MEMORANDUM. Executive Summary. DATE: August 18, Joe DePinto, Todd Redder, Dan Rucinski

Transcription

1 DATE: August 18, 2009 MEMORANDUM FROM: PROJECT: TO: Joe DePinto, Todd Redder, Dan Rucinski LOIERM Sandra George, Environment Canada Kim Waltz, Wisconsin Department of Natural Resources International Upper Great Lakes Study - Environmental Technical Working Group CC: SUBJECT: Lessons Learned from the Ecological Modeling Conducted for the IJC Lake Ontario St. Lawrence River Study Executive Summary The International Upper Great Lakes Study Board is in the third year of a five-year study for the International Joint Commission to evaluate the overall impacts of current and fencepost Lake Superior regulation plans on the Upper Great Lakes system. The International Upper Great Lakes Study (IUGLS or Study) is considering the potential future impact of regulation plans and basin water supply scenarios on the environment, as well as various economic sectors, including hydropower, commercial navigation, and recreational boating. With regard to the environmental interest, the Environmental Technical Work Group (ETWG) has been charged with developing a quantitative understanding of the response of key ecological performance indicators (PIs) to alternative regulation plans when applied to a range of net basin water supplies, including potential climate change conditions. This memorandum provides a detailed review and evaluation of the IJC Lake Ontario St. Lawrence (LOSL) Integrated Ecological Response Model (IERM) with respect to its utility for the IUGLS. Specifically, this memorandum provides the following: A review of the overall Integrated Ecological Response Model (IERM) framework in terms of the applicability and portability to the UGL study; Review of the IERM ecological sub-models in terms of the applicability and portability to the UGL study; and Results and findings of an illustrative application of the LOSL IERM to a suite of regulation plans and basin supply scenarios for Lake Superior. Key components of the IERM structural framework, including the core source code and the supporting database design, were reviewed with respect to their applicability and portability to the Upper Great Lakes system. Many of the structural features of the IERM framework, including the database structure, visualization tools, and algorithms for facilitating aggregation and weighting of PI results were determined to be readily portable to an IERM application for the IUGLS, with little or no modification required. 501 Avis Drive Ann Arbor, MI Fax:

2 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 2 In addition to the structural features of the IERM, ecological sub-models and performance indicators developed for the LOSL Study were reviewed and evaluated with respect to their portability to the Upper Great Lakes system. In general, it appears that it may be possible to adapt the core algorithms of the wetland vegetation, fish, and wetland bird sub-models to Lakes Superior, Michigan, and Huron provided that sufficient physical data are available. Further literature review, model testing, and discussion with ETWG scientific experts will be required to determine whether sufficient data are available to adapt these three sub-models to various locations within the Upper Great Lakes system. Given the overall paucity of physical and biological data at regional scales, application of these existing sub-models would likely need to occur on a site-specific or reach-specific basis. At a minimum, detailed bathymetric/topographic data would be required to facilitate application of the ecological sub-models to a particular site or reach. For example, the wetland vegetation algorithm could potentially be applied to wetlands associated with Saginaw Bay based on existing bathymetric and LiDAR survey datasets available for this area. The existing version of the IERM was applied to investigate the potential sensitivity of the suite of ecological performance indicators originally developed for the LOSL study to Lake Superior water level regimes resulting from various regulation plans generated to date by the Shared Vision Model (SVM) and two basin supply scenarios (historic and a climate change scenario). The results shown here indicate that there is potential for identifying sensitivities for at least a subset of the performance indicators that were developed for the LOSL study. Similar to findings from the LOSL Study, these results demonstrated that there are likely to be both positive and negative Performance Indicator (PI) responses to any given regulation plan with a range of sensitivities among the PIs being tested. Therefore, comparison of any given pair of plans will require evaluating the body of results in the context of the relative significance, certainty, and sensitivity associated with each individual PI. However, the results of these runs should not be used to draw any conclusions about the relative benefits of any plans because 1) water level changes exceeded the bounds of some algorithms, and 2) it was not possible to specify appropriate initial conditions for the wetland vegetation sub-model, which directly or indirectly drives many of the PI responses. One potentially important aspect of the IUGLS modeling that was not addressed in the LOSL IERM is the potential for various ecosystem components to evolve/adapt to long-term changes in water level magnitude, timing, and frequency that may occur as a result of climate change in the Great Lakes basin. For example, the wetland vegetation model developed for Lake Ontario operates within a very specific elevation range ( meters IGLD85) based on the bathymetric/topographic datasets collected for the 32 wetland study sites. The use of this restricted elevation range implicitly limits the application of the LOSL implementation of the vegetation sub-model to water level regimes that fall within this range. However, it is entirely possible that water levels in the Great Lakes will decrease by up to a meter as the result of climate change. Therefore, there is a need to consider the potential migration of vegetation down (or up) slope in response to the climate-driven trends. Additionally, fauna may adapt in terms of their habitat preferences or tolerances in response to significant changes in water level regime. Again, this analysis will likely require detailed bathymetric datasets that extend a meter or more below the current low water datums for Lake Superior, Michigan, and Huron.

3 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 3 Review of IERM Modeling Framework This section provides a review of the overall Integrated Ecological Response Model (IERM) framework with respect to its applicability and portability to the International Upper Great Lakes Study. The term framework is used here to describe the structural, object-oriented code and database components of the IERM used to support simulations of ecological response, but not the functional ecological sub-models developed for the LOSL study. The individual ecological sub-models are addressed in the next section. Overview The IERM framework provides a suite of model analysis and visualization tools to support the evaluation of ecological responses across multiple regulation plans and basin supply scenarios. The IERM database, which is developed in Microsoft Access, provides the backbone for these tools by storing information and results for all performance indicators (PIs) calculated via the ecological sub-models, as well as hydrologic criteria. The IERM database was intentionally structured to provide flexibility in specifying all kinds of PIs and criteria using robust, yet flexible protocols for storing documentation, model inputs, and model results within the database structure. Additional databases were developed for the LOSL application of the IERM to support specific ecological sub-models, including those for Lake Ontario wetland vegetation and fish habitat and population. An IERM simulation involves the execution of multiple ecological sub-models to compute the response of various ecological PIs to a specified time period of quarter-monthly water level conditions. An annual result is computed for each individual PI and stored in the IERM database. Once all of the ecological sub-models have been executed, an aggregation routine is run to distill the annual time series of PI results to a single PI metric for the simulation. For example, for a PI that describes the annual habitat area available for a specific fish species, the aggregation metric might be the annual average area of available habitat. For other types of PIs, maximum or minimum value predicted during the simulation period may be of interest. The aggregation routine in the IERM provides the flexibility to compute a variety of these metrics for a given PI. The annual time series and aggregate metric results generated for each individual PI are stored in the IERM database, where they are utilized by the suite of visualization tools available in the IERM framework. The overall storage and flow of information is illustrated in Figure 1.

4 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 4 IERM Program (Visual Basic 6) IERM Main Menu (IERM.exe) IERM Simulation Routines Hydraulics Performance indicators Criteria Model Input/Output IERM Simulation Databases IERM_LO_wetland.mdb IERM_LO_Fish*.mdb IERM_LO_WLBird.mdb IERM_USL_Fish*.mdb IERM_LSL_v5.mdb IERM Visualization Routines Hydraulics Performance Indicators Criteria DAO 3.6 Library Simulation Simulation Results Results IERM Main Database (IERM_v5.mdb) IERM Databases (MS Access) Figure 1. IERM Framework Flow Chart Key tools available for visualizing PI results in the current IERM framework include the target plot (Figure 2), output summary (Figure 3), a time series plot (Figure 4), and plots comparing PI results across all regulation plans and basin supply scenarios (Figure 5). The IERM also provides tools for visualizing key hydraulic time series results (e.g., water levels in various locations), evaluating hydrologic criteria between plans, and reviewing documentation for individual performance indicators. The target plot allows the user to select a baseline and an alternative plan and then calculates the ratio (i.e., alternative:baseline) of the PI metrics calculated for the selected plans as described above. In Figure 1, the center of the target diagram represents a bullseye (ratio 2.0), and the red circle represents a ratio of unity (ratio = 1.0), indicating that there is no perceived difference between the selected baseline and alternative plans. Based on the ratio approach, PI points that fall inside the red circle suggest an improvement for the alternative plan relative to the baseline plan, and vice versa for points that fall outside the red circle. For the LOSL application, individual ecological PIs were organized and displayed by region (Lake Ontario and upper/lower St. Lawrence River reaches) and by category (vegetation, fish, birds, herptiles, mammals, and species-at-risk). The organization and display of PIs for the UGLS can be easily modified to categorize results by lake, by PI category, and/or by site ecosystem type. Clicking on a PI point displayed on the target diagram generates an output summary such as that shown in Figure 3. The user can drill down further into the results for a specific PI by clicking the View Time Series button, which generates the visualization shown in Figure 4. The time series visualization shows annual PI results for the entire simulation period in the upper pane, and also provides a time series of the hydrologic forcing function (e.g., water level) that drives the PI response (either directly or indirectly). The time series plot allows the user to select different baseline and alternative plans to explore how the annual results change based on

5 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 5 different regulation schemes and/or basin supply scenarios. The target and time series visualization tools also allow the user to view documentation for individual PIs ( PI Browser button) or export results to a Microsoft Excel worksheet. Additional visualization tools are also available for comparing PI metrics across the full suite of regulation plans and supply scenarios for which simulations have been conducted (see Figure 5). Figure 2. IERM Performance Indicator Target Plot

6 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 6 Figure 3. IERM Performance Indicator Output Summary Figure 4. IERM Performance Indicator Time Series Plot

7 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 7 Figure 5. IERM Performance Indicator Comparison across Regulation Plans In addition to providing a series of visualization tools to evaluate individual PI responses, the IERM framework provides the capability of applying weighting factors to individual PIs, regions, and PI categories to generate region-specific, category-specific, and overall weighted ratios of ecological response (Figure 6). Once weights have been specified for each PI, region, and category (either via the user interface or directly in the IERM database), the results can be displayed on the target diagram in addition to, or in place of, the individual PI results.

8 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 8 Figure 6. IERM Weighting Factors and Results Applicability of IERM Framework to IUGLS As indicated above, the IERM framework code structure and supporting database provide considerable flexibility in terms of accommodating a variety of ecological PIs, hydrologic criteria, and associated documentation. The general aggregation and weighting algorithms and the visualization tools described in the previous sections can all be readily applied for the IUGLS with little or no modification required to the source code. Therefore, the following features of the IERM framework can be utilized for the IUGLS application: Overall IERM code and supporting database structure; Target and time series visualization tools; Other plotting options for comparing results for PIs and hydrologic criteria across regulation plans and basin supply scenarios; General aggregation algorithm for calculating PI metrics; Algorithm for weighting PI results by location/region and category; and PI documentation capabilities. A majority of the features and capabilities listed above can be used as-is for the IUGLS IERM application, although it will be necessary to develop a revised suite of ecological performance indicators and hydrologic criteria to feed into the framework. Some modifications or enhancements may be required to the algorithms that allow for aggregation of PI time series results and weighting of PI results. However, these adjustments are expected to be relatively

9 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 9 minor and straightforward to implement. Other adjustments will likely include reconfiguring the PI categories and the location/regions that are used to group the individual PIs. Due to limited resources and limited large-scale datasets to support ecological modeling for the Upper Great Lakes system, it will likely be necessary to focus on site-specific ecological submodels and performance indicators. Ideally, the selected sites will encompass a variety of regions and shoreline types (e.g., beach) within Lakes Superior, Michigan, and Huron. Sitespecific models will likely rely on detailed local bathymetry/topography and other supporting datasets. Therefore, it may be of value to incorporate an interactive, programmable map interface into the IERM framework to support 1) geoprocessing of localized datasets, and 2) spatial visualization of PI results across the various sites and regions used to represent the Upper Great Lakes system. has worked with several programmable mapping tools that may be considered for this purpose, including MapWindow ( and ESRI s MapObjects. The potential value/need for mapping and geoprocessing utilities will be assessed as data acquisition and model development activities progress over the next few months. Because the general IERM framework, including the code and supporting database structure, was developed in a flexible and extensible manner, the bulk of the effort in developing the IUGLS IERM application will be directed at developed of new and/or revised ecological sub-models and associated performance indicators. The next section provides an overview of ecological submodels developed for the LOSL application that may be applicable in some form to the Upper Great Lakes system. Review of IERM Ecological Sub-Models As discussed previously, the LOSL application of the IERM includes a suite of ecological submodels and associated performance indicators (PIs) that were developed by LOSL ETWG researchers to represent flora and fauna responses in Lake Ontario and the St. Lawrence River resulting from alternative regulation plans and various basin supply scenarios. This section provides a brief description of each major sub-model developed for Lake Ontario and the uppermost reach of the St. Lawrence River (i.e., Thousand Islands area) and discusses potential application to the Upper Great Lakes for the IUGLS. It should be noted that sub-models developed for the St. Lawrence River below the Iroquois Dam are not discussed here because they represent PIs specific to riverine conditions, and are therefore not applicable to the IUGLS. Table 1 provides a summary of the key performance indicators selected and developed for Lake Ontario and the upper St. Lawrence River for the LOSL application of the IERM, including ratings of significance, certainty, and sensitivity that were assigned to aid evaluation and interpretation of model results.

10 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 10 SVM ID Table 1. Key Ecological PIs for Lake Ontario and the Upper St. Lawrence River from the LOSL Study Region PI Group E1 Lake Ontario Vegetation E2 Lake Ontario Fish E3 Lake Ontario Fish E4 Lake Ontario Fish E5 Lake Ontario Fish E6 Lake Ontario Fish E7 Lake Ontario Birds E8 E9 E10 E11 E12 Lake Ontario Lake Ontario Lake Ontario Lake Ontario Upper SL River Speciesat-Risk Speciesat-Risk Speciesat-Risk Speciesat-Risk Fish PI Description Wetland Meadow Marsh Community - total surface area, supply-based (Lake Ontario) Low Veg 18C - spawning habitat supply (Lake Ontario) High Veg 24C - spawning habitat supply (Lake Ontario) Low Veg 24C - spawning habitat supply (Lake Ontario) Northern Pike - YOY recruitment index (Lake Ontario) Largemouth Bass - YOY recruitment index (Lake Ontario) Virginia Rail (RALI) - median reproductive index (Lake Ontario) Least Bittern (IXEX) - median reproductive index (Lake Ontario) Black Tern (CHNI) - median reproductive index (Lake Ontario) Yellow Rail (CONO) - preferred breeding habitat coverage (Lake Ontario) King Rail (RAEL) - preferred breeding habitat coverage (Lake Ontario) Low Veg 18C - spawning habitat supply (Upper St. Lawrence) PI Units Researchers Significance Certainty Sensitivity ha Wilcox, Ingram ha-days ha-days ha-days Minns, Doka, Chu, Bakelaar, Leisti Minns, Doka, Chu, Bakelaar, Leisti Minns, Doka, Chu, Bakelaar, Leisti index Minns, Doka index Minns, Doka index index index ha ha ha-days DesGranges, Ingram, Drolet DesGranges, Ingram, Drolet DesGranges, Ingram, Drolet Lantry, Schiavone Lantry, Schiavone Minns, Doka, Chu, Bakelaar, Leisti

11 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 11 SVM ID Region PI Group PI Description PI Units Researchers Significance Certainty Sensitivity E13 Upper SL River Fish High Veg 24C - spawning habitat supply (Upper St. Lawrence) ha-days Minns, Doka, Chu, Bakelaar, Leisti E14 Upper SL River Fish Low Veg 24C - spawning habitat supply (Upper St. Lawrence) ha-days Minns, Doka, Chu, Bakelaar, Leisti E15 Upper SL River Fish Northern Pike - YOY recruitment index (USL) index Minns, Doka E16 Upper SL River Fish Largemouth Bass - YOY recruitment index (USL) index Minns, Doka E17 Upper SL River Fish Northern Pike - YOY net productivity (USL - Thousand Islands) grams/ha Farrell E18 Upper SL River Birds Virginia Rail (RALI) - median reproductive index (Lake St. Lawrence) index DesGranges, Ingram, Drolet E19 Upper SL River Mammals Muskrat (ONZI) - house density in drowned river mouth wetlands (Thousand Islands area) #/ha Farrell, Toner 4 4 5

12 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 12 Wetland Vegetation Sub-model The wetland vegetation sub-model for Lake Ontario was developed based on extensive wetland plant species composition sampling and bathymetric/topographic surveys conducted for 32 wetlands around the lake shoreline and the Thousands Islands area (Wilcox and Xie, 2007; Wilcox et al., 2005). Selected wetland sites were evenly distributed among the four geomorphic wetland types found in the Lake Ontario environment: Barrier beach (8 sites total); Drowned river mouth (8 sites); Open embayment (8 sites); and Protection embayment (8 sites). Plant species composition and bathymetric/topographic data collected at each site were compiled and used to support the development of a generalized vegetation model that predicts annual plant community distributions for each of the four geomorphic types based on past water level history. Information regarding the total surface area of each geomorphic type was compiled for all wetland areas in Lake Ontario and the Thousand Islands area. Detailed bathymetry/topography was not available for a majority of the non-studied wetland sites; however, the 32 selected study sites represented 12-26% of the total inventory area for each geomorphic wetland type. Bathymetry and topography data collected for the 32 study sites were used to develop typical wetland geometry for each of the geomorphic types. Because the configuration for a given wetland geomorphic type is generally consistent from site to site, the use of a typical/average geometry was considered adequate for representing wetlands occurring across the Lake Ontario and Thousand Islands shoreline areas. Plant quadrat sampling was conducted along 7 transects for each of the 32 wetland study sites. Each of the selected transects represented a unique hydroperiod condition, including elevations that had not been flooded in decades, elevations that had been flooded and/or dewatered in recent years, and elevations that had not been dewatered in decades. Plant species composition data collected at the 7 transects were analyzed across the 8 study sites for each wetland geomorphic type using an ordination technique, which provided a means for evaluating similarities in species richness and diversity among the various samples. Based on this evaluation, five unique plant communities were identified across the four geomorphic types: Upland transition community ( U ) last flooded > 30 years ago; Meadow marsh community ( ABC ) last flooded 5-30 years ago; Emergent marsh community #1 ( D ) last flooded < 5 years and/or last dewatered < 4 years ago; Emergent marsh community #2 ( EF ) last dewatered 4-39 years ago; and Submerged/floating leaf community ( G ) last dewatered > 39 years ago. In addition, the plant species composing these five communities were assigned to 15 wetland plant structural categories (e.g., grasses, sedges, shrubs, floating leaf), in order to support the linkage of wetland sub-model predictions to fauna sub-models that required more detailed

13 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 13 vegetation inputs. It is worth noting that while there are significant differences in geometry between the four geomorphic types, the plant communities defined for each geomorphic type were very consistent. This finding reinforced the primary importance of lake water level dynamics in driving wetland vegetation response. The wetland sub-model computes the distribution of the five plant communities described above based on the flooding and dewatering history of a specific elevation at a given point during the simulation. Flooding is defined as the continuous inundation of a particular elevation for an approximately one-month period during the summer season, while dewatering occurs when a particular elevation remains uninundated throughout an entire growing season (April-October). The flooding and dewatering elevations are calculated for each year of a simulation, and this information is used to determine the number of years elapsed since each elevation has been flooded or dewatered. Given this information, a plant community is assigned to each elevation interval represented in the generalized wetland geometry for each year. Finally, the total percent coverage of each plant community within each geomorphic type is calculated on an annual basis and combined with the total system-wide areas of each geomorphic type to generate predictions of areal coverage for each plant community. The key performance indicator derived from the wetland vegetation sub-model was the areal extent of meadow marsh, a critical habitat zone that has been compressed due to decades of water level regulation in Lake Ontario. Expansion of meadow marsh was considered a desirable outcome when comparing alternative regulation plans (Wilcox and Xie, 2007). Application to the Upper Great Lakes The wetland vegetation sub-model certainly has potential value for application to the Upper Great Lakes system. As described above, the model algorithm tracks flooding and dewatering events at specified elevations, a concept that is transferable to Lakes Superior, Michigan, and Huron. The inputs required to drive the wetland vegetation sub-model include plant community types associated with various flooding/dewatering intervals and the geometry associated with the wetland(s) of interest (i.e., the distribution of area across a given elevation range). A review of the existing literature for Lakes Superior, Michigan, and Huron should be conducted to determine if the plant communities and associated flooding/dewatering intervals established for Lake Ontario could be applied in the IUGLS IERM application. Assuming that vegetative communities are reasonably similar, then the largest obstacle to applying the wetland sub-model for wetland sites in the Upper Great Lakes is obtaining detailed, site-specific bathymetry and topography datasets based on LiDAR or other collection techniques. The availability of bathymetric/topographic datasets for sites in the Upper Great Lakes is currently being evaluated. At a minimum, it is anticipated that detailed datasets will be available for Georgian Bay and Saginaw Bay. Based on the above considerations, it is recommended that a modified version of the wetland vegetation sub-model developed for Lake Ontario be strongly considered for application to sites where appropriate bathymetric/topographic datasets are (or will be) available. At a minimum, it will be necessary to reconfigure the wetland geometry represented in the database that supports this sub-model. Pending a review of available literature pertaining to Upper Great Lakes vegetative communities, it may also be necessary to modify the plant communities or the associated flooding/dewatering intervals from those used for Lake Ontario. Finally, attention should be paid to predictions of vegetation response developed using other models (e.g.,

14 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 14 Georgian Bay models developed by Dr. Chow-Fraser) to insure a reasonable level of consistency between the models, particularly for sites/reaches in close proximity. Fish Habitat & Population Sub-models The fish subgroup of the overall LOSL ETWG developed a series of habitat supply and population models and associated performance indicators to evaluate fish response to water level regulation in Lake Ontario and the upper St. Lawrence River. Habitat supply models were developed for 8 fish guilds representing a range of thermal (10, 14, 18, 24 C) and vegetation (high/low) preferences. An equivalent set of habitat supply models was developed for 4 individual species (northern pike, smallmouth bass, largemouth bass, and yellow perch) and 2 species-at-risk (pugnose shiner, bridle shiner). In addition, population models were developed to predict relative population response for northern pike, smallmouth bass, largemouth bass, and yellow perch. Each of the habitat supply models predicts daily weighted suitable habitat area for 5 life stages: spawning, fry, young-of-year, juvenile, and adult. The weighted suitable area calculations are based on relatively complex algorithms that determine local habitat suitability based on water levels and a range of factors, including: Daily water temperature (adjusted for nearshore locations); Site-specific hypsographic profiles for nearshore and wetland areas; Substrate type; Depth-based predictions of submerged aquatic vegetation (SAV) cover; and Predictions of emergent vegetation cover generated by the wetland plant sub-model. Daily estimates of habitat supply were summed across each calendar year to obtain total annual habitat supplies in units of hectare-days. The 4 species population models predict annual youngof-year, juvenile, and adult population densities based on calculations of daily growth and survival for each of the 5 life stages. The population models require several inputs, including: Daily weighted suitable area estimates for each life stage (as generated by the habitat supply models); Daily water temperature (adjusted for nearshore locations); Species-specific bioenergetic and mortality rates; and Probability of stranding based on water level fluctuations. The guild and species habitat supply models were applied for representative sub-regions in Lake Ontario and the upper St. Lawrence, including the Thousand Islands area (IJC Shoreline Unit R1). For Lake Ontario, 6 sub-regions were represented in the model: Bay of Quinte, Presquile, North Central Shore, West Shore, South Central Shore, and the Outlet Basin. The species population models were applied to simulate local population trends within each of the LO/USL sub-regions listed above. The pugnose shiner and bridle shiner habitat supply models were applied only for local areas where these species have been observed, including Sodus Bay and portions of the eastern Lake Ontario shoreline and outlet basin. Complete documentation for the fish habitat supply and population models can be found in Minns et al. (2005).

15 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 15 As shown in Table 1, the performance indicators derived from the fish sub-models included both suitable habitat availability (i.e., areal extent) and population indices for the species described above. Application to the Upper Great Lakes Similar to the wetland vegetation sub-model, the fish habitat and population models have potential with regard to application to the Upper Great Lakes system. However, the site-specific input data needs for the fish sub-models are much more intense than for the vegetation submodel. While the wetland sub-model could potentially be applied to the Upper Great Lakes by incorporating local or regional bathymetry, the fish sub-model would require reach-specific estimates of substrate type, estimates of daily water temperature, and more extensive (e.g., reachscale) bathymetry and topography datasets. In addition to identifying sites/reaches where these various data needs could be adequately met, it would need to be determined if the four species selected for the Lake Ontario model (northern pike, smallmouth and largemouth bass, and yellow perch) are appropriate indicator species to model in Lakes Superior, Michigan, and Huron. Based on the above considerations, it is recommended that work with Susan Doka, one of the primary developers of the original fish models, to determine how best to leverage the considerable model development work conducted for Lake Ontario for IUGLS IERM application. Wetland Bird Sub-model The wetland bird sub-model was developed based on non-linear regressions of breeding pair density and nest losses due to flooding and stranding events (DesGranges et al., 2005). These regressions were based on extensive field density data collected in the lower St. Lawrence River and Ontario and Quebec nesting chronology records. The wetland bird performance indicators developed for Lake Ontario represent indices of reproductive success for 8 species: veery, song sparrow, American bittern, marsh wren, common moorhen, least bittern, black tern, and Virginia rail. Of these 8 species, all except veery (treed swamp) and song sparrow (shrub swamp) utilize emergent marsh vegetation as principle nesting habitat. The reproductive success indicators are the product of two factors: wetland bird breeding pair density and nesting success related to potential flooding/stranding events. The breeding pair density is computed annually for the breeding season (roughly late May July) based on a species-specific function of water depth in the preferred breeding habitat (emergent marsh, treed swamp, or shrub swamp). The wetland bird habitat types were defined relative to the wetland plant sub-model communities as follows: Treed swamp ( A for all geomorphic types); Shrub swamp ( B for barrier beach / drowned river mouth wetlands); Meadow marsh ( BC for open/protected embayments, C for barrier beach / drowned river mouth wetlands); and Emergent marsh ( DEF for all geomorphic types). A second calculation is performed to estimate the nesting success of each species during the breeding season for each year. The nest success estimate is based on the probability of flooding

16 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 16 and/or stranding of the nest due to water level fluctuations occurring during the initial nesting period and subsequent re-nesting attempts. After the breeding pair density and nesting success has been computed for a given simulation year, these two values are multiplied together to obtain an overall reproductive index. For Lake Ontario, fluctuations in water level during the wetland bird breeding season are typically minor; therefore, the annual PI response is largely determined by the breeding pair density, which is driven by the seasonal (late May-July) availability of water in the preferred breeding habitat. Application to the Upper Great Lakes The wetland bird sub-model developed for the LOSL study relies strongly on predictions from the Lake Ontario wetland vegetation sub-model and supporting bathymetric/topographic data, with the only additional input being the species-specific breeding pair density functions. The breeding pair density functions for the 8 selected wetland bird species were developed based on a variety of nesting chronology records from riverine and lacustrine systems in Ontario and Quebec; therefore, it may be possible to use these functions directly to describe density distributions for the Upper Great Lakes. A review of any pertinent literature for Lakes Superior, Michigan, and Huron should be conducted to determine if this approach is reasonable. If so, or if alternative density functions can be developed for one or more of the upper lakes, then it should be feasible to apply the wetland bird sub-model to the Upper Great Lakes system. Because the wetland bird algorithms rely on the inter-annual interplay between water levels and habitat zones, this sub-model provides additional value beyond the wetland or fish sub-models in terms of identifying potential issues with significant transitions in water level regime that may occur across a series of years. Herptile Sub-model This herptile (reptile/amphibian) sub-model is based on logistical regression analyses of field capture data collected for 300 trap locations within eastern Lake Ontario and upper St. Lawrence River (Thousand Islands area) protected embayment wetlands. The regression models predict the probability of occurrence for each species as a function of water depth and percent cover of several plant communities, including upland, shrub/scrub, fine-leaved emergent vegetation, broad-leaved emergent vegetation, and submerged/floating vegetation. The herptile performance indicators represent the probability of occurrence, or relative habitat suitability, on a 0-1 scale for six species: Midland painted turtle, snapping turtle, Blanding s turtle, green frog, leopard frog, and American toad. Application to the Upper Great Lakes It is probably feasible to apply the herptile sub-model to the Upper Great Lakes system, assuming that the probability of occurrence functions developed for the LOSL system can be considered to be reasonably representative of the upper lakes. However, the herptile performance indicators are unlikely to provide additional value in terms of indentifying critical thresholds related to water level timing, duration, and frequency metrics. Because herptile occurrence is strictly tied to the occurrence of various wetland vegetative communities, the response of these performance indicators can simply be inferred from the wetland vegetation sub-model results. For this reason, herptiles were not among the 19 key PIs selected to characterize the Lake Ontario and upper St. Lawrence River ecological response (Table 1).

17 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 17 Muskrat Sub-model The muskrat sub-model was developed based on two primary data sources: 1) winter house censuses conducted for the period, and 2) digital elevation maps constructed for 6 upper St. Lawrence wetland (drowned river mouth) study sites in the Thousand Islands area. Logistic regression analysis of the field data indicated that fall and winter water depth within the wetland cattail communities and winter air temperature are important variables controlling the density and sustainability of muskrat populations in the Thousand Islands area. The muskrat sub-model combines calculations of presence/absence (probability of occupancy) and the annual density of active muskrat houses into a single muskrat house density performance indicator. The muskrat sub-model computes the probability of muskrat occupancy based on the mean winter (December-February) water depth within the wetland emergent marsh (cattail) plant community. If the probability of occupancy is less than the threshold value of 0.35 for a given year, the wetland cannot support a muskrat population for that year. If the probability of occupancy is greater than 0.35, the actual muskrat house density is calculated based on the mean fall (September-November) water depth within the emergent marsh plant community and the mean winter air temperature. Application to the Upper Great Lakes The muskrat sub-model was developed using site-specific data available from a limited sampling period for wetlands in the Thousand Islands area of the upper St. Lawrence River. Therefore, it is difficult to ascertain if the algorithm used for the St. Lawrence system, could be applied for the Upper Great Lakes system. It is recommended that the pertinent literature (if any) regarding muskrat abundance/success in the upper lakes be reviewed to determine whether there is any basis for applying the regression developed for the LOSL study, or some variation of this algorithm. If a muskrat PI is determined to be important to assess ecological response for the IUGLS, then it will probably be necessary to obtain field data, either from the literature or from this study, to parameterize the empirical algorithm used for LOSL. Species-at-Risk The development of the suite of species-at-risk performance indicators generally relied on the sub-models that were developed to simulate wetland vegetation, fish habitat/population, and wetland birds responses to water level regulation plans. The following is a listing of all speciesat-risk for which performance indicators were developed for Lake Ontario and the Thousand Islands area: Vegetation: Carex atherodes, various dune plant species; Fish: pugnose shiner, bridle shiner; Wetland Birds: least bittern, black tern, king rail, yellow rail; and Herptiles: Blanding s turtle, spiny softshell turtle. As noted in the fish sub-model discussion, performance indicators for fish species-at-risk (pugnose shiner and bridle shiner) were based on predictions of the fish habitat sub-model. Similarly, the vegetation, wetland bird, and herptile sub-models were used to predict the performance of these categories of species-at-risk.

18 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 18 Application to the Upper Great Lakes The applicability of any of the species-at-risk performance indicators developed for the LOSL study depends first and foremost on whether the same species are considered at risk for any area(s) within the Upper Great Lakes system. In general, there is less literature and field data available to support the species-at-risk predictive models than other species of vegetation, fish, bird, and herptile; therefore, it may be difficult to develop meaningful indicators of performance for the IUGLS. Ultimately, results obtained for LOSL species-at-risk performance indicators did not significantly impact the evaluation of alternative regulation plans, and a similar result could be anticipated for the Upper Great Lakes given the likely paucity of data for any such species. IERM Sensitivity Testing for Lake Superior Water Levels The existing version of the IERM was applied to investigate the potential sensitivity of the suite of ecological performance indicators originally developed for the LOSL study to Lake Superior water level regimes resulting from various regulation plans and basin supply scenarios. This section describes the approach undertaken for this application, followed by a discussion of the results and associated findings. Approach Water level time series were obtained from the Shared Vision Model (SVM) for two net basin supply scenarios (historical and climate change scenario 2A ) and five regulation plans (Table 2). Each supply scenario was applied to all five regulation plans to produce a total of 10 water level time series for Lake Superior, each consisting of monthly water levels interpolated to a quarter-monthly interval for use in the IERM. Table 2. Summary of Regulation Plans Plan ID Plan Description 77A PP MH The current release plan The current infrastructure with no regulation on releases Plan attempts to reduce variability in Lakes Michigan-Huron water levels. 121 Plan attempts to maintain Lake Superior monthly long-term average levels. HP Plan releases a constant release that favors hydropower. Because the existing IERM application is configured to simulate ecological responses associated with Lake Ontario water levels, it was necessary to adjust the water level time series obtained from the SVM to obtain appropriate water level magnitudes and fluctuations for input to the LOSL IERM application. This was accomplished based on the following approach: 1. Raw water level time series for Lake Superior (obtained from the SVM) were used to compute the quarter-monthly water level residuals relative to the low water datum

19 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 19 (LWD) for Lake Superior (183.2 meters). (For example, a water level of meters would translate to a residual of 0.4 meter.) 2. The residual water level time series were then added to the Lake Ontario low water datum (74.2 meters) to obtain the water level magnitude in the context of the Lake Ontario system. Based on this approach, intra- and inter-annual water level fluctuations were preserved relative to the original Lake Superior time series. 3. A second set of water level time series were developed to represent conditions in the Thousand Islands area of the upper St. Lawrence River by subtracting 0.14 meter from the Lake Ontario time series. 4. The two sets of water level time series (corrected to the Lake Ontario LWD and adjusted for the Thousand Islands area) were then input to the IERM to drive the ecological submodels developed for the Lake Ontario and upper St. Lawrence River system. The historical and climate change 2A net basin supply scenarios represent different time periods. The historical climate condition corresponds to the period of (107 years), while the climate change scenario is represented by a 50-year period corresponding to An example of the difference in water level resulting from these supplies is shown in Figure 7. This figure compares the water levels under current conditions (Plan 77A) for the historical supply (blue line) compared to the climate change supply (red line) applied to the same historical time period ( ). Note that the climate change scenario produces significantly lower water levels for this time period. Figure 7. Time Series Plot Comparing Lake Superior Water Levels under the Existing Regulation Plan (77A) for Historical and Climate Change Supply Scenarios The full suite of performance indicators included in the original IERM application for the LOSL system include indicators for Lake Ontario and various locations within the upper and lower St.

20 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 20 Lawrence River reaches. However, only the performance indicators for Lake Ontario and the Thousand Islands area located at the upper extent of the St. Lawrence River were included in the IERM simulations based on Lake Superior water levels. The rationale for adopting this approach was two-fold: 1. Performance indicators for Lake Ontario and the Thousand Islands area of the upper river (IJC Shoreline Unit R1) are driven by raw or adjusted Lake Ontario levels. However, St. Lawrence River locations downstream of the Thousands Islands area require more complex hydrologic information, including release flows at various downstream dam structures. These conditions could not be specified when testing the Lake Superior water level regime. 2. Performance indicators downstream of the Thousand Islands area represent ecological responses to riverine or quasi-riverine conditions that are not reflective of the conditions experienced in the Upper Great Lakes, including Lake Superior. Based on this rationale, all LOSL ecological performance indicators for locations downstream of the Thousand Islands area (IJC Shoreline Unit R1) were excluded from the analysis. Results & Discussion As highlighted previously in this memorandum, the IERM framework allows users to quickly compare the response of many performance indicators under for a variety of regulation plans and basin supply scenarios. This is accomplished by aggregating (via average, sum, quartile, etc.) the annual response into a single performance metric. The aggregated scores are displayed on a target diagram, which is generated based on the calculated ratio of user-selected baseline and alternative plans. For the Lake Superior sensitivity analysis, the current release plan (77A) is a logical choice to represent a default baseline plan in the target visualization. An example comparing the 19 key LOSL Lake Ontario and upper St. Lawrence River PI results for the MH release plan relative to the 77A plan is shown in Figure 8 for the historical basin supply scenario. This example shows that several of the fish PIs respond relatively poorly to the Michigan-Huron release plan, while many of the other ecological PIs appear to respond favorably.

21 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 21 Figure 8. Target Diagram Comparing PI Results for Plan 77A and the Michigan-Huron Plan under Historical Basin Supply Conditions The target diagram portion of the interface compares the aggregate response scores of the release plans for each PI. However, there can be significant temporal variation in the responses that is not apparent from the target diagrams. The time series plots can be used to investigate these differences. Figure 9 compares the temporal results for the wetland meadow marsh vegetation PI between the baseline (77A) and Michigan-Huron plans under historical supply conditions. This plot indicates the strong positive response that low water level has on the PI. In each of the low water periods, the meadow marsh PI responds favorably by taking over areas vacated by emergent vegetation.

22 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 22 Figure 9. Time-Series Plots Comparing Meadow Marsh PI Results for Plan 77A and the Michigan-Huron Plan (top panel). Simulated water levels for each plan are shown in bottom panel. Figure 10 shows the same meadow marsh PI for the baseline release plan under differing basin supply scenarios. Again, the consistent low water levels appear to favor this PI, although the differences are less pronounced than under the intermittent changes in water level seen in Figure 9. Conversely, the fish habitat supply performance indicator (Pugnose Shiner spawning habitat supply), shown in Figure 11, displays an adverse response to the lower water levels in the climate change scenario. Similar to findings from the LOSL Study, these results demonstrate that there are likely to be both positive and negative PI responses to any given regulation plan; therefore, it is necessary to evaluate the body of results in the context of the relative significance, certainty, and sensitivity associated with each individual PI. Unexpected results are generated for some PIs (e.g., muskrat house density) under the climate change scenario because this supply scenario results in water levels that are generally out of bounds for the algorithms that drive these indicators. Another cause for unusual results is the inappropriateness of initial conditions used for some PIs (e.g., meadow marsh vegetation). The results presented here for the application of the LOSL version of the IERM to Lake Superior water level conditions demonstrate the value of the IERM framework in assessing the response and sensitivity of various ecological performance indicators in the Upper Great Lakes system. The results shown in Figures 8-11 indicate that there is potential for identifying sensitivities for at least a subset of the performance indicators that were developed for the LOSL study. However, these results are only intended to be illustrative, and any specific conclusions based on these results should be avoided at this time because of the limitations described above.

23 Lessons Learned from the Ecological Modeling Conducted for the IJC LOSL Study page 23 Figure 10. Time Series Plot Comparing Meadow Marsh PI Results for Plan 77A for Historical and Climate Change Supply Scenarios (top panel). (Simulated water levels for each plan are shown in bottom panel.) Figure 11. Time Series Plot Comparing Fish Habitat PI Results for Plan 77A for Historical and Climate Change Supply Scenarios (top panel). (Simulated water levels for each plan are shown in bottom panel.)