Validation of a habitat suitability index for oyster restoration

Habitat suitability index (HSI) models provide spatially explicit information on the capacity of a given habitat to support a species of interest, and their prevalence has increased dramatically in recent years. Despite caution that the reliability of HSIs must be validated using independent, quantitative data, most HSIs intended to inform terrestrial and marine species management remain unvalidated. Furthermore, of the eight HSI models developed for eastern oyster (Crassostrea virginica) restoration and fishery production, none has been validated. Consequently, we developed, calibrated, and validated an HSI for the eastern oyster to identify optimal habitat for restoration in a tributary of Chesapeake Bay, the Great Wicomico River (GWR). The GWR harbors an unparalleled, restored oyster population, and therefore serves as an excellent model system for assessing the validity of the HSI. The HSI was derived from GIS layers of bottom type, salinity, and water depth (surrogate for dissolved oxygen), and was tested using live adult oyster density data from a survey of high vertical relief reefs (HRR) and low vertical relief reefs (LRR) in the sanctuary network. Live adult oyster density was a statistically significant sigmoid function of the HSI, which validates the HSI as a robust predictor of suitable oyster reef habitat for rehabilitation or restoration. In addition, HRR had on average 103-116 more adults m-2 than LRR at a given level of the HSI. For HRR, HSI values ≥0.3 exceeded the accepted restoration target of 50 live adult oysters m−2. For LRR, the HSI was generally able to predict live adult oyster densities that meet or exceed the target at HSI values ≥0.3. The HSI indicated that there remain large areas of suitable habitat for restoration in the GWR. This study provides a robust framework for HSI model development and validation, which can be refined and applied to other systems and previously developed HSIs to improve the efficacy of native oyster restoration.


Introduction
Habitat suitability indices (HSI) are a commonly developed and often robust spatially 4 explicit, decision support model used to identify the capacity of a given habitat to support a species of interest (U. S. Fish andWildlife Service 1981, Roloff andKernohan 1999). In 6 1981, the United States Fish and Wildlife Service proposed and developed the first HSI models, which were intended to quantify the value of habitats when considering 8 management alternatives in species-specific conservation and restoration (U. S. Fish and Wildlife Service 1981). HSIs are commonly generated through application of 10 wildlife-habitat relationships to relevant geospatial environmental data within a Geographic Information System (GIS) to develop a composite HSI score with a range of 0 12 to 1, representing unsuitable to optimal habitat (Brooks 1997). Depending on the relevance of the selected habitat variables, quality of the geospatial environmental data, 14 and reliability of the applied wildlife-habitat relationships, these models can serve as robust spatial tools to inform species management. 16 Although the USFWS emphasized the need for validation of HSIs, or the quantitative assessment of an HSI's ability to predict habitat suitability via an independent data set, 18 most HSIs intended to inform the management of terrestrial and marine species have not been validated (Brooks 1997, Araújo andGuisan 2006). Recent HSI validation studies 20 have indicated that unvalidated HSIs can be unreliable indicators of habitat quality (Reiley et al. 2014). Although widely used, HSI models have been criticized as unreliable 22 and lacking scientific rigor (Cole andSmith 1983, Roloff andKernohan 1999). Despite the potential cost associated with obtaining these independent datasets, the validation process 24 is required to determine the reliability and utility of these models if they are to inform species conservation and management (Brooks 1997, Tirpak et al. 2009, Reiley et al. 2014).
To implement HSI models confidently, they must be tested for accuracy in a four-step 2 process (Brooks 1997, Tirpak et al. 2009, Reiley et al. 2014: development, calibration, verification, and validation. Development involves the use of wildlife-habitat relationships 4 to generate an HSI ranging from 0, representing unsuitable habitat, to 1, representing optimal habitat. Calibration aims to ensure that the HSI spans the full range of values 6 from 0 to 1. Brooks (1997) notes: "The intent is for sites of excellent habitat quality to receive high scores (e.g., 0.7-1.0), and sites of poor habitat quality to receive low scores 8 (e.g., 0-0.3). If the HSI scores do not ordinate across the entire range of values from 0 to 1, then they will be of little use in describing differences among sites." Verification entails 10 assessment of performance of an HSI model against independent qualitative or categorical (ranked) data. A positive correlation of ranked data, such as presence/absence, and HSI 12 values would provide verification. Validation involves testing the performance of an HSI model against independent quantitative data in space and time, such as against population 14 density or abundance. If validation has been accomplished, verification is not necessary. A note of caution pertains to the use of different, non-independent data for validation. For 16 example, a dataset for a single ecosystem, such as a tributary, could be split in half. One half of the dataset could be used to develop an HSI, and the second half used to test the 18 fit of the HSI model. This would not constitute validation because the data used to test the HSI are not statistically independent of the data used to develop the model. With the advent of contemporary geographic information systems (GIS) software, 12 many HSIs for oysters have been developed within the past decade ( an oyster HSI. Here, we describe the development of a simple, reliable HSI for the eastern oyster and present the results from a direct field validation of the model. 6 Material and methods

Study area 8
The GWR is a tributary on the western shore of the lower Chesapeake Bay ( Figure 1).
The GWR is located approximately 10 km south of the Potomac River and 25 km north of the Rappahannock River, and has a small watershed consisting predominately of forested and agricultural lands (Southworth et al. 2010). The GWR is mesohaline and is considered 12 a trap-type estuary with gyre-like water circulation patterns that has contributed to its history of significant natural oyster recruitment (Andrews 1979, Southworth et al. 2010).
14 The system is characterized by a single, central deep channel with an extensive sand shoal near the river mouth (Southworth et al. 2010 independent survey data. The steps in HSI development were as follows: (i) Assimilation of data sets on environmental variables (e.g., salinity);

24
(ii) Assessment of habitat requirements for eastern oyster from a literature review; (iii) Construction of ArcGIS environmental layers; (iv) Formulation of suitability functions for each of the environmental variables; 2 (v) Calculation of the HSI for the river system; (vi) Collection of eastern oyster abundance data for the system from a population 4 survey; and, (vii) Comparison of HSI values and oyster abundance. 6 The HSI was derived from Geographic Information System (GIS) layers of environmental and biotic variables, including bottom type, land use, salinity, existence of 8 private oyster leases and public oyster grounds, seagrass cover, dissolved oxygen, and water depth for most of the GWR at depths deeper than 2 m. From these variables, we 10 selected those of greatest relevance to site suitability for oyster restoration in the GWR and for which there was river-wide data, which included bottom type, depth, and salinity   processed in the laboratory, rather than in the field, due to the high probability that individual oysters would not be easily seen in the field, resulting in biased (inaccurate) 2 samples. Parameter estimates for density and abundance were obtained using the R statistics package (R Core Team 2015). 4   contained within the private shellfish aquaculture leases.

18
Habitat suitability indices provide a quantitative tool that integrates the best available environmental data and corresponding science to identify locations with the most potential Some of the previous oyster HSI models have included additional variables, such as 10 turbidity and predation intensity (Table 2), for smaller or more intensively studied waterbodies, where these spatial data sets are available. For larger or less studied 12 waterbodies, spatial data sets for large suites of environmental variables are often rare.
Further, based on the wide range of environmental and biotic variables included in prior 14 HSI models (Table 2), it is clear that a "one size fits all" approach to HSI modeling is inappropriate. While many variables that determine habitat suitability overlap between 16 species and systems, variables that are significant drivers in a given system may not be as critical as other variables. For instance, of the four oyster HSI models developed for were not in common ( Table 2).
The eastern oyster provides an excellent example of this variability in requirements in 24 that its geographic range encompasses the (i) Atlantic coast from Canada to Florida, (ii) Gulf of Mexico coast, and (iii) Caribbean from the Yucatan Peninsula to the West Indies suite of habitats across temperate, subtropical and tropical areas (Kennedy et al. 1996).
When using HSI models that incorporate a small subset of variables, such as only salinity 2 and substrate (e.g., Soniat et al. (2013), Swannack et al. (2014)), it is especially valuable to conduct model validation. Such simple models trade incorporation of large suites of 4 environmental layers (for which there are rarely comprehensive spatial datasets for large areas) for broader spatial coverage, with the underlying assumption being that oyster 6 habitat suitability can be adequately described by salinity and substrate alone. In the case of Chesapeake Bay, for which seasonal anoxia and hypoxia are prevalent in deeper waters, 8 omission of a variable that encapsulated dissolved oxygen (i.e., water depth) may lead to an HSI erroneously overstating the extent of suitable oyster habitat for restoration. In this 10 case, water depth was an effective, if imperfect, surrogate for dissolved oxygen, and was largely accountable for the performance of the HSI model along with bottom type.

12
Bathymetric information is often available for most waterbodies, and an understanding of the relationship between depth and dissolved oxygen concentrations can be useful to 14 eliminate areas of hypoxia or anoxia in restoration efforts by use of HSI models. In the case of the GWR, seasonal hypoxia in areas deeper than 4 m, reduced salinity in upriver 16 locations, and subsidence of reef material in areas of soft sediments had previously been identified as priority factors that could negatively impact the success of oyster restoration.

18
Thus, developing an HSI that integrates the variables known to be major drivers of restoration success in a particular system with subsequent validation and model refinement 20 is the optimal, robust approach.
Given the commonly stated goal for oyster restoration projects of oyster abundance 22 and biomass enhancement, live adult oyster density data derived from the 2011 survey of the ACE restored reefs in the GWR were used to validate the model. Our use of live adult Ecology,Vol. 00,No. 0 total live oyster density. Immediately following restoration, oyster sanctuaries can experience major recruitment pulses that can temporarily inflate total oyster density with 2 size structure skewed towards high densities of recruits and sub-adults, which have reduced probabilities of survival relative to adults (Puckett and Eggleston 2012). However, in areas of low-relief reef with an HSI ≥0.3, density rapidly increases to nearly double the 50 live adult oysters m −2 target in low-relief reef ares with an HSI of 1.
Contrasting this, for high-relief reef in areas with an HSI as low as 0, the 50 live adult 12 oysters m −2 target is exceeded with a rapid increase in density to nearly 5x the target at an HSI of 1, although the long-term sustainability of reefs in areas with low associated HSI 14 values is questionable. This finding also provides further support to the existence of bistability on the restored oyster reefs in the GWR in that under similar environmental 16 conditions (i.e., for a given level of HSI), low-and high-relief reefs remain differentiated by oyster density.

18
With the increasing availability of spatial datasets for environmental variables in marine and estuarine systems, habitat suitability indices will likely continue to be 20 developed to inform species conservation and management. However, as cautioned by the USFWS shortly after the development of the first HSIs in the 1980s, the performance of 22 these models must be quantitatively assessed via an independent dataset. This study provides a robust framework for HSI model development and validation, which can be 24 refined and applied to other systems and previously developed HSIs to improve the efficacy of native oyster restoration. with independent quantitative data in space and time, such as population density or abundance. If validation has been accomplished, verification is not necessary.
8 Table 2. Variables used in habitat suitability index models developed for oyster aquaculture, fishery production and restoration, and the models using each variable.  Table 1. Habitat suitability index models developed for oyster aquaculture, fishery production, and restoration. Calibration is achieved when the HSI approximately spans the full range from 0 to 1. Verification is achieved when the HSI values are positively correlated with independent qualitative or categorical (ranked) data, such as presence/absence data. Validation is achieved when the HSI values correlate positively with independent quantitative data in space and time, such as population density or abundance. If validation has been accomplished, verification is not necessary.