. ZONEDNS – zoning housing density; extracted from zoning descriptions

16. COST_NYC – commuting cost to New York City through major roads. It was measured as cost-weighted distance by assigning different weights to interstate highway, US highway, State highway and local road and then calculating the cumulative cost of the least-cost route from each sample point to New York City.

17. PRIME -- binary variable indicating whether it is prime farmland.

18. POPDNS90 – 1990 census population density based on TIGER 1990 census block data.

19. HHDENS90 – 1990 census household density based on TIGER 1990 census block data.

20. POPDNS00 – 2000 census population density based on TIGER 2000 census block data.

21. HHDENS00 – 2000 census household density based on TIGER 2000 census block data.

22. DST_D95 – proximity to 1995 existing developed lands calculated as linear distance.

23. DST_D00 – proximity to 2000 existing developed lands measured as linear distance.

24. COST_JOB – commuting cost to nearest employment center through major roads. Within or around Highlands region, 22 employment centers were identified, including Vernon township, Waterbury town, Flemington borough, Flemington, Oakland, Parsippany-Troy Hills, Mount Olive, Morristown, Wayne, Bridgewater, Newton, Phillipsburg, Newburgh, Mount Kisco, Poughkeepsie, Beacon, Middletown, Monroe, Peekskill, West Haverstraw, Ossining, and Suffern. By assigning different weights to interstate highways, US highways, State highways and local roads, it measures the cumulative cost-distance for traveling from each sample point to an employment center that involves the least cost path.

25. DST_H2O – proximity to open water body calculated as linear distance.

26. FLDPLAIN – binary variable indicating whether it is within floodplain.

27. STEWARD – forest stewardship measured in percentage.

28. HVALUE90 – 1990 census housing value. It was mainly based on TIGER 1990 census block data. And for those blocks whose housing values were missing, they were substituted with housing values extracted from census block group data.

Table 2-27. Descriptive statistics of the variables used in the econometric analyses.

Variable	Mean	Std.Dev.	Minimum	Maximum	Cases
PLOT	2875.5000	1660.0264	1.0000	5750.0000	5750
UPPERNY	0.2845	0.4512	0.0000	1.0000	5750
LOWERNY	0.1397	0.3467	0.0000	1.0000	5750
UPPERNJ	0.2459	0.4307	0.0000	1.0000	5750
LOWERNJ	0.3299	0.4702	0.0000	1.0000	5750
CHANGE	0.0362	0.1867	0.0000	1.0000	5750
SLOPE	12.6186	10.1668	0.0000	70.3677	5750
DST_TRN	9801.9997	6572.4063	108.1670	29300.4004	5750
ZONE_C	0.0343	0.1819	0.0000	1.0000	5750
ZONE_I	0.0583	0.2343	0.0000	1.0000	5750
ZONE_M	0.0012	0.0349	0.0000	1.0000	5750
ZONE_ONR	0.0216	0.1453	0.0000	1.0000	5750
ZONE_OS	0.0024	0.0493	0.0000	1.0000	5750
ZONE_R	0.8823	0.3223	0.0000	1.0000	5750
ZONEDNS	1.3811	2.3426	0.0000	8.0000	5750
COST_NYC	131401.4830	42812.2140	30918.1992	272714.0000	5750
PRIME	0.1901	0.3924	0.0000	1.0000	5750
POPDNS90	0.4344	0.8966	0.0000	26.1800	5750
HHDENS90	0.1551	0.3243	0.0000	9.3000	5750
POPDNS00	0.5197	1.1199	0.0000	36.6200	5750
HHDENS00	0.1887	0.4822	0.0000	16.9400	5750
DST_D95	187.0789	188.1106	30.0000	1765.6700	5750
DST_D00	173.2743	182.4722	0.0000	1765.6700	5750
COST_JOB	54580.0356	34016.1588	0.0000	151527.0000	5750
DST_H20	1517.2313	1275.2775	15.0000	7542.9570	5750
FLDPLAIN	0.0384	0.1923	0.0000	1.0000	5750
STEWARD	1.3946	7.4529	0.0000	90.7584	5750
HVALUE90	228900.5220	108983.9600	0.0000	600000.0000	5750

Results

On average, roughly four percent of agricultural and forest points in the New York and New Jersey Highlands region changed to developed uses between 1995 and 2000. This conversion was highest in the upper New Jersey (six percent) and lower New York (five percent) subregions, followed by the lower New Jersey and upper New York subregions (three and two percent, respectively). Because these are average rates of change, some localized areas within the Highlands region changed use even more rapidly. Although the Highlands landscape is changing rapidly relative to other forested regions, from the point of view of the estimation procedure, these proportions are still relatively small. For example, the odds that an individual point drawn at random will remain in its agricultural or forest condition are better than twenty to one. Considering the relatively small probability of development, the models fit the data reasonably well.

The c² statistics for each subregional model appear to support a relationship between the probability of development and at least a few of the explanatory variables included (Table 2-26). It is also likely that co-linearity among the variables (especially the location variables) may result in lower statistical significance for at least some coefficients within the group. To the extent that co-linearity exists between other variables (for example population density and household density), the statistical significance of the related variables may be reduced. It could also be argued that proximity to development is correlated with minimum lot size, population/household density, and housing value.

As a group, the location variables appeared most strongly related to the probability of development. For example, proximity to land that was already developed by 1995 (the opposite of the distance to developed land variable included in Table 2-26) increased the probability of development in the New York and upper New Jersey Highlands regions. This suggests agglomeration of development between 1995 and 2000. In the lower New Jersey and upper New Jersey Highlands, proximity to train stations and water, respectively, also increased the probability of development. Other than location, prime farmland and land of high development value showed higher probability of change in upper New York, while land of shallower slope showed higher probability of change in lower New Jersey.

A major advantage of the combined geospatial and econometric approaches used here to study the Highlands is that the econometric model can be incorporated into a geographic information system (GIS) and used to map the model results. Each of the explanatory variables (x_it) is a layer within the GIS. A new layer representing the estimated probability of development was developed by using the fact that the estimated probability of development is (1+exp[-g(x_it'b_t)])^-1. Using the parameters from Table 2-26, a map of estimated probabilities of development was constructed (Figure 2-39).

Discussion

The econometric model of land use change should not be viewed as a crystal ball infallibly predicting the future but rather as providing a hazy view of possible changes in the near term future. Visual examination of the predicted likelihood of change map (Figure 2-39) reveals the general patterns and highlights potential hotspots of future development. The central portions of the Highlands in northern New Jersey and southern New York show higher probabilities of land use change in the near term. The northeastern portion of the Highlands study area in Dutchess and Putnam counties and the extreme western portion of study area in Hunterdon and Warren counties show lower probabilities of change. Whereas on the whole the model results appear to reasonably reflect present development trends, there are some areas where the near term change appears to be underestimated. For example, the Interstate 78 corridor of southwestern Warren County would appear to be such an area of underestimated near term future change. It must be recognized that the econometric model is empirically based and was calibrated on the 1995 to 2000 land use and land cover change data. During this 1995 to 2000 time period, the northeastern and western portion of the study area experienced comparatively lower levels of development and land use change, which then translated into lower estimates of modeled future change.

The econometric modeling approach can also be used to examine the effect of changes in any of the variables on the estimated probabilities. For example, the effects of a hypothetical change in the minimum lot size could be translated into a map of new and potentially different probabilities of development. It is a relatively simple procedure to change any of the variables in x_it and then create new maps of estimated probabilities. Considerable skepticism must be used in viewing a map created in this way, because the estimated changes in probabilities are no more reliable than the underlying parameter estimates in Tables 2-28 and 2-29. Only a few of the variables displayed statistical significance, and most of these were distance measures. Although technically possible, no simulated probabilities for changes in x_it were attempted for this reason.

The parameter estimates shown in Table 2-27 do not show the change in the predicted probability associated with changes in these variables (such changes are referred to as marginal effects). Evaluating marginal effects is accomplished by deriving elasticities from the estimated logit parameters (Table 2-28 and Table 2-29). The elasticity in the probability of development with respect to the kth element of x_it is defined here as the percentage effect of a one unit increase in x_ijtk on the proportion of land in forests (suppressing the subscripts i and t):

e_k(x) = ¶P(x)\¶x_k × x_k\P(x)

= -¶g(x'b)\¶x_k × exp[-g(x'b)]/[1+exp[-g(x'b)]]² × x_k\P(x).

It is clear that the elasticity of probability depends on the attributes of the sample point, x, and can be calculated using this formula for any value of x. Maps of elasticities could potentially highlight the percentage change in the probability of development in response to a percentage change in population density, household value, or any other of the variables in x. However, the same qualification made earlier for predicted probabilities applies equally well to elasticities. This type of analysis was outside the scope of the Study Update but would appear to be a fruitful area for future research.

Table 2-28. Logit parameter estimates^a for change in agricultural and forest lands in the New York Highlands region between 1995 and 2000.

Variable Lower New York Upper New York

Slope -0.00691 0.0111

(0.0206) (0.0170)

Prime farmland -0.247 0.713*

(0.509) (0.440)

Population density 0.154 0.0279

(0.519) (0.453)

Household density 0.316 -0.0489

(1.53) (1.18)

Value of housing 0.0639 0.320*

(0.201) (0.181)

Maximum housing -0.0319 0.0127

density (0.0981) (0.0724)

Proportion of land -0.0919 -0.0469

in Forest Stewardship (0.120) (0.0744)

Distance to train -0.00114 -0.000171

(0.00445) (0.00450)

Distance to water 0.0235 -0.000806

(0.0188) (0.0295)

Distance to developed -0.858* -0.522*

land (0.277) (0.177)

Commuting cost to -0.487 -0.325

nearest center (1.05) (1.11)

Commuting cost to 0.311 0.421

New York City (0.666) (1.12)

Constant -2.52* -4.31*

(0.957) (1.15)

_____________________________________________________________

c²(12) ^b 34.5* 23.7*

N 803. 1636

_____________________________________________________________

* Indicates estimate is significantly different from zero (P£0.10).

^a Numbers in parentheses are asymptotic standard errors.

^b Tests the joint hypothesis that all coefficients are simultaneously zero.

Table 2-29. Logit parameter estimates^a for change in agricultural and forest lands in the New Jersey Highlands region between 1995 and 2000.

Variable Lower New Jersey Upper New Jersey

Slope -0.0635* -0.0141

(0.0288) (0.0138)

Prime farmland 0.240 -0.0834

(0.328) (0.358)

Population density -0.0509 -0.202

(0.221) (0.272)

Household density -0.174 0.657

(0.776) (0.695)

Value of housing 0.0730 0.0639

(0.155) (0.201)

Maximum housing -0.0205 -0.0132

density (0.0588) (0.0511)

Proportion of land -0.00159 0.0248

in Forest Stewardship (0.0452) (0.0184)

Distance to train -0.0134* 0.00285

(0.00402) (0.00317)

Distance to water 0.0101 -0.0514*

(0.0114) (0.0189)

Distance to developed -0.234 -0.443*

land (0.143) (0.129)

Commuting cost to -0.0251 -0.155

nearest center (0.748) (0.365)

Commuting cost to -0.663 -0.477

New York City (0.825) (0.480)

Constant -0.298 -1.58*

(0.982) (0.594)

_____________________________________________________________

c²(12) ^b 47.3* 37.6*

N 1897 1414

_____________________________________________________________

* Indicates estimate is significantly different from zero (P£0.10).

^a Numbers in parentheses are asymptotic standard errors.

^b Tests the joint hypothesis that all coefficients are simultaneously zero.

Figure 2-38. Map of the four sub-regions used for the econometric analysis.

Figure 2-39. Map of likelihood of change according to the econometric analysis.

Possible Consequences of Future Change to Resources

Changes in Land Use and Land Cover

Objectives

Continued human development is the primary factor shaping the New York–New Jersey Highlands region, and so a better understanding of past and present trends in land use and land cover change is critical. The objective of the land use and land cover analysis is two-fold: 1) to provide a consistent assessment of present day land use and land cover across the broader four State Highlands region; and 2) to provide a more detailed analysis of the New York–New Jersey Highlands by mapping changes over the past three decades. A combination of Landsat Thematic Mapper satellite imagery, digital orthophotography, and existing (land use and land cover) data sets were used to undertake the above analyses. The resultant land use and land cover data set enabled the derivation of landscape level indicators of watershed condition and forest integrity for comparative purposes across varying levels of analysis, e.g., region, watershed, county and municipality.

Land Cover Mapping Methods

The objective of the land cover change analysis component was to assess past changes and present trends in land cover conversion from the 1970’s through 2000. The only consistent land cover data set available for the Highlands study region prior to this project was the USGS National Land Cover data set that dates from the early 1990’s. Earlier, the USGS mapped land use and land cover from high altitude aerial photography for the mid 1970’s time period (this data set was called LUDA). Satellite and digital ortho-photographic imagery were used as the basis for the land cover change analysis component of the New York–New Jersey Highlands Regional Study: 2002 Update. Use of these data permitted the creation of standardized land cover information that is consistent across the bi-State region.

The USGS National Land Cover data set was used to provide a general background of land cover patterns across the broader four State highlands region. The USGS data set did not meet project requirement of consistent land use and land cover classification and mapping for the bi-State study region, across multiple time periods, up to and including the year 2000. Therefore the decision was made to create a unique land use and land cover product for the New York–New Jersey Highlands study region. Land cover was mapped in four instances that span 28 years: 1972, 1984, 1995, and 2000. Standard protocols developed as part of the landscape program mapping undertaken for the State of New Jersey (Lathrop, 2000) were used to provide for a land cover database consistent with the existing New Jersey statewide data set.

The land cover mapping was undertaken at two levels of generalization: Level I, the most generalized with 8 classes; and Level 2, with 15 classes (Table 2-30). The Level I and Level II classification schemes were designed to follow the NOAA Coastal Land Cover Classification System (Dobson et al., 1995). The more generalized Level I classification scheme was used for comparison of the 1972, 1984, 1995 and 2000 classifications.

Table 2-30. Classification scheme used in land use and land cover mapping.

Table 2-30a. Land cover level I (8 classes)

1 1.10 Developed (Classes 1.11-1.14)

2 1.20 Cultivated and grassland (Classes 1.20, 1.30)

3 1.40 Woody land (Classes 1.41-1.44)

4 1.60 Bare land (Class 1.60)

5 2.00 Unconsolidated shore (Class 2.00)

6 2.10 Estuarine emergent wetland (Class 2.10)

7 2.40 Palustrine wetland (Classes 2.30, 2.40)

8 2.50 Water (Class 2.50)

Note:

Developed land includes impervious, bare or partially vegetated land surfaces due to commercial, industrial, residential and transportation land uses.

Forest and wetland land covers include upland and wetland forests, scrub and shrub, and emergent vegetation communities.

Cultivated and grassland includes agricultural lands (including cultivated land, pastures and hay fields), managed grasslands (e.g., large areas of mowed, irrigated and fertilized lawn, and golf courses) and unmanaged grassland.

Bare land includes lands made barren by quarrying and mining activities.

Table 2-30b. Land cover level II (15 classes)

1 1.11 Highly developed (> 75 percent impervious surface)

2 1.12 Moderately (25-75 percent impervious surface)

3 1.20 Cultivated

4 1.30 Grassland

5 1.41 Upland forest: deciduous dominant (>66 percent)

6 1.42 Upland forest: coniferous dominant (>66 percent)

7 1.43 Upland forest: mixed deciduous and coniferous (>66 percent)

8 1.45 Upland scrub and shrub

9 1.60 Bare land

10 2.00 Unconsolidated shore

11 2.10 Estuarine emergent wetland

12 2.30 Palustrine emergent wetland

13 2.40 Wetland forest: palustrine

14 2.45 Wetland shrub and scrub: palustrine

15 2.50 Water

A combination of digital image analysis techniques were used to classify the Landsat Thematic Mapper (TM) and Multispectral Scanner (MSS) images into land cover maps. Additional digital data sets were incorporated in the context of a geographic information system (GIS) to provide further classification improvement. Two major types of GIS data were used to guide the classification process: wetlands data (e.g., U.S. Fish and Wildlife Service National Wetland Inventory (NWI), New Jersey Department of Environmental Protection (NJ DEP) freshwater wetlands, New York State Department of Environmental Conservation (NYS DEC) wetlands, USDA Natural Resources Conservation Service (NRCS) soils, and land use data (NJ DEP and New York county-level land use, U.S. Geological Survey land use and land cover)). These data sets were incorporated into the classification process for either pre-classification stratification or post-classification modification. For example, the various wetland data sets were combined to create a new data theme that showed the likelihood that any particular grid cell should be classified as a wetland and this information was used in the classification process. The land use data were used to stratify the study area into primary land use types (e.g., developed, agriculture, vacant). These were then used to constrain the land cover classification (i.e., developed land cover types could only be found in grid cells that were mapped as developed land use).

The land use maps were created independently for New Jersey and New York. The 1995 digital orthophotographs of New York (1994/1995) and New Jersey (1995/1997) were simultaneously available, and so served as the baseline for land cover mapping. Approximately 50 percent of the New York imagery was from 1994 and 50 percent from 1995. A vast majority of the New Jersey imagery was from 1995 with only a small portion of western Hunterdon and Warren counties covered by 1997 imagery. A mosaic of color infrared digital orthophotographic quarter-quadrangles (DOQQs) at 1-meter grid cell resolution was created for the bi-State Highlands region. The 1995 land use and land cover data were available for the portion of the Highlands study area that falls in New Jersey (NJ DEP, 2000). Land use maps were available at the county level in New York State and generally dated from the early to mid 1990’s. These land use maps were overlain on the 1995 DOQQs. The land cover was updated as necessary using on-screen interpretation and digitization. The land use categories mapped are listed in Table 2-31.

Table 2-31. Land use categories (11 classes).

Class # Description

1 Residential

2 Industrial/commercial

3 Institutional/recreational

4 Transportation

5 Utility

6 Quarry/landfill

7 Agricultural

8 Forest

9 Undeveloped/vacant

10 Transitional

11 Wetlands/water

A similar process was used to overlay the original county level land use maps on the 1984 Landsat TM imagery and updated accordingly. In this case, areas of new development (i.e. development occurring from 1984 to the 1990’s) were removed. Unfortunately, the coarser spatial resolution of the Landsat TM imagery (30 meters as compared to the 1 meter of the DOQQ’s) limited our ability to interpret land use and create highly accurate land use maps for the New York portion of the study area for 1984. Land use and land cover data was available for the New Jersey portion of the study area for the year 1986 (NJ DEP, 1996).

A combination of SPOT panchromatic and Landsat TM imagery were used to provide updated land use for the year 2000. The SPOT panchromatic imagery (10-meter spatial resolution) was available as a region-wide mosaic of individual SPOT scenes that spanned a range of dates from 1998 to 2000 (only a small subset of the study area had 1998 data, the majority was from 2000). Thus, the same SPOT scene did not cover the entire New York–New Jersey Highlands region. This panchromatic data was merged with the September 23, 1999 Landsat TM imagery using a Principal Components resolution merge algorithm. The 1995 land use maps were overlain on the composite image, and areas of new development (development subsequent to 1995) were interpreted and digitized, on-screen, to produce a year 2000 land use map.

Landsat imagery served as the remotely sensed source data used to consistently map land cover across the entire New York–New Jersey Highlands study region throughout the period of interest. Landsat TM images were acquired for relatively cloud-free dates in 1994 and 1995 (November 4, 1994 and September 4, 1995) for the 1995 baseline. Cloud covered areas were replaced with December 22, 1994 imagery. The November "leaf-off" imagery was taken after normal deciduous plant leaf fall, allowing the clearer differentiation of evergreen and deciduous forests as well as developed areas. The September "leaf-on" imagery permits the further discrimination of cultivated, wetland and developed areas. Landsat TM images from April 5, 2001 and September 23, 1999 were used to provide more recent land cover information for the year 2000. Corresponding images from November 8, 1984 and September 21, 1984 were acquired in order to perform change detection. Earlier generation Landsat Multi-spectral Scanner (MSS) imagery from October 10, 1972 was also acquired to extend comparisons further back in time, albeit at a coarser spatial resolution and more generalized level of categorization.

A "ground-truthing" field campaign was undertaken as part of an earlier project to map land cover across the State of New Jersey. Over 300 field sites were visited during the fall of 1994 and winter/spring months of 1995, simultaneously with image acquisition and prior to any classification activities. Over 1,400 field sites in New Jersey were visited in the spring of 1997 and 2000 to serve as additional, post-classification accuracy assessment ground reference sites. Results of the accuracy assessment suggests that the 1995 Level I land cover map is approximately 92 percent correct, while the more detailed Level II land cover map is correct approximately 80-85 percent of the time. Additional field checking of over 360 candidate land use change polygons was undertaken during the October 2002 to assess the year 2000 land use change map. The results suggest that land use alteration and change (i.e., from a non-developed to a developed or transitional land use category) between 1995 and 2000 was correctly identified 89 percent of the time. No attempt was made to independently assess the accuracy of the 1972 or 1984 image maps due to the absence of appropriate field reference data for those years. While these maps are not completely error free, we conclude that the land cover data are of sufficient quality that they can be used in further regional to landscape scale analysis with confidence. Greater detail concerning the accuracy assessment can be found in Appendix 2-A.

Results: Land Use and Land Cover Change

Forest land dominates the greater Highlands region that stretches across the states of Pennsylvania, New Jersey, New York and Connecticut (Figure 2-40). Farmland is clustered in the southwestern portions of the Highlands in Pennsylvania and New Jersey, in the western portions of Orange County and the northeastern corner of the study area in Dutchess and Putnam counties in New York. The southeastern edge of the study area is adjacent to the greater New York City metropolitan area and represents the most heavily urbanized portions of the Highlands (e.g., Rockland County, NY). Other largely urban areas occur around some of the major cities and towns elsewhere in Highlands: Easton and Phillipsburg in the southwest; Morristown in the center; and Peekskill in the north. A more comprehensive land use and land cover analysis was conducted for the New York–New Jersey Highlands study area. This analysis shows that the Highlands study area (as of 2000) is dominated by upland forest land cover types at 51 percent of the total Highlands area, followed by developed land at 24 percent, cultivated/grassland at 12 percent, and wetlands/water at 12 percent (Table 2-32).

Comparison of the 1972, 1984, 1995 and 2000 land cover maps shows that human development of the Highlands has increased steadily over the past three decades (Table 2-32). Changes occurring to the Highlands landscape are largely the result of human activities, namely residential, commercial, and, to a lesser extent, industrial development. The amount of developed land cover area in the New York–New Jersey Highlands increased from approximately 280,000 acres in 1984 to nearly 345,000 acres in the year 2000 (Table 2-32). Thus developed land cover types comprise approximately 24 percent of the New York–New Jersey Highlands area in the year 2000, a five percent increase from 1984. This growth in developed area equates to an increase of 1.3 to 1.6 percent annually (Table 2-33). The total area covered by urban land uses (based on the land use map rather than land cover map), which includes urban/suburban/rural/commercial/industrial development, lawn/yards, and developed park areas, is estimated to be even larger at more than 450,000 acres in the year 2000. It should be noted that the developed land use figure is greater than the developed land cover because a tract of land mapped as residential land use may include various land covers from moderate to lightly developed to grass or forest land cover.

The 1984, 1995, and 2000 land cover sets were created from the same satellite sensor and comparable auxiliary data sets, and so provide a good estimate of land cover change during this time period. The 1972 land cover map provides a coarser view of land cover during this earlier time period due to the cruder technology of the Landsat Multispectral Scanner platform. The area estimates are, therefore, not directly comparable to the later 1984 and 1995 map series. Our qualitative accuracy assessment for the 1972 land cover map suggests that developed area is underestimated due to the larger minimum mapping unit (coarser scale) of the developed land cover mask (derived from 1:250,000 scale USGS land use and land cover data). Housing or commercial/industrial developments and other developed land covers that are smaller than approximately 10 acres (4 hectares) are not consistently mapped in the 1972 time period. The minimum mapping unit for the 1984, 1995 and 2000 developed land cover mask (derived from NJ DEP ITU and CRSSA photo-interpreted data) is approximately 2.5 acres. Comparison of the developed land cover area estimates across the three time periods shows an increase of 82,200 acres between 1972 and 1984 and an increase of 39,769 acres between 1984 and 1995 (Table 2-32). Thus, over a comparable 11-12 year period, the change between 1972 and 1984 is approximately twice that observed between 1984 and 1995. A substantial (but un-quantified) portion of the developed area increase between 1972 and 1984 is an artifact of a change in delineation methods. Dispersed suburban/exurban development that may have existed in 1972 was not mapped, and is found for the first time in the 1984 land cover map.

Both farm and forest land has been converted to residential and commercial land uses to meet the demands of a regional population change that grew by more than 11 percent between 1990 and 2000. Analysis of the year 1995 to 2000 change indicates that the annual rate of forest loss to development is increasing, while the amount of farmland loss is decreasing. In the New York–New Jersey Highlands study area, the rate of forest and wetland land covers (predominantly upland forest) converted to developed land covers increased from a rate of over 1500 acres/year (or 0.2 percent/year) between 1984 and 1995 to 3,400 acres/year (0.4 percent/year) between 1995 and 2000 (Table 2-33). The majority of this change is explained by the loss of upland forest (1984: 757,115 acres; 1995: 737,996 acres; 2000: 721,293 acres) when examined separately. Estuarine (along the Hudson River) and palustrine wetlands (e.g., freshwater marshes, swamps and riparian forests) were comparatively stable across time (1984: 100,309 acres; 1995: 103,555 acres; 2000: 102,253 acres). The rate of cultivated and grassland land covers (predominantly farmland) converted to other land covers, either developed or, in some cases, proceeding through natural succession to forest land, decreased from a rate of nearly 2,240 acres/year (or 1.1 percent/year) between 1984-1995 to nearly 1,600 acres/year (0.9 percent/year) between 1995-2000 (Table 2-33). The amount of water cover has actually increased over the study period due to the impoundment of Merrill Creek and Monksville Reservoir.

Table 2-32. New York–New Jersey Highlands land cover trends (acres).

Land Cover	1972	1984	1995	2000
Developed	197,002	278,999	318,768	344,569
Cultivated/ grassland	223,732	208,790	184,190	176,200
Forest	804,766	757,115	736,996	721,293
Wetland	127,312	100,309	103,556	102,254
Barren	3,201	10,069	10,262	9,652
Water	61,946	62,587	64,502	64,305
Total*	1,417,959	1,417,869	1,418,273	1,418,273

*Note: differences in area totals due to rounding errors.

Table 2-33. New York–New Jersey Highlands: land cover changes.

Type	1984-1995 Area change Percent change	1995-2000 Area change Percent change
Developed	+39,769 acres +14.2 percent	25,801 acres +8.1 percent
	+3,615 acres/year +1.3 %/year	5,160 acres/year +1.6 %/year
Forest/Wetland	-16,873 acres -2.0 percent	-17,004 acres -2.0 percent
	-1,534 acres/year -0.2 %/ year	-3,400 acres/year -0.4 %/year
Farmland/Grassland	-24,600 acres -11.8 percent	-7,990 acres -4.3 percent
	-2,236 acres/year -1.1 %/ year	-1,598 acres/year -0.9 %/year

Figure 2-40. Map of land cover of four State Highlands region.

Source: National Land Cover Data Set, U.S. Geological Survey.

Discussion

The Highlands contains a diversity of land uses and land covers. While extensive areas of the Highlands consist of large contiguous tracts of semi-wilderness forest and watershed lands, the Highlands study area also contains other landscape types including: river valley agricultural areas with scattered villages; rural areas with a mix of housing, woods and fields; suburban towns; and small cities. The land use and land cover analysis shows that while forest land still dominates, human development has increased steadily over the past three decades. Typical of the spatial patterns associated with urban sprawl, the tracts of new development are widely dispersed throughout the New York–New Jersey Highlands region (Figure 2-41). Thus, rather than a single “sprawl line” that separates suburban from exurban (i.e. rural) areas and cuts across the larger study region, the land use and land cover change analysis suggests a much more chaotic pattern. While the new urban growth has been most heavily concentrated along the southeastern edges of the study area, it has sprawled along major highway corridors, and leapfrogged across the Highlands region. As interstate highways (e.g., Routes 78, 80, 84, 287, the New York State Thruway) have crisscrossed the region, and State highways have expanded into major commuting routes (e.g., New Jersey Routes 15, 23, 206, and New York Route 17), no area is beyond the reach of the New York City metro commuter. Office park and retail development has followed suit as residential development has moved outward, creating a positive feedback cycle of further development in more rural settings.

The rate of development per year increased from 3,695 acres per year between 1984 and 1995 to 5,159 acres per year between 1995 and 2000, which equates to an increase of 1.3 to 1.6 percent annually (Table 2-33). Both farm and forest land has been converted to residential and commercial land uses to meet the demands of a growing regional population. Analysis of the change between 1995 and 2000 indicates that the annual rate of forest loss to development is increasing, while the amount of farmland loss is decreasing. This shift may reflect the amount of readily available land close to the New York City metropolitan area with farmland developed first and a more recent focus shifting to forested tracts.

While the best possible effort was made to map land use and land cover in a consistent manner and with high accuracy across the entire area and throughout various time periods, some error is inevitable. Thus, the land use and land cover data should be considered estimates with some margin of error. Further accuracy assessment is being undertaken to examine the accuracy of the 1995 land cover mapping in the New York State portion of the study area. The 1995 to 2000 land cover change estimates also need further review due to the coarser spatial resolution of the year 2000 land use mapping (i.e. based on SPOT panchromatic 15 meter imagery rather than 1 meter digital orthophotography). These studies are ongoing, and will be reported as part of the final metadata for these data sets.

Figure 2-41. Map of urbanization in the New York–New Jersey Highlands.

.

Landscape Indicators of Forest and Watershed Integrity

Introduction

There is a great need for the simplification and synthesis of land use and land cover change data to provide information that is useful to land managers and policy makers. The simple metrics for analyzing, monitoring, and communicating information about change are often referred to as environmental indicators within the environmental management literature (ERMS 2000; Jones and Simmers 2001). There has been a great push by Federal agencies to develop land cover data sets and indicators that are suitable for measuring and monitoring land cover and associated environmental change at broad regional scales (Jones et al., 1997; USEPA, 1998). Similarly, since 1995, the New Jersey Department of Environmental Protection (NJ DEP) has embraced a results-based management system that relies on indicators to ascertain progress toward environmental goals (Kaplan and McGeorge, 2001). Many of the NJ DEP measures are statewide and, where applicable, are disaggregated to finer watershed scales (NJ DEP 2000, 2001a, 2001b).

The USDA Forest Service employed the environmental indicator approach in its nationwide study, National Projections of Forest and Rangeland Condition Indicators (Hof et al., 1999). This study analyzed past change, but also projected these indicators into the future (to the year 2020). The Hof et al. study identified the New York–New Jersey Highlands and the neighboring ridge & valley region of Pennsylvania, New Jersey and New York as representative of a hotspot of degrading forest condition indicators that require further management consideration from an ecosystem perspective. The Hof et al. study, as a nationwide investigation, was very broad in scope. We adopted a similar conceptual framework for the New York–New Jersey Highlands Regional Study: 2002 Update, but employed a different methodology and suite of indicators at a much finer level of resolution.

Objectives

A suite of landscape level indicators were chosen to quantify important components of the Highlands land use and land cover as one means of measuring the condition of the New York–New Jersey Highlands forests and watersheds: 1) percentages of altered and unaltered land cover; 2) indices of forest fragmentation; 3) percentage of impervious surface cover; and 4) percentage of the riparian areas of permanent streams that is in a vegetated, as compared to developed state. The land use and land cover described in a previous section served as the basis for the development of these landscape indicators. The environmental indicators were analyzed on a watershed basis, aggregating results to the HUC 11 (USGS Hydrologic Unit Code 11) level. There are 51 complete or partial HUC 11 watersheds within the New York–New Jersey Highlands study region. The four indicators were calculated for each of the 51 watersheds for each of the three years for which land use and land cover was established (1984, 1995, and 2000). This was done to permit analysis of existing trends and to estimate possible future conditions (low- and high-constraint build-out scenarios). The relationships between selected landscape indicators and independently measured environmental parameters were examined to assist in identifying important thresholds that may signify high potential for environmental degradation.

Background and Methods

Percentage of Altered vs. Unaltered Land Cover

The original, and still primary, land cover of the New York–New Jersey Highlands is forest and, to a lesser extent, wetlands. Undisturbed forest and wetlands provide the essential ecosystem services of natural water filtration and groundwater recharge. Allan and Flecker (1993) identified land use as one of the most important factors determining non-point source pollution. Clean, high quality water is one of the New York–New Jersey Highlands most important natural resources. Urban and suburban areas generate significant pollution loadings from street litter deposition, atmospheric fallout, road traffic, soil erosion and soil-adsorbed pollutants (Novotny and Chesters, 1981). While erosion and soil loss by surface runoff is considered a predominant source of pollution from agricultural lands, leaching of fertilizers (nutrients N and P) and toxic pesticides and herbicides are also of concern (Novotny and Chesters, 1981). Extractive mining can also be a local source of sediment erosion but, to the best of our knowledge, there is no evidence of acid mine drainage in this region. As forest and wetlands are replaced by developed and agricultural land, downstream water quality suffers due to the combined impacts of point and non-point source pollution and soil erosion. Thus, there is a clear connection between land use and land cover and the integrity of Highlands water sources. The connection between forest and water quality has been appreciated for well over one hundred years as is evident in a reading of the 1899 New Jersey State Geologist’s Report on the forests of New Jersey (Russell, 1988).

U.S. EPA has developed a number of landscape indicators for watershed to regional scale assessments to quantify environmental conditions. One of the simplest, yet most powerful indicator, is the U-index which is the proportion of watershed area that is altered due to urban or agricultural land cover (EPA, 1997; 1994; O’Neill et al., 1998). We used a similar index that was modified slightly to include barren lands due to mining and quarrying activities, in addition to cultivated land, grassland and developed (urban) land cover as altered land. A simple majority threshold of 50 percent altered land was chosen as the critical threshold signifying increased risk of environmental degradation. This threshold was chosen based on several lines of evidence presented below.

Aquatic communities such as benthic macroinvertebrates (e.g., insects) and algae are used as biological indicators of stream health because of their ability to discriminate human influences on the environment in a predictable way (USGS, 2002). A study of more than 800 sites in New Jersey found that the primary factors related to degradation of benthic communities are the percentage of urban land use within the associated drainage basin as well as the amount of upstream wastewater discharges (Kennen, 1999). Conversely, the total amount of forested land within a drainage basin was the best predictor of an unimpaired community. The USGS National Water Quality Assessment (NAWQA) study compared the aquatic community status as indicators of the water quality of selected watersheds to the proportion of urban land use in the watershed (Ayers et al., 2000). Several Highlands streams had some of the lowest scores (least degraded sites) nationally for aquatic community status (Figure 2-42). The land use in these basins is less than 34 percent urban. Analysis of the NAWQA study for other central and northern New Jersey watersheds shows that those basins with greater than 40 percent urban land cover had either medium or high stream degradation scores (Figure 2-42).

Figure 2-42. Comparison of biological status of New Jersey Highlands stream sites to 140 nationwide NAWQA sites using an urban land use gradient (highlighted stream names are Highlands streams. Source: US Geological Survey.

The surface waters (major streams, rivers and lakes) of New York and New Jersey are classified according to water quality and protection status (NJ DEP, 2001; NYS DEC, 2001). The percentages of developed land cover versus stream water classification for each of the 196 HUC-14 watersheds were compared as part of the New York–New Jersey Highlands Regional Study: 2002 Update. The New York and New Jersey classifications were compiled and ranked on a scale of 1 to 5. Surface waters ranked 4 and 5 represent the highest water quality, and should be protected from any measurable change to existing water quality (e.g. trout production or C1 in New Jersey; trout stream (Ts) or AA/A in New York). The surface water and stream segments within each HUC 14 basin were extracted and averaged to determine a composite value for each basin. Comparisons of average surface water rank and percentage of developed land cover in each HUC 14 show that highest quality surface waters (e.g., Class 3 or greater) do not occur in watersheds that are more than 50 percent altered (Figure 2-43). A Wilcoxon Rank Sum test shows a statistically significant (p < 0.0005) difference in percentage of altered land cover between those watersheds with a stream ranking of 3 or better versus those ranked less than 3. A similar analysis was conducted to compare the average surface water ranking and percentage of altered land (developed and agricultural land) by watershed. This analysis shows that agricultural land does not appear to have as much of an impact on surface water ranks, as do urban land cover types. Watersheds may contain high percentages of agricultural land and, therefore, higher proportions of altered land (e.g. 50-80 percent), and still are classed as rank 3 or better (Figure 2-44). The Wilcoxon rank sum test showed no statistical difference in this case.

The proportion of altered to unaltered land was calculated for each of the 51 HUC 11 watersheds for 1984, 1995, and 2000 by use of the New York–New Jersey Highlands land use and land cover database. It was necessary to establish a relationship between unaltered land area and housing unit density in order to estimate the amount of residual unaltered land under build-out. U.S. Census 2000 block group housing unit density data and the 2000 land cover map were used to establish the relationship for present day conditions. The census data were aggregated into housing density classes corresponding to commonly used municipal zoning densities. This housing unit density map was then cross tabulated with the land cover map to determine the percent area of each density class that was mapped as unaltered (e.g., forest and wetlands) land cover. Figure 2-45 shows that there is a smoothly asymptotic relationship of decreasing unaltered land with increasing housing unit density. Municipal zoning under the two build-out scenarios was used to estimate the housing unit density after build-out. Then, based on the observed relationship between housing unit density and unaltered land, the amount of unaltered land after build-out was estimated for each zoning area. The percent of unaltered land was then calculated on the HUC 11 watershed basis.

Figure 2-43. Surface water quality scores vs. percent developed land cover for HUC 14 sub-watersheds.

Figure 2-44. Surface water quality scores vs. percent altered land cover for HUC 14 sub-watersheds.

Figure 2-45. Graph of percent unaltered land cover vs. housing unit density (hu/acre).

Impervious Surface Cover

Impervious surface (e.g., asphalt, concrete, buildings, road surfaces) is an important environmental indicator of the intensity of human land use, and closely correlates with water quality degradation and altered runoff patterns in urban and urbanizing areas (Novotny and Chesters, 1981; Driver and Troutman, 1989; Arnold and Gibbons, 1996; Bolstad and Swank, 1997; Charbeneau and Barrett, 1998; Wang 2001). As forests, wetlands, grasslands and to a lesser extent cultivated land is paved over with impervious surface, groundwater recharge is lessened, resulting in greater surface runoff, higher storm runoff peaks and downstream flooding (Brown, 1988; Ferguson and Suckling, 1990). Impervious surface cover is increasingly used as a landscape-level indicator of non-point source pollution and watershed health. Arnold and Gibbons (1996) compared data from several studies and found that at 10 percent impervious surface cover, water quality begins to show signs of impact. Water quality is often seriously degraded at more than 25-30 percent impervious surface cover.

Stream water quality and the percentage of impervious surface cover were compared on a HUC 14 watershed basis for the New Jersey portion of the study area. The New Jersey classifications were ranked on a scale of 1 to 5. Surface waters ranked 4 and 5 represent the highest water quality, and should be protected from any measurable change to existing water quality (e.g. trout production or C1 in New Jersey). The surface water and stream segments within each HUC 14 basin were extracted and averaged to determine a composite value for each basin. Comparison of average surface water ranks and the percentage of impervious surface cover showed that those basins with water quality rank 4 (i.e., class C1/Tp, the highest water quality) generally experienced less than or equal to 10 percent impervious cover (except in one instance) (Figure 2-46). In other words, the overall stream water quality decreased as impervious surface cover increased above 10 percent. The Wilcoxon Rank Sum test showed that there was no statistically significant difference in the percentage of impervious surface cover between basins with surface water quality rank 4 or better than in those ranked below 4. Further examination showed that those basins with a surface water quality rank 3 or better generally contained less than 15 percent impervious cover. The Wilcoxon rank sum test showed that there was a statistically significant difference (p < 0.0001) in the impervious surface cover between basins with a surface water quality rank of 3 or better versus those ranked less than 3. We chose an impervious surface cover threshold of 10 percent based on our interest in protecting existing high quality water (i.e., Rank 4 or above).

The proportion of impervious surface was calculated for each of the 51 HUC 11 watersheds for 1984, 1995, and 2000 by use of the New York–New Jersey Highlands land use and land cover database. The 1995 NJ DEP land use data included estimates of impervious surface cover for each land use polygon. These data were rasterized to match the Landsat TM grid for the New Jersey Highlands study area. The mean impervious surface cover was determined for each of the Landsat TM spectral training classes used to map developed land cover for 1995 based on these data. The impervious surface cover was then estimated across the entire New York–New Jersey Highlands study area based solely on the Landsat TM data. The mean impervious surface cover for the NJ DEP data set were regressed against the Landsat TM data from a large sample of polygons for each of 13 different land use classes as a validation check of the utility of this relationship. The regression showed a good fit with a coefficient of determination (R²) value of 0.91. This same relationship was then applied to the 1984 and 2000 data sets.

In order to estimate the amount of impervious surface under build-out, it was necessary to establish a relationship between impervious surface cover and housing unit density. U.S. Census 2000 block group housing unit density data and the 2000 impervious surface map were used to establish the relationship for present day conditions. The census data were aggregated to housing density classes that correspond to commonly used municipal zoning densities. This housing unit density and impervious surface maps were cross-tabulated to determine the average impervious surface cover of each density class. Figure 2-47 shows the relationship between percent impervious surface cover and housing unit density. The percentage of impervious surface increases in a nonlinear fashion to a housing density of approximately 1 unit/acre, and then continues to increase at a more linear pace. The municipal zoning under the two build-out scenarios was used to estimate the housing unit density after build-out. The amount of impervious surface after build-out was then estimated for each zoning area based on the observed relationship between housing unit density and impervious surface. The percentage of impervious surface was then calculated for each HUC 11 watershed.

Figure 2-46. Surface water quality scores vs. percent impervious surface cover for HUC 14 sub-watersheds.

Figure 2-47. Graph of percent impervious surface cover vs. housing unit density (hu/acre).

Riparian Corridors

Protected buffer strips where human development is excluded or minimized is a “best management practice” that is often advocated as a means to reduce the impact of human developed land uses on adjacent riparian areas and downstream water quality (NJ DEP, 1999; Zampella et al., 1994; Muscutt, 1993; Welsch, 1991). Protected riparian buffers serve as vital habitat for both upland and wetland-dependent species in addition to reducing non-point source pollution.

Riparian areas were defined as those areas that are adjacent or hydrologically connected to the surface water network (e.g., streams, rivers, lakes or reservoirs). Riparian areas of the Highlands were delineated as those areas that are adjacent to the stream corridor and classed as either: 1) 100 year floodplain as mapped by the Federal Emergency Management Agency; 2) wetlands as mapped by the US Fish & Wildlife Service National Wetland Inventory or NJ DEP; or 3) hydric soils as mapped by the USDA National Resources Conservation Service based on the New York–New Jersey Highlands GIS database. Isolated wetland areas (i.e., those not adjacent to a stream corridor) were excluded from this analysis. Freshwater streams and rivers were extracted from the USGS 1:24,000 scale digital GIS data set. A 90 meter buffer on both sides of all mapped streams and rivers was delineated to create a 180 meter wide riparian corridor and included as riparian area. The relative percentage of the riparian that is in an altered land cover (e.g., developed, cultivated/grassland or barren) was calculated. The alteration of the riparian zones was then summarized on a HUC 11 watershed basis. The proportion of the riparian zones in altered land cover was estimated based on the amount of altered land calculated under the various build-out scenarios.

Forest Fragmentation

The conservation of large tracts of contiguous forest habitat and the minimization of fragmentation were identified as major issues of concern in the New York–New Jersey Highlands study region. Large contiguous tracts of forest and wetland that are not fragmented by human development are especially valuable as wildlife habitat and recreational open space, as well as watershed protection. Human development has the direct impact of removing existing natural habitat as well as fragmenting the habitat that remains into smaller pieces. Paved roads, residential and commercial development often serve as barriers or hazards to wildlife movement and native plant dispersal, as well as altering “natural” disturbance regimes. Human development also has "indirect" impact by creating a number of different kinds of intrusions with varying depth of impact into adjacent natural habitat and recreational open spaces. These intrusions include increased air, water, noise and light pollution; changes in microclimatic conditions due to higher sunlight and wind levels; increased populations of invasive "weed" species; and increased frequency of disturbance due to direct contact with humans, human pets and associated "rural/suburban pest" species. The border area affected by these disturbances is labeled edge, as compared to the undisturbed interior habitat (Zipperer, 1993).

A number of passerine songbirds such as warblers and vireos, that breed and forage in forested uplands and wetlands, are particularly associated with forest interior rather than edge habitat. A recent decline in the breeding populations of these migratory songbirds has been linked to the effects of forest fragmentation (Bohning-gaese et al., 1993; Robinson et al., 1995). Fragmentation has led to isolation of interior forest habitat (Whitcomb, 1977; Butcher et al., 1981; Blake and Karr, 1984) and increased pressure by nest predators (Wilcove, 1985) and brood parasitism by cowbirds, which show higher frequency closer to the forest edges (Brittingham and Temple, 1983). In addition to forest interior nesting songbirds, there are a number of other so-called area-sensitive species that depend on large tracts of undisturbed interior habitat to maintain viable populations. Large raptors such as red-shouldered hawks and barred owls are area-sensitive species that require large blocks of mature forested wetlands and adjacent upland forest.

Many characteristic Highlands amphibian and reptiles are sensitive to habitat fragmentation and human disturbance through a variety of mechanisms. Slow moving amphibians and reptiles are especially susceptible to road-kill and are therefore impacted by increasing densities of roads and traffic volumes (Mitchell, 1992). Timber rattlesnakes, a species of particular concern in the Highlands, are considered a restricted range species because they rely on winter denning sites in rocky talus areas. During the periods immediately before and after hibernation, the snakes congregate around these sites, making them particularly susceptible to human disturbance (Brown, 1993). In addition, there are a number of area-sensitive mammal species that rely on large contiguous tracts of high quality habitat (Mitchell, 1992). Bobcats, a State threatened species in New Jersey, require comparatively large home ranges of relatively intact forest area. River otters, though not considered threatened or endangered, are rare and receive special management consideration in New York. Black bears are also sensitive to human disturbance, especially during the winter denning months (Goodrich and Berger, 1994).

The land use and land cover change data set was used to examine the issue of forest fragmentation in several ways. One approach examined was to delineate contiguous tracts (i.e. patches) of forest (upland and wetland forest combined). Major roads (i.e., county 500 and 600 level highways and higher) were included in the analysis as a fragmenting influence or barrier; such that a tract of forest that might otherwise be considered contiguous, if it were subdivided by a major road would be mapped as two separate parcels. While some of the smaller tracts may represent a single ownership, most of the larger tracts will be composed of multiple ownerships, both public and private. Based on the size distribution, the contiguous forest tracts were then broken into size classes: 1) 25-100 acres; 100-500 acres; 3) 500-1000 acres; 4) 1000-5000 acres; and 5) greater than 5000 acres. This analysis was conducted for the year 2000 time period.

Due to the importance of forest interior habitat in the New York–New Jersey Highlands, we used GIS analysis to examine the present status and recent trends of this key habitat type. Forest interior habitat was mapped using similar methods to those employed as part of the New Jersey Endangered & Nongame Species Program Landscape Project (Niles et al., 1999). All upland and wetland forest types were combined to create a simple binary forest and non-forest map. Using a GIS spatial analysis procedure called buffering, a 90 meter buffer was delineated inside the boundary of every area of forest habitat patch to exclude this edge zone and leave only interior forest habitat. Contiguous forest tracts smaller than 25 acres in size (10 hectares) were further excluded. This interior forest analysis was conducted for the 2000, 1995 and 1984 time periods. Due to the great difference in spatial resolution of the 1972 data, this time period was not examined.

Three general ways of calculating indicators of forest fragmentation at the watershed scale were considered: 1) a patch centered approach; 2) the area of interior forest habitat approach; and 3) forest cover percentage of Breeding Bird Atlas blocks. The patch-centered method calculates the percent of the watershed that is composed of the largest single patch of contiguous forest (Wickham and Norton, 1994; Wickham et al., 1999). While this measure is useful in characterizing existing patterns of forest fragmentation as is evident in the 2000, 1995 and 1984 time series of land use and land cover data, it is limited for estimating future fragmentation under build-out. This method relies on a “wall-to-wall” map of forest cover and the spatial configuration of forest patches. Under the build-out scenarios, while the amount of residual interior forest cover can be reasonably estimated, the future spatial configuration of the forest patches cannot. Based on this need for estimating forest fragmentation under both existing conditions and at build-out, the amount of interior forest per watershed was selected as the primary environmental indicator of fragmentation.

A final method of characterizing forest integrity was an analysis of the Highlands forest cover at the scale of the Breeding Bird Atlas blocks (NJ Audubon, 2002). The entire State of New Jersey was subdivided into atlas blocks that were approximately 10 square miles in size. Each block was thoroughly surveyed and observers noted the breeding status of all birds in the area during the years of 1994 to 1997. New York State was similarly surveyed during the years of 1980-1985 (NYS DEC, 2002). Due to the more contemporaneous nature of the New Jersey data, these survey results were closely examined. The atlas survey block grid was overlaid on top of the New York–New Jersey Highlands land cover data for 1995 and the percent of forest cover was determined for each block. Based on input from the NJ DEP Endangered & Nongame species program and the New York Natural Heritage Program, a list of forest interior dependent raptors and neotropical migrant birds was enumerated (Table 2-34).

Table 2-34. List of selected forest interior breeding birds.

Forest Interior raptors

Barred owl (Strix varia)

Northern goshawk (Accipiter gentilis)

Red-shouldered hawk (Buteo lineatus)

Forest Interior Neotropical Migrants

Acadian flycatcher (Empidonax virescens)

Veery (Catharus fuscescens)

Black & white warbler (Mniotilta varia)

Worm-eating warbler (Helmitheros vermivorus)

Cerulean warbler (Dendroica cerulea)

Ovenbird (Seirus aurocapillus)

Louisiana waterthrush (Seirus motacilla)

Kentucky warbler (Oporornis formosus)

Hooded warbler (Wilsonia citrina)

American redstart (Setophaga ruticilla)

Scarlet tanager (Piranga olivacea)

The number of forest interior raptors and neotropical migrants observed to be breeding was plotted vs. the percent forest area for each atlas block (Figure 2-48). Based on this data set several thresholds were selected. Analysis of the 1995 New Jersey Breeding Atlas data in relation to the Highlands land use and land cover data revealed that there were some thresholds were evident: 1) at less than 70 percent forest cover, the number of observed forest interior nesting neotropical migrant bird species was quite variable, above 70 percent forest cover all the blocks had 7 or more species; 2) at less than 70 percent forest cover, the number of blocks with the full suite of forest interior raptors was uncommon, while above 70 percent forest blocks more commonly contained all three raptors; 3) at less than 25 percent forest cover, there was a notable drop in bird diversity with no blocks containing the full suite of 10 species; and 4) at less than 25 percent forest, the number of forest interior raptors were generally less than two.

In order to estimate the amount of interior forest habitat under build-out, it was necessary to establish a relationship between interior forest area and housing unit density. U.S. Census year 2000 block group housing unit density data and the year 2000 interior forest area map was used to establish the relationship for present day conditions. The census data were aggregated into housing density classes corresponding to commonly used municipal zoning densities. To minimize the bias introduced by other land covers (i.e., not forest or classes related to developed land uses), water, emergent wetlands, cultivated and barren land cover areas were removed, leaving only upland and wetland forest, developed, and grass land covers. This housing unit density map was then cross-tabulated with the forest interior map to determine the percent area of each density class that was mapped as interior forest. Figure 2-49 shows that there is a smoothly asymptotic relationship of decreasing interior forest habitat with increasing housing unit density. A threshold of 40 percent of the watershed as interior forest was selected based on our comparison of the present day distribution of interior forest and the Breeding Bird Atlas block forest cover data.

Figure 2-48. Forest interior dependent bird species richness vs. the percent forest area for Breeding Bird Atlas data blocks.

Figure 2-49. Percent forest and interior forest vs. housing unit density (hu/acre).

Results

The land use and land cover analysis served as the basis for the development of several landscape indicators of forest integrity and watershed condition: 1) unaltered land cover (percent forest and wetland); 2) percent cover of impervious surface; 3) percent of the riparian buffer of permanent streams that is in a vegetated, as compared to developed or altered state; and 4) indices of forest fragmentation. These indicators were computed on a percentage basis for each of the 51 HUC 11 watershed basins within the New York–New Jersey Highlands study area. The landscape indicators were mapped for the years 1984, 1995 and 2000 to monitor region-wide trends. These indicators were also estimated for high- and low-constraint build-out scenarios to evaluate potential future conditions.

The land cover change analysis shows a general trend towards increasing altered land cover during the 1980’s and 1990’s (Figure 2-50). The number of watersheds with greater than 50 percent altered land went from 12 in 1984, to 13 in 1995 to 17 in 2000 (i.e., from 24 to 25 to 33 percent of the watersheds). Depending on the build-out scenario, the number of watersheds expected to have greater than 50 percent altered land cover could more than double. Under the high-constraint scenario, the number of watersheds with greater than 50 percent altered would be 24, and under the low-constraint scenario the number would increase to 36 (Figures 2-51a, b). Under the high-constraint scenario, most of the predicted increase would be less than 10 percent change as compared to 2000 (Figure 2-51c). Under the low-constraint scenario, a significant number of watersheds (i.e. 10 out of 51 or nearly 20 percent) are expected to show a greater than 20 percent change over 2000 conditions. This increase in altered land under the two build-out scenarios, and especially under the low-constraint scenario, indicates that threats to Highlands water quality would be expected to increase.

The land cover change analysis shows a general trend towards increasing impervious surface cover. The number of watersheds with more than 10 percent impervious surface cover increased from 6 in 1984, to 9 in 1995 and 2000 (Figure 2-52). While the number of watersheds with greater than 10 percent impervious cover stayed stable between 1995 and 2000, the number of watersheds with between 5 and 10 percent impervious surface cover increased from 20 in 1995 to 24 in 2000. Depending on the build-out scenario, the number of watersheds expected to have greater than 10 percent impervious surface cover could more than triple to quadruple. Under the high-constraint scenario, the number of watersheds with greater than 10 percent impervious surface cover would be 33 (i.e., 33 out of 51 or 65 percent), and under the low-constraint scenario the number would increase to 41 (i.e., 41 out of 51 or 80 percent) (Figure 2-53a, b). Under both scenarios, most of the predicted increase would be less than a 10 percent change as compared to 2000 (Figure 2-53c, d). Under the low-constraint scenario, a significant number of watersheds (i.e. 23 out of 51 or 45 percent) are expected to show a 5 to 10 percent change over 2000 conditions. This increase in impervious surface cover indicates that negative impacts to Highlands water quality would be expected to increase.

The analysis shows that alteration of riparian zones has not dramatically increased between 1984 and 2000. The number of watersheds with greater than 50 percent altered riparian zones has remained stable at 5 to 7 while the number between 25 and 50 percent altered zones has increased slightly, from 25 in 1984 to 26 in 1995 to 32 watersheds in 2000 (Figure 2-54). The two build-out scenarios show a very different response in relation to riparian zone protection. The low-constraint scenario is expected to show a large increase in riparian zone development and alteration with the number of watersheds with greater than 50 percent altered land going to 24 (i.e. 24 out of 51 or 44 percent) (Figure 2-55b). Of special notice is the large predicted increases in those watersheds that feed the Croton Reservoir system (Figure 2-55d). The high-constraint scenario (which incorporates wide wetland buffer distances) remains largely unchanged from the present situation with 10 watersheds at greater than 50 percent altered (Figure 2-55a). The results of the high-constraint build-out scenario suggests that increasing the wetland buffer distance will help to protect sensitive riparian zones (and thereby surface water quality) even in the face of increasing development.

Two parameters were analyzed as an indicator of forest integrity: 1) the amount of interior or core forest habitat (i.e., the forest that is unfragmented with minimal “edge”) in each watershed basin; and 2) the percent of overall forest cover by Breeding Bird Atlas survey block. The land use and land cover analysis shows that the amount of overall forest and the unfragmented interior forest is decreasing. The interior forest indicator shows a steady decline from 15 (i.e., 15 of 51 or 29 percent) of the watersheds having greater than 40 percent interior forest cover in 1984 to only 9 (i.e., 9 out of 51 or 18 percent) of the watersheds in 2000 (Figure 2-56). The number of watersheds with more than 40 percent interior forest is expected to decrease to 7 and 4 under the high- and low-constraints respectively (Figure 2-57). The low-constraint build-out scenario shows over a third of the watersheds (i.e. 19 out of 51 watersheds or 37 percent) with greater than 10 percent decline in interior forest area the northeastern corner of the New York Highlands region and the central Highlands of New Jersey show up as a hotspots for interior forest loss.

The number of prime interior forest habitat blocks similarly declines. In the year 2000, less than 22 percent of the New York–New Jersey Highlands region was considered ‘prime’ habitat for forest interior nesting birds or raptors. Under the various build-out scenarios, the amount of prime forest habitat is expected to further decline. Under the low-constraint scenario, where the remaining Pequannock Watershed lands are opened to future development and limited protection of steep slopes, the number of prime habitat blocks decreases further such that only 13 percent of the New York–New Jersey Highlands region was considered ‘prime’ habitat for forest interior nesting birds or raptors (Figure 2-57). The high-constraint scenario shows a more modest decline with 20 percent of the New York–New Jersey Highlands region considered prime habitat (Figure 2-58).

Figure 2-50. Percent altered land cover for New York–New Jersey Highlands HUC 11 watersheds: 1972, 1984, 1995, and 2000.

Figure 2-51. Estimated changes in percent altered land at build-out.

Figure 2-52. Percent impervious cover for New York–New Jersey Highlands HUC 11 watersheds: 1984, 1995, and 2000.

Figure 2-53. Estimated changes in percent impervious surface at build-out.

Figure 2-54. Percent altered riparian zones for New York–New Jersey Highlands HUC 11 watersheds: 1984, 1995, and 2000.

Figure 2-55. Estimated changes in percent altered riparian zones at build-out.

Figure 2-56. Percent interior forest for New York–New Jersey Highlands HUC 11 watersheds: 1984, 1995, and 2000.

Figure 2-57. Estimated changes in percent interior forest at build-out.

Figure 2-58. Estimated changes in prime forest habitat at build-out.

Discussion

The coupled build-out-landscape indicator analysis was designed to serve as a planning tool to provide a way to assess the potential impacts to forest and watershed integrity based on “what if” scenarios. This analysis is not an “absolute” prediction of future conditions at any particular point in time. Rather it suggests what might be expected to happen based on existing patterns and trends, and under the various assumptions codified in each build-out scenario. This analysis was undertaken with two main objectives: 1) to compare the relative impact of the low- vs. high-constraint scenarios at the regional scale; and 2) to identify potential hotspots of environmental change that deserve greater consideration in future land management decisions.

The indicator analysis reveals that there has been a steady loss of forest cover and an increase in human altered land and impervious surface cover. The landscape indicators highlight future hotspots of change such as the intensification of human development impacts in the northeastern corner of the Highlands in New York State when coupled with the build-out analysis. The low-constraint scenario, which simulates the status quo in terms of zoning and environmental regulations, shows large land cover and landscape indicator change, signaling significant environmental degradation. The high-constraint scenario, which incorporates greater wetland and riparian zone buffers, limited development on steep slopes, and no development of watershed management lands, shows a more limited change in landscape indicators and a lower degree of associated environmental degradation.

Continued development of the Highlands is inevitable and in some locations welcomed. However, greater attention must be paid to protecting the integrity of the Highlands forests and watersheds. The Highlands native forests and wetlands serve as natural water filtration systems, protecting the integrity of the Highlands’ vital water supplies. The larger forest tracts that form the core of the Highlands serve to protect the integrity of Highlands’ natural communities by reducing human-induced edge effects; protecting area-sensitive native animals; and helping to minimize conflicts between forest land management practices and adjacent human development. Smaller, fragmented forest patches may still have significant conservation value, especially in urban and suburban areas where they may serve as the only available wildlife habitat over large areas.

Resources at Risk

Objectives

The adequacy of present conservation management and ownership in protecting these critical and sensitive conservation areas was evaluated using several approaches. The first approach was to examine the efficacy of zoning as a land conservation management technique by measuring the extent that large lot zoning was more closely associated with the high conservation values assessment (CVA) ranking. The second approach was to identify areas that have important conservation values (i.e. high CVA scores as identified above) but that are not presently protected by the existing network of publicly owned open space and conservation lands. The third approach was to identify those areas with high conservation value that may have a higher likelihood of change in the near-term.

Methods

To examine the efficacy of zoning as a land conservation management technique, the CVA model results were geographically cross-referenced with the zoning map used in the build-out scenarios. To a certain extent, large lot zoning (i.e., down-zoning) is considered to enhance the conservation of the natural resources and limit environmental degradation and is a common tool used by local planning boards. We used the spatial analytical capabilities of the GIS to geographically cross-reference the zoning map with the CVA map and summarize the information accordingly.

For the second objective of identifying those areas that have important conservation values but that are not presently in some form of conservation ownership, we used the spatial analytical capabilities of the GIS to geographically cross-reference the existing framework of publicly owned open space with the CVA map and summarize the information accordingly. As this approach attempts to identify the gaps in conservation protection, it has been labeled a gap analysis (Scott et al., 1993). Using the GIS, the existing publicly owned lands are “masked out,” and only the remaining lands are visible, thereby highlighting gaps in existing conservation protection. Major clusters and large contiguous tracts of high CVA ‘gap’ areas (CVA combined rank of 4 or 5) were identified as ‘conservation focal areas’ for special consideration in open space protection. These regional focal areas include high CVA lands that serve to connect existing publicly or privately owned conservation lands into larger local n