Developing ecological endpoints for valuation of semi-arid riparian ecosystem services

Creating measurable ecological accounting units has become a point of emphasis in valuing ecosystem services. Understanding which ecological endpoints, which emanate from biophysical production functions, are important to individuals could help to create measurable ecological accounting units. Using two semi-arid riparian ecosystems we create a suite of ecological endpoints and using benefits transfer techniques compare their ability to be transferred to similar riparian ecosystems. If clearly defined, ecological accounting units can be developed for ecosystem services. This could lead to ecosystem services being properly incorporated into benefit cost analyses that maximize economic product of both market and non-market goods and services.


Introduction
Ecosystems throughout the world provide mankind with a variety of marketable and non-marketable goods and services that provide life support systems. Mankind's use of natural ecosystems has sparked a debate in the literature on how to quantify and account for these systems in economic terms. Daily (1997) introduced ecosystem services as "the benefits of nature to households, communities, and economies." Despite the importance of ecosystem services to mankind, they are often overlooked or taken for granted in typical benefit cost analyses (Bingham et al. 1995;Heal 2000). As an example, the Millennium Ecosystem Assessment reported that 60% of global ecosystem services are being degraded or used unsustainably (Millennium Ecosystem Assessment 2005).
One reason for this degradation may be the lack of clearly defined ecological accounting units, making it difficult to incorporate the true value of ecosystem services into a benefit cost analysis. Costanza et al. (2017) point out that "until recently, policy proposals such as road construction and land use changes were formally evaluated based on a narrow financial cost-benefit analysis only, in which the costs of development as well as the benefits recognized in the market were included. Informally, other values often played a role as well, but in most of these cost-benefit analyses, costs of loss and benefits of conservation of non-market benefits, as most regulating ecosystem services produce, were ignored." (8). We hypothesize that clearly defining ecological accounting units will create a system of ecosystem service valuation that leads to value inclusion in benefit cost analyses going forward.
As proposed by the Millennium Ecosystem Assessment, ecosystem services can be categorized into four broad types: 1) provisioning services, 2) regulating services, 3) cultural services and 4) supporting services. While it is helpful to categorize ecosystem services, it is problematic as they are complex and as Costanza et al. (2017) explain, ecosystems are multidimensional and the distinction between direct and indirect contributions to human wellbeing is complex and not completely understood. Further, as de Groot et al. (2010) point out, valuing ecosystem services is a key area that is in need of further research. How these valuation studies are to be conducted is a current topic of debate in the literature. Boyd and Banzhaf (2007) have advocated for clearly defined ecological accounting units that focus on final services and not ecosystem processes. Their definition is that "final ecosystem services are components of nature, directly enjoyed, consumed, or used to yield human well-being." They list three key components in defining ecosystem services: 1) they are directly enjoyed consumed or used creating end-products to the users; 2) services are ecological characteristics not ecosystem functions or processes; and 3) the quantity of the service differs from the value of the service. This is a more formal definition than Daily (1997), but quantifying ecological accounting units has proved problematic in the literature. Muller and Burkhard (2012) demonstrate how ecosystem functions are necessary components in deriving ecosystem services, which ultimately lead to human wellbeing. However, as de Groot et al. (2012) point out, no one method works in assessing ecosystem tradeoffs for valuation. Further, Mart ınez-Harms and Balvanera (2012) argue for a standardized approach similar to Boyd and Banzhaf (2007) while Crossman et al. (2013) provide an example of a blueprint model in an effort to standardize ecological accounting units. A conclusion of Crossman et al. (2013) is that no one approach can do it all, which is the same argument as de Groot et al. (2012). A further example is Barkmann et al. (2008) identifying a set of ecological endpoints from science models to value ecosystem services. They find that while this approach is beneficial it comes with a cost, as "basic science" models that explain ecosystem processes and functions are difficult for participants to understand. Boyd and Krupnick (2013) have worked to further expand the definition of ecological endpoints as " … meaningful biophysical production functions to determine their economic value." But as Boyd and Banzhaf (2007) point out, what are the units that are to be defined in order to track and measure ecosystem services?
This research builds upon these previous studies as we seek to address the question, can clearly defined ecosystem endpoints be developed and transferred between similar sites for two river systems? Working off the Boyd and Banzhaf (2007) criteria we characterize ecosystem services for two study sites in the southwestern US as "final" goods and not intermediate goods that are directly enjoyed by users. Further, we recognize that ecological processes transform biophysical conditions into outputs enjoyed by users (Boyd and Krupnick 2013). As such, we work in harmony with scientific modelers to define the quantities of ecosystem processes that are used to create the ecological endpoints. Our approach follows Pandeya et al.'s (2016) argument that ecosystem services should be valued at the local scale, as their benefits arise predominantly at the local scale. In addition, we integrate ecological process models into the valuation exercise, similar to Barkmann et al. (2008) in developing and defining ecological endpoints that are similar across two riparian ecosystems, the San Pedro Riparian National Conservation Area (SPRNCA) and Middle Rio Grande, NM (MRG). In both systems, ecological endpoints are grounded in ecological process models to create our ecological endpoints that are a true representation of the ecological system and not based on hypothetical outcomes. We developed ecological endpoints for two riparian ecosystems so that a benefits transfer can be carried out having similar biophysical units of measurement across regions.
With a growing stream of literature focusing on Benefits Transfer (BT) approaches, having clearly defined ecological accounting units that can be used in BT approaches could ensure more accuracy in the valuation of ecosystem services. Our approach is to employ a basic value transfer, as defined by Kubiszewski et al. (2013). The need for this type of approach is highlighted by Johnston and Rosenberger (2010) where, in reviewing the BT literature, they found a lack of studies provide high quality, policy-relevant, replicable, empirical estimates of non-market values for many environmental commodities. Recognition of this problem stems from Boyd and Banzhaf 's (2007) insight that clearly defined ecological accounting units do not currently exist.
As highlighted throughout the literature, no one method of quantification is perfect. As Mart ınez-Harms and Balvanera (2012) and Crossman et al. (2013) have argued, it is necessary to create a set of ecological indicators that can inform ecosystem service values. We seek to develop these indicators for a local scale ecosystem, as Pandeya et al. (2016) argue is important. These indicators have the same accounting units across the two study sites allowing for a BT to be conducted which could aid in overcoming some of the challenges raised by Johnston and Rosenberger (2010). This paper proceeds as follows. First, we detail our study design and the elicitation format for valuation estimation. Second, we explain our estimation procedure and valuation calculations. Third, we present the results with a discussion of the findings. We conclude with some reflections on the implications of this research to define quantifiable and measurable units for ecosystem services.

Study design and elicitation format
A traditional split sample approach was employed to elicit Marginal Willingness To Pay (MWTP) values for ecological endpoints for two riparian ecosystems in the southwestern US: 1) the San Pedro Riparian National Conservation Area (SPRNCA) located in southeastern Arizona, and 2) the Middle Rio Grande (MRG) located in central New Mexico. Ecological endpoints for water resource and vegetation management were developed for these two systems with the goal of quantifying metrics that are transferable across semi-arid riparian ecosystems. This approach builds upon the Boyd and Banzhaf (2007) and Boyd and Krupnick (2013) approaches that ecological endpoints should be grounded in biophysical models. Upon identifying the biophysical inputs and processes for the two river systems from natural scientists we conducted two sets of focus groups to aid in developing the ecological endpoints used in our valuation exercise.
The first set of focus groups presented participants with a scientific model to define the SPRNCA (shown in Figure 1) and the MRG (shown in Figure 2). A focus group facilitator led a group discussion to obtain written and oral feedback about the inputs and processes they deemed important. These processes included the average cubic feet per second at river guages, depth to groundwater for each river section, and the average length of wetted stream over the year. Vegetation features such as the type of vegetative species, root depth for species and which species were native versus  non-native and invasive. The wildlife supported by the systems focused up on avian species. Focus groups were presented with types of breeding birds, vegetation required for breeding purposes and the number of migratory and type of migratory bird species.
Feedback from the first set of focus groups provided insight into which biophysical processes were important as we worked to quantify the ecological endpoints based upon ecological production theory (Boyd and Krupnick 2013). Focus group feedback resulted in our characterization of the ecosystems into three endpoints: 1) average miles of surface water flow, 2) composition of riparian vegetation, and 3) abundance of avian species by breeding habitat and surface water dependency. In addition, the average number of migratory birds was deemed an important endpoint and included in the survey. A second set of focus groups was conducted to gather feedback on the presentation and organization of an educational component to inform survey participants about "basic science" models. A focus group facilitator led the discussion to gather information about graphics and tables to determine which presentation format was desirable. During this set of focus groups, feedback was solicited on the choice question design to present the ecological endpoints finding the preferred design to be a choice question card with pictorial summaries of the ecosystems, rather than tables and charts. The following sections provide background information on the two study sites with a description of how the ecosystem inputs and processes were presented to focus group participants in order to develop the ecological endpoints. Following this description, the choice experimental format, our elicitation format, is presented along with a summary of the final survey design and implementation procedures.

San Pedro Riparian National Conservation Area (SPRNCA)
Flowing northward across the US Mexico border, the San Pedro River cuts through the desert of southeastern Arizona. On 18 November 1988, the US Congress designated 40 miles of the upper San Pedro as a Riparian National Conservation Area (SPRNCA). The primary purpose for this designation was to protect and enhance a rare remnant of what was once a network of similar desert riparian ecosystems throughout the southwestern US. The SPRNCA contains nearly 57,000 acres of public land and is home to 84 species of mammals, 14 species of fish, 41 species of reptiles and amphibians, and over 100 species of breeding birds (Tellman and Huckleberry 2009). In addition, the SPRNCA provides habitat for over 250 species of migrant and wintering birds traveling between temperate and tropical or sub-tropical regions (Tellman and Huckleberry 2009).
Extensive human use of the upper San Pedro River has led to groundwater depletion, potentially threatening the historic cottonwood-willow (Populus-Salix) woodlands and leading to shifts toward non-native Tamarix shrub lands on some sites (Stromberg 1998;Lite and Stromberg 2005). Formation of the SPRNCA has provided protection for the river and riparian zone and some passive restoration has resulted from the removal of cattle (Krueper, Bart, and Rich 2003), but threats surrounding water use still exist Noon 2010, Brand et al. 2011). In the SPRNCA, ground and surface water are in high demand for the competing uses of municipal, agricultural, industrial and riparian ecosystem services (Stromberg et al. 2006;Brand and Noon 2011). Changes in the hydrology of the SPRNCA can end up in resultant changes in physiognomy, abundance and the composition of riparian vegetation (Pettit, Froend, and Davies 2001;Shaw and Cooper 2008;Stromberg et al. 2006Stromberg et al. , 2009) that, in turn, affect bird abundance and bird species assemblages (Brand et al. 2011).
A research record over the last 20 years has led to substantial physical and biological science knowledge, enabling the development of a decision support system (see Sumer et al. 2006) that allows for the prediction of different future groundwater scenarios for the SPRNCA, as displayed in Figure 1. This allows for an understanding of the impact that policy decsions that affect groundwater could have upon the composition of vegetation and bird abundance throughout the SPRNCA. The basic premise is that a change in groundwater availability creates a change in the composition of vegetation which leads to a change in the abundance of breeding and migratory birds throughout the SPRNCA (Brand et al. 2011). Nine policy scenarios were developed using input from the Upper San Pedro Partnership (USPP) which is a consortium of twenty-one non-governmental organizations, a private water company and local, state and federal agencies (Richter et al. 2009).
Of the nine constructed scenarios (see Figure 1), scenarios 1-3 illustrate the impacts of a uniform change in groundwater levels from current conditions through all sections of the SPRNCA. Scenarios 4-7 illustrate the impacts of spatially variable changes in groundwater levels, as defined in Figure 1. Scenarios 8 and 9 represent extreme possible outcomes that are achieved through large changes in population growth rates and groundwater pumping rates. More detail on the development of these scenarios can be found in Brookshire et al. (2010).
In each scenario, the conditions of each section of the river are described as either a wet, intermediate or dry condition class (Stromberg et al. 2006). The wet condition class is defined as a section of the river that has near perennial surface water flows with the major vegetative species being a cottonwood-willow forest where the depth to ground water is less than 2.5 meters. In the intermediate condition class surface water flows are intermittent; roughly 60-90% of the year flows are present, with a mixture of cottonwood-willow and saltcedar (Tamarix sp.) being present. Depth to ground water is between 2.5-3.5 meters. The dry condition class has surface water flow less than half of the year being dominated by Tamarix with isolated cottonwood snags and depth to groundwater is greater than 3.5 meters. This categorization of river conditions through these condition classes allows for the quantification of two ecological endpoints: 1) presence of surface water flows and depth to groundwater, 2) composition of vegetation for a reach (i.e. cottonwood-willow, Tamarix or no-vegetation). A third ecological endpoint is employed to describe wildlife by way of avian species.
As part of the formation of the SPRNCA, cattle grazing in and around the floodplain was prohibited, resulting in a significant increase in herbaceous riparian vegetation. Krueper, Bart, and Rich (2003) investigated how this change in herbaceous cover impacted avian species finding dramatic increases in the abundance of breeding and migratory bird species in the years that followed the removal of cattle from the SPRNCA floodplain. We describe breeding birds in the SPRNCA in two ways: 1) by nest height and 2) by surface water dependency. In total three main ecological endpoints are utilized to describe current and future conditions of the SPRNCA: 1) presence of surface water flows and depth to groundwater, 2) vegetation composition, 3) abundance of breeding and migratory birds.

Middle Rio Grande (MRG)
Located in central New Mexico, the Middle Rio Grande (MRG) cuts through the desert landscape creating a riparian ecosystem approximately 150 river miles in length.
Historically this ecosystem was dominated by woodlands of native cottonwood (Populus deltoids subsp. wislizeii) and willow (Salix gooddingii) species (Crawford, Ellis, and Molles 1996). For millennia this ecosystem was sustained by naturally occurring overbank floods that provided habitat to a diverse wildlife (Najmi et al. 2005). In more recent decades, the Rio Grande's flow regime has become highly regulated through an engineering system that includes multiple dams and diversion channels, allowing for the invasion of dense exotic vegetation such as salt cedar (Tamarix spp.) and Russian Olive (Elaeagnus angustifolia), leading to fire becoming a dominant force in shaping this riparian ecosystem (Crawford and Grogan 2004). Recent restoration efforts in the MRG have focused upon mechanically thinning non-native understory vegetative species in an effort to reduce fire risk (USFS 2005).
Employing a previously developed model to a 128 km segment of the MRG that had twelve vegetation composition-structure types and five guilds of birds (canopy, midstory, understory, water-obligates and spring migrants) four management options within seven management scenarios applied across the MRG, as displayed in Figure 2, were created. These scenarios allow for an exploration of the impact that different clearing regimes of the woody understory along the MRG could have upon the composition of vegetation and bird abundance (Brand et al. 2013). The basic premise is that changing vegetation composition, through management options, alters the composition of vegetation and, in turn, changes the abundance of breeding and migratory bird guilds.
Sections of the MRG were classified into four different management options based upon active management activities. Management Option 1 (MO1) incorporates intensive mechanical clearing of all native and non-native understory to reduce the risk of fire. Management Option 2 (MO2) applies selective hand-thinning of the nonnative understory while retaining the native understory in an effort to reduce fire risk while maintaining native vegetation. Management Option 3 (MO3) involves no clearing activities in the next 5-10 years, creating a dense understory. Management Option 4 (MO4) represents the projected future composition of the MRG in the absence of active management, allowing for the natural succession process to occur.
Applying these management options across the MRG, seven hypothetical scenarios were developed in consultation with the Middle Rio Grande Conservancy District (see Figure 2). Scenarios 1-4 are a representation of applying each of the four management options across all reaches respectively. Scenarios 5-7 are three scenarios where a mixture of the management options are applied based on whether the reach is in a rural versus an urban region. For instance, Scenario 5 applies MO1 to the urban reaches 1 and 3-6, while MO2 is applied to the rural reaches of 2 and 7-10. Scenario 6 applies MO1 to the same urban reaches, while MO3 is applied to the rural reaches. Scenario 7 applies MO2 to the urban reaches, while MO3 is applied to the rural reaches. Unlike the SPRNCA, management in the MRG is primarily driven by potential mechanical alterations to the vegetative structure and composition rather than through ground water pumping alterations.
Ecological endpoints are defined for the MRG to mirror those developed for the SPRNCA, with the main difference being that depth to groundwater and the presence of surface water are not employed for the MRG endpoints, as these two endpoints are invariant across the future policy scenarios displayed in Figure 2. Vegetative composition is presented for each reach along with the abundance of breeding birds by nest height and surface water dependency. Creating a classification system for the SPRNCA and MRG with similar delineated ecological endpoints allows us to conduct a non-market valuation exercise for each ecosystem independently and then transfer the results of each independent study to the opposing site, effectively conducting a BT between the two sites with standardized ecological accounting units.

Choice Experimental format
To elicit values for the ecological endpoints a Choice Experimental (CE) model was employed. This method can be traced back to the postulate made by Lancaster (1966;1971) that utility is derived from the characteristics that a good possesses, rather than the good per se. The use of a CE allows a researcher to calculate marginal values for the attributes of a good, as the good is defined through these attributes and participants make choices over a bundle of these attributes (see Louviere, Hensher, and Swait 2000). This technique typically relies on asking individuals multiple questions where at least one of the attributes for the good varies. In so doing, marginal values for the good's attributes can be calculated.
The ecological endpoints developed for the SPRNCA and MRG were the attributes used to describe the two ecosystems shown in Figure 3. Within each of the three attributes, sub-classifications were employed to describe the categories. For instance, miles of surface water present is an average of surface water conditions across a river system, for the SPRNCA there is an average of 31 miles of surface water present over a year, for the MRG this attribute does not vary and, as such, it was not reported in the choice questions to the participants. For the second attribute, vegetative composition, each river system is characterized by condition classes. For the SPRNCA three condition classes were employed (wet, intermediate, dry) and the percentage of the total and number of acres for each class, along with the location of each class, was presented. For the MRG four condition classes were employed as management options and the percentage of the total and number of acres for each class, along with the location of each class, was presented. Avian species were described in three ways: 1) migratory species, reported as the average total number of migratory bird species over the migratory seasons, 2) breeding birds based on nest height, which was subdivided into three nest heights (Canopy, High Shrub, Low Shrub) and 3) breeding birds based on surface water dependency (i.e. surface water dependent birds and non-surface water dependent birds). The three ecological endpoints are displayed in Figure 3 depicting the current conditions for each system with the SPRNCA on the left and the MRG on the right.
One of the features of a CE is the ability to calculate Marginal Willingness to Pay (MWTP) for each attribute of the good rather than a mean value, as is employed in the traditional Contingent Valuation Method (CVM) for non-market goods and services. To do this, a CE presents multiple alternatives for the good in question, asking respondents to select their most preferred alternative from this set of alternatives. This process can then be repeated varying the attribute levels across the choice questions. For each ecosystem, four choice questions were developed with two choices (i.e. option A and option B) compared to the status quo (i.e. current conditions from Figure 3). In order to introduce variability in the attributes across participant choices, seven versions of the choice questions were developed for the SPRNCA and nine versions for the MRG. To calculate MWTP for the ecological endpoints it was necessary to include a cost variable for each choice. The payment vehicle (cost to the household) was a one-time payment to the America the Beautiful National Parks and Federal Recreational Lands Pass. The bid array for the cost variable was taken from Kirchhoff (1994) and applied to the choice questions. Table 1 presents the levels for each attribute that were used to seed the choice questions. As can be seen in Table 1, three levels for each attribute were employed to ensure non-linearity in the attributes. The bid array for the cost variable ranged from no payment up to $415 as developed by Kirchhoff (1994).

Survey design and implementation
To elicit responses to the CE questions, a mail and paper survey was developed for each river system. The surveys consisted of six main sections. The first section presented an introduction to the study site, (i.e. history and location) and an explanation of possible future scenarios. The second section presented a detailed explanation of each of the main attributes for the study site: 1) water availability, 2) vegetative composition and the linking of vegetation with water availability, 3) bird abundances presented as both breeding birds by nest height and by surface water dependency and migratory birds. The third section presented the current conditions for both study sites, as displayed in Figure 3. Section four provided an explanation of proposed water use programs for the SPRNCA and mechanical management options for the MRG, while the fifth section elicited responses to the four CE questions. Finally, section six collected demographic information for each respondent (e.g. gender, age, mean household income) 1 . Because the survey was a hypothetical exercise and participants did not actually pay for the proposed changes, we employed a cheap talk script similar to Cummings and Taylor (1999). As Broadbent (2014) has demonstrated, the use of a cheap talk script in a CE reduces MWTP estimates helping to overcome any concerns about hypothetical bias due to the hypothetical nature of the valuation exercise. The procedure outlined by Dillman (2000) was employed to implement internet and mail surveys to residents in Arizona for the SPRNCA and New Mexico for the MRG. A cluster sampling procedure was employed using zip codes in the states to create the sampling clusters. A representative sample based on these zip code clusters was purchased from a commercial sampling firm that produced 2,000 potential respondents for Arizona and New Mexico residents, respectively. Potential respondents were contacted four times over the course of eight weeks inviting them to participate in the survey. Each participant was provided with a web address and a unique username and password to access the survey along with a paper version of the survey which could be filled out and returned via standard US mail. Online responses for 155 individuals were received for the SPRNCA survey and 208 for the MRG. The paper version of the survey yielded 148 and 144 responses for the SPRNCA and MRG respectively. This led to a response rate of 15.15% and 20.1% for the SPRNCA and MRG surveys respectively.

Estimation procedure and willingness to pay estimates
In designing the choice questions for the SPRNCA and MRG, the procedure developed by Huber and Zwerina (1996) was followed. Huber and Zwerina (1996) identify four principles for efficient choice designs: 1) orthogonality, 2) level balance, 3) minimal overlap, and 4) utility balance. Any design that fulfills these four principles has a minimal D-error. Minimal D-error was accomplished through a computerized search approach developed by Kuhfeld (2005) for the statistical package SAS. 2 This resulted in seven choice sets with four choice questions per set for the SPRNCA and nine choice sets with four choice questions for the MRG.

Choice experimental theoretical model
The theoretical model for a CE is based on the notion that individuals make a single decision from three alternatives for the river system described by the ecological endpoints for the system. Because choices are not ordered, a random utility model is used for the i th individual choosing among n alternatives. Assume the i th individual's utility for choosing option n is given by: where V i n is the systematic portion of the utility function that is determined by the attributes of the river system, and e i n is the stochastic element. Making the assumption that V i n is linear in the parameters, the functional form of the utility function for alternative n is expressed as: whereX i n is a vector of characteristics for alternative n and the i th individual, c n is the coefficient for alternative n representing the mean of the distribution of unobserved effects on the random component and k n is a vector of coefficients representing the effect of the attributes for alternative n on utility. The probability that the i th individual chooses alternative n over alternative k is: where V i n is the systematic utility for alternative n for the i th individual and V i k is the systematic utility for alternative k for the i th individual with e i n and e i k being the stochastic elements.
The conventional econometric model employed in the literature is the conditional logit model. This model dates to McFadden (1974) when he proposed modeling expected utilities in terms of the attributes of the good rather than the attributes of respondents. We recognize the restrictive assumptions of the conditional logit model (i.e. independence of irrelevant alternatives and a restraint on preference heterogeneity) but choose to employ this model as it has been the conventional model providing accurate parameter estimators in the literature. To account for each participant answering four questions, clustered standard errors are estimated by participant for the SPRNCA and MRG respectively.

Marginal Willingness To Pay estimation
One of the salient features of the CE framework is the ability of a researcher to calculate MWTP for each sub-attribute 3 . In the case of the ecological endpoints, MWTP can be calculated for each category (i.e. vegetation composition, breeding birds by nest height and surface water dependency, water availability). This calculation is shown in Equation (4): where k i is the estimate for the different ecological endpoints and k cost is the estimate for the cost of the choice to a household. Standard errors are obtained using the bootstrap procedure developed by Krinsky and Robb (1986) where the estimated standard error of the mean WTP is the standard error of the estimated empirical distribution of the mean, as explained by Poe, Severance-Lossin, and Welsh (1994).
These MWTP estimates for each study site can be transferred to the other site, known as a policy site, using BT techniques. One of the primary requirements for a BT to be accurate is correspondence or similarity between sites (Loomis and Rosenberger 2006). Creating a suite of ecological endpoints for the two river systems that are similar fulfills the correspondence criteria, allowing for a transfer between the two sites that enables us to observe whether the MWTP estimates differ between the original estimates and the transferred estimates. We accomplish the BT using a unit value transfer technique, or the transfer of the set of MWTP (see Kubiszewski et al. 2013;Johnston and Rosenberger 2010;Rosenberger and Loomis 2000).
To compare whether the MWTP values statistically differ between the two sites a two sample t-test is conducted given in Equation (5): wherek sp are the MWTP estimates for the SPRNCA andk mrg are the MWTP estimates for the MRG. b se 2 sp and b se 2 mrg are the squared standard errors for the SPRNCA and MRG, respectively, using the Krinsky and Robb (1986) bootstrap procedure.

Results and discussion
Results of the conditional logit model can be found in Table 2 using the breeding birds by surface water dependency rather than by nest height as both variables describe the same ecological endpoint, creating a co-linearity problem in model estimation. In addition, breeding and migratory bird numbers found in Table 1 are adjusted to be in thousands of birds, while acres of vegetation are adjusted to be a percentage of the total to account for heteroscedasticity that was found in the unadjusted model. Conditional logit results for the SPRNCA are all statistically significant except for the miles of surface water present. In addition, in using percentages for the vegetative classes one of the categories needed to be dropped due to multicollinearity, the dry vegetative condition was omitted as it was the smallest percent of acreage. Results for the MRG found that breeding birds by water dependency are statistically significant with MO#1 being the only vegetative class that is significant, MO#4 was omitted due to multicollinearity.
We chose to use breeding birds by surface water dependency to represent the wildlife component for our parameter estimator rather than breeding birds by nest height due to multicollinearity; and in running different versions of the model we found that none of the categories for breeding birds by nest height were statistically significant. This could be an indication that while the focus groups deemed this as an important classification of the wildlife component, classification of breeding birds by surface water dependency is the preferred classification. The estimate for length of wetted stream (surface water) was insignificant; however the classification of breeding birds by surface water dependency was significant. Length of wetted stream may be viewed as an input into the production process for vegetative conditions and wildlife species, the ecological endpoints for the survey respondents. The results for the MRG find less statistical significance in the ecological endpoints for the wildlife species; however we still find the classification of breeding birds by surface water dependency as the preferred classification.
Calculation of the MWTP estimates from Equation (4) can also be found in Table 2. All of the MWTP estimates for the SPRNCA are found to be statistically significant while only the water bound bird category and MO#1 are significant for the MRG. An interesting finding for the SPRNCA is that the MWTP for additional increases in breeding birds are found to be negative. This could be a result of over two decades worth of restoration which has resulted in current conditions in the SPRNCA that satisfy the general public and there is no willingness to engage in further restoration in an effort to increase breeding birds by surface water dependency (see Broadbent et al. 2015). There is a desire to engage in activities that increase the wet and intermediate condition classes in the SPRNCA, as the positive and significant MWTP estimates indicate. For each percentage increase in acreage for the intermediate class we find a WTP of $5.74 and $3.16 for the wet class. This demonstrates that while WTP is increasing across the condition classes, it is increasing at a decreasing rate. Currently, 10% of the SPRNCA is classified as the dry condition class, 30% as the intermediate and 60% as the wet condition class. A one percent increase in a condition class is roughly 20 acres. By acre this means a change to the intermediate class is $0.29 and $0.16 for the wet class.
For the MRG, the results differ from the SPRNCA. We find a willingness to increase the amount of water bound birds as the MWTP is $8.30, meaning participants are WTP $8.30 to obtain 1,000 more breeding birds that are dependent upon surface water flows. We do not find the MWTP to be statistically significant for the non-water bound breeding birds. This result could be tied to the management option that is currently being conducted in the MRG. In the MRG, the driver of change to the ecosystem is mechanical clearing of the understory to reduce fire risk, which we denote as MO#1 and it is the only statistically significant vegetative class. This willingness to pay may be a result of current active management in the MRG, resulting in the other management option not being a realistic alternative to the survey respondents.
Turning to the benefits transfer, Table 3 reports the results of a two sample t-test, as given in Equation (5). If a point estimate BT approach is taken between these two sites one could be confident in using either the migratory bird estimates for the SPRNCA or the MRG, as they are not found to be statistically different, meaning the estimates for the SPRNCA or MRG provide the same MWTP values. The magnitude of difference in the two MWTP estimates, as shown in Table 2, could be adjusted for the difference in bird densities between the two regions (refer back to Figure 3 to see the difference in migratory bird species). The same cannot be said for breeding birds by surface water dependency, as these estimates are negative for the SPRNCA and positive for the MRG leading to a test statistic that rejects the hypothesis of equity in the MWTP estimates. Again, this result could be due to the different drivers of change in the two ecosystems. For the vegetative classes, in comparing the wet condition class to MO#1 there is no difference in these two, while the intermediate class and MO#2 do differ. While not perfect, the vegetative condition classes were developed to be as close as possible for ecological accounting units. Surface water could not be compared between the two regions, as surface water is invariant in the MRG as described by average miles of surface water available.

Reflections and conclusions
In an attempt to create transferable ecological accounting units for economic valuation, this research develops a suite of ecological units for two semi-arid riparian ecosystems in the southwestern US, the SPRNCA and MRG. We employed the Boyd and Banzhaf (2007) and Boyd and Krupnick (2013) criteria in creating our ecological accounting units, which are important inputs in benefit costs analyses to account for the impact of change on environmental amenities. If ecological systems are to be included in benefit cost analyses, it is important to quantify their value so that their values are included in benefit cost analyses to protect and preserve ecosystems so that degradation, as the Millennium Ecosystem Assessment (2005) has found, does not continue. In using our developed ecological accounting units to estimate MWTP for both the SPRNCA and MRG we found that defining avian species by their dependence on surface water was the most statistically significant ecological endpoint. As Boyd and Krupnick (2013) point out, ecological services are complex and, in addition, as Barkmann et al. (2008) have shown, participants have a difficult time understanding "basic science" models. Presenting breeding birds by nest height is an example of the complexity in understanding "basic science" models. The results of our study suggest a more basic approach in classifying breeding birds by surface water dependency and not by nest height. In creating attributes that are based on vegetation classes we find less evidence as to their importance in estimating MWTP, finding it difficult to standardize ecological accounting units for this attribute similar to Crossman et al. (2013) andde Groot et al. (2012) where they argue that no one approach can do it all. Some of the insignificance in the attributes of the two sites may be due to the different drivers of change in the SPRNCA and MRG. The SPRNCA is driven from changes to groundwater levels due to human-induced pumping, while the MRG is driven by mechanical alterations to the ecosystem from management options to reduce the risk of fire.
While both systems experience ecological changes, how these changes are brought about could be altering the economic value of the system to the respondents. This is a challenge in developing ecological accounting units, as the drivers of change in the two ecosystems are not standardized. Future research should focus on creating standardized ecological accounting units with similar drivers of change. In doing so, we advocate incorporating scientific models to estimate the magnitude of change in the ecosystem attributes, as we have done in this study, and not rely upon hypothetical scenarios in developing a CE survey.
We conducted a BT approach to compare MWTP for ecological endpoints which we designed to be similar. While BT techniques have become more popular in the published literature due to their ability to provide quick and inexpensive estimates for valuation, problems still arise. As this study has highlighted, the accuracy of point estimate transfers deteriorate as the policy site's characteristics differ from those of the study site. Johnston and Rosenberger (2010) have highlighted issues that surround BT techniques, such as differentials in policy and study sites. Recently, in an effort to create a database of studies that value environmental amenities the Environmental Valuation Reference Inventory (EVRI) 4 has been created. This is a valuable resource in conducting BT, but it does not solve the issue that differences may exist between policy and study sites. We hope to see further emphasis in this area in the literature as ecosystem values are calculated and transferred across sites.
As the literature evolves on defining ecological accounting units there is a need for additional studies, such as this one, that compare the use of standardized ecological accounting units across similar and diverse ecosystems. While this is a challenging process, it is necessary to understand differences in the valuation of similar and diverse ecosystem services. In choosing to apply our standardized accounting units we chose two ecosystems that appeared to be relatively similar. This proved problematic as the driver of change in the two ecosystems differed, which could be why we observe differences in the MWTP for the two ecosystems, even though both systems are described through a similar set of ecological endpoints. Future studies should take this into account to verify whether homogenous systems with similar drivers provide similar estimates or whether the public views these ecosystems as unsubstitutable and heterogeneous.

Disclosure statement
No potential conflict of interest was reported by the author(s).