STUDY PERFORMANCE REPORT State: Michigan Project No.: Study No.: 237000 Title: Development of a GIS for Inventory, Classification, and Management of NonGame Wildlife in Great Lakes Waters. Period Covered: T-10-T-3 October 1, 2011 to September 30, 2012 Study Objective: The overall goal of this project is to provide a geospatial tool to assist in developing management plans to implement the MWAP. This goal will be addressed by meeting the following objectives: Objective 1. To develop an ecological database containing information on species distributions and aquatic habitats of the Great Lakes. Objective 2. To conduct ecological classification of nearshore and offshore Great Lakes habitats in Lakes Huron, Superior and Ontario, and compare with existing classifications of nearshore zones and streams. Objective 3. To determine suitable indices of relative habitat quality for sensitive life stages of important non-game species. Objective 4. To develop GIS-based decision support projects to facilitate evaluation of potential impacts to non-game wildlife habitat from a variety of pressures. Objective 5. To develop and deliver educational programs for end-users on the use of GIS-based decision support projects for science inquiry and management. Objective 6. To develop and implement long-term, Internet-based strategies for project maintenance and distribution. Summary: Jobs 2-6 were active this year. We made progress on all active jobs. We continued to update fish spawning locations in the Goodyear et al. (1982) spawning atlas, and revised spawning habitat suitability models and maps for 5 native fish species. We added several fisheries data layers to the Lakebed Alteration Decision Support Tool (LADST), expanded it to all Great Lakes waters, and made it available on the internet. We provided technical GIS support for for several members of the Michigan DNR and DEQ, and made conference presentations, meetings, and webinars about the Great Lakes GIS, the Great lakes Aquatic Habitat Framework (GLAHF), and the LADST. Findings: Work was scheduled for Jobs 2–6 for 2011-12, and progress is reported below. Job 2. Title: Determine indices of relative habitat quality.–During FY12, we developed data to further describe the thermal habitats of the Great Lakes coastal zone. These data included cumulative degree-days, spring rate of warming, and upwelling index (see below). We also added several other data layers including seasonal ice thickness, wave height and wave energy (NOAA GLERL), and chlorophyll a, suspended sediment, transparency, dissolved organic carbon, and submerged aquatic vegetation (satellite data layers from Michigan Technical Research Institute). In collaboration with the University of Michigan (Catherine Riseng), The Nature Conservancy (Mary Khoury, Scott Sowa, Lindsay Chadderton), and the USGS Great Lakes Science Center 1 T-10-T-3, Study 237000 (Jim McKenna, Chris Castiglione), we are adding data on zooplankton, benthos and fish densities and distributions for coastal Great Lakes waters. These data on lower watersheds and coastal and offshore waters are being attributed to the Great Lakes Aquatic Habitat Framework, a project led by Catherine Riseng (UM) to facilitate modeling, assessment and classification of Michigan’s coastal habitats. Temperature is an important metric in explaining the growth rate and development of fish. One way of relating temperature over time to growth rates over time is the cumulative degree-day (CDD, ºC∙day) metric. CDD has been used for centuries in agriculture, more recently in entomology, and only in the past decade being used to characterize ecotherm growth. Neuheimer and Taggart (2007) were able to show that variation in fish length is a strong linear function of the variation in GDD (r2≥0.92). Venturelli et al. (2010) studied populations of immature walleye in Ontario and Quebec, Canada and found 96% of the variation in length was explained by GDD, and that the GDD model outperformed the age-based equivalents. Using modeled vertical water column temperature data from the National Oceanic and Atmospheric Administration (NOAA) Great Lakes Coastal Forecasting System, the average daily 0 to 20 m water column temperature was calculated and summed (when the temperature was above 0ºC) over the entire year. The summed GDD layer was calculated for each year from 2006 through 2011 for all of the Great Lakes. Figure 1–Cumulative degree-days calculated for the Great Lakes for 2010. The rate water warms in the nearshore zone is a measure of potential success for spawning and larval fish. Bush et al. (1975) found the rate and regularity of water warming between April 1 and May 15, 1960-1970 were strongly related (r = 0.90) to year class strength for spawning and incubation periods of walleye in Lake Erie. Other studies have also found that spring rate and regularity of warming in Lake Erie and Saginaw Bay has an important role in reproductive success of walleye and yellow perch (Eshenroder 1977; Koonce et al. 1977; Shuter and Koonce 1977). NOAA’s Coastwatch program provides Great Lakes Surface Environmental Analysis 2 2 T-10-T-3, Study 237000 (GLSEA2) Sea Surface Temperature (SST) data, which were used to calculate spring rate of warming, instead of modeled vertical water temperature, since spring warming is most critical in shallow, nearshore areas where fish spawn. To calculate spring rate of warming the average daily temperature image for March 1 was subtracted from the June 1 image and divided by 93, the number of days between the chosen dates. The larger window of March 1 to June 1 was selected to capture the seasonal warming rate that has an effect on other biological processes besides walleye spawning, such as diatom blooms. The resulting layer gives the average daily rate of warming for a given pixel. This has been calculated for all of the Great Lakes from 1995 through 2011. Figure 2.–Estimated spring rate of surface water warming for the Great Lakes in 2010. Coastal upwelling occurs when cold water from the hypolimnion reaches the surface in a vertically stratified lake, potentially bringing nutrient-rich water to the surface, and thereby stimulating phytoplankton blooms and altering species distributions. Upwelling is caused by wind action driving the epilimnetic waters away from the coast until winds reduce in magnitude or change direction causing the epilimnetic water to return to the coast. Following the methods described in Plattner et al. (2006) and Wegscheider (2006), we calculated an annual upwelling index for the years 1994 through 2011. The index was calculated for all of the Great Lakes using the GLSEA2 SST data, which is an average daily surface temperature product covering all the Great Lakes from remotely sensed satellite imagery. A cumulative index was created for the entire year showing the frequency of upwelling for a given pixel for each year. 3 T-10-T-3, Study 237000 Figure 3.–An index of cumulative upwelling estimated from AVHRR surface temperature data for the Great Lakes in 2010. 4 T-10-T-3, Study 237000 We used Classification and Regression Tree (CART) models and binary logistic regression models to quantify suitability of critical spawning habitats of five selected native fishes (lake herring, emerald shiner, lake whitefish, yellow perch, and lake trout). These species were key prey or predators in historic Great Lakes food webs whose reproductive success and relative abundance have been compromised by habitat degradation and destruction, overfishing and impacts from non-native species (Wells and McLain 1973). Data Sources We quantified habitat suitability of spawning and nursery areas for several native species including Emerald shiner, Lake herring, Lake trout, Lake whitefish, and Yellow perch in coastal and open waters of the Great Lakes. The spawning presence locations of the native species were obtained from the Goodyear Spawning Atlas (GSA) (Goodyear et al. 1982) and were manually entered into a spreadsheet, which was imported as a geolocated point file using ArcGIS for Desktop 10 (ESRI, 2011). Six physical habitat datasets were geospatially represented and attributed to the native species presence locations. The physical habitat data sets included lake depth in meters, slope in degrees, substrate, distance from river mouth in kilometers, average surface temperature during spawning in °C, and cumulative degree days (CDD) in days. Lake depth by relevant lake datum was acquired from the National Oceanic and Atmospheric Administration’s (NOAA) National Geophysical Data Center (NGDC) (Great Lakes Bathymetry, 2000). Slope of the lakebed was calculated using Spatial Analyst Extension for ArcGIS with the Great Lakes Bathymetry raster as input. Distance from river mouth was calculated using the Euclidean Distance Tool in Spatial Analyst Extension for ArcGIS. Substrate was digitized by several agencies from printed maps and figures. The Lake Erie substrate layer was digitized by Ohio Sea Grant (Haltuch and Berkman, 1999); Lake Huron by the Institute for Fisheries Research (IFR) (Thomas et al., 1973; Robbins, 1986); Lake Michigan by IFR (Powers and Robertson, 1968); Lake Ontario by NOAA-GLERL (Thomas et al., 1972); Lake St. Clair by IFR (Lake St. Clair Substrate, 2000); and Lake Superior by IFR (Thomas et al., 1978). The substrate categories and corresponding integer codes are listed in Table 1 below. Table 1.–Codes used to classify Great Lakes substrate types. Substrate type Code undefined Clay Sand Mud Silt Hard 0 1 2 3 4 5 A river mouths database (Great Lakes Basin Major River Mouths, 2011) was created by locating all rivers above a 250-km2 threshold drainage area. The threshold drainage area was selected based on available nutrient loading data and additional monitoring data for these rivers. The nutrient loading data are relevant to other projects and research, and have been accepted by the Core Working Group of scientists advising the Great Lakes Environmental Assessment and Mapping Project. 5 T-10-T-3, Study 237000 Average surface temperature during spawning was calculated across the entire Great Lakes for each native species. The average surface temperature was calculated using the Great Lakes surface temperature data from NOAA GLERL CoastWatch (Great Lakes Surface Environmental Analysis, 2011) from 2008-2011. Lane and others (1996) listed the typical spawning season for each of the native species, which included emerald shiner from May to mid-July; lake herring in November; lake trout from November through December; lake whitefish from mid-October through December; and yellow perch from April through early May. The average surface temperature was calculated during the previously stated typical spawning periods. The Cumulative Degree Day (CDD) data layers for each lake were calculated using the Great Lakes Coastal Forecasting System vertical water temperature data from NOAA GLERL (Great Lakes Coastal Forecasting System, 2012). Surface temperature data were obtained for everyday of the year from 2008-2010. CDD was calculated by summing the temperature for all days of a given year. To remove annual variability, the 3 data layers were averaged to result in the final CDD used. Data Processing Data processing utilized the ArcGIS and Geospatial Modelling Environment (GME) (Beyer, 2011). All spatial data were projected to the Albers Equal Area Conic USGS version for the contiguous USA. The Great Lakes bathymetry layer was used as the masking, extent, and cell size raster for standardizing the other 4 physical habitat data layers. The CDD layer required extra processing since the shoreline had large data gaps when overlayed with the bathymetry layer. A raster calculator was used to fill the gaps in the CDD layer. The spawning location dataset contains only confirmed presences of spawning and not absence data. In order to statistically model spawning habitat suitability, absence points are needed. After reviewing literature on binary logistic regression with presence-absence data, we confirmed that we could create “pseudo-absence” data for each species. Using a binary logistic regression model, Zaniewski et al. (2002) found that environmentally weighted, pseudo-absence data generated for generalized additive models results in models that are closely correlated with true presenceabsence data, and perform better than random absence data or ecological niche factor analysis. For our study, we created environmentally-weighted, pseudo-absence data based on spawning depth information for each of the 5 native fish species. We created an equal number of absence points as there were presence points. We randomly distributed absence points at depths >30 m for lake whitefish and yellow perch; >100 m for lake trout and lake herring; and depths >10 m for emerald shiner. For the analysis of spawning habitat suitability using classification and regression trees (CART), we created an exclusion zone of 5 kilometers around each present spawning location, and then removed the exclusion zone from the waters of the Great Lakes. The remaining areas of the Great Lakes were entered into GME’s Generate Random Stratified Points Tool to randomly create absence points. We then generated a second set of absence points for the CART model, so that the environmental weighting by depth did not drive the first division of the classification tree. We again created an equal number of absence points as there were presence points. We coded presence and absence points with 1 and 0, respectively. Once we prepared the physical habitat data layers and the species presence and absences points, we attributed the spawning point data with the physical habitat data found at each point. In GME, the “isectpntrst” command attributed each point with a new column for each of the physical habitat variables. We then reviewed the point layer in ArcGIS to check for null values. Any null values found were attributed with nearby information or removed from the calculations as 6 T-10-T-3, Study 237000 appropriate. The point layer at this time was prepared for statistical analysis, which requires exporting the data table to .dbf and converting to a .csv file. Binary Logistic Regression Model Development We used binary logistic regression to calculate spawning site habitat suitability of native species in the Great Lakes. We used R for Statistical Computing software (R Development Core Team, 2011) for statistical calculations. The .csv files created above were import into R and a binary logistic regression model was developed using the generalized linear model command and the complementary log-log function. The canonical logit function was also tested, but consistently had a higher residual error value, therefore we used the complementary log-log function for all of the models. The “step” function in R allowed for selection of the optimal model based on the Akaike information criterion (AIC) for each of the 5 native species. The probability was then calculated for each presence and absence point in a spreadsheet using the equations below. P(z) = 1 – EXP( -EXP(z) ), where z = β0 + β1x1 + β2x2 + … + βnxn The optimal probability value was selected from the calculated probability values by determining the highest number of correctly classified presence points and the least incorrectly classified absence points. For each native species, a habitat suitability area was calculated using the 6 standardized physical habitat rasters and the above probability equations. The calculation was performed in ArcGIS using Raster Calculator Tool. The output probability rasters were then reclassified using the Reclassify Tool in ArcGIS into areas of 0 (not suitable) or 1 (suitable). CART Model Development Relationships between spawning site location and physical habitat variables were also assessed statistically using classification and regression trees (CART). CART has been used extensively in the past decade to explore ecological relationships with habitat variables on the landscape scale, particularly in the Great Lakes region (Stanfield 2006; Brewer 2007; Steen et al. 2008). CART has become prevalent for landscape scale analysis because of its ability to utilize an assortment of response types, efficiently process large datasets, handle missing values, and the ease of construction and interpretation (Breiman et al. 1984; De’ath and Fabricius 2000; De’ath 2002). CART® data mining software, part of the Salford Systems Predictive Mining Suite (Salford Systems, Inc. 2011; Steinberg and Colla 1997), was used to develop the presence-absence classification trees to predict suitable spawning habitat for the 5 native fish species. Due to the relatively small sample size of the data sets (N < 1,000) the v-fold cross validation was selected as the testing method. The 10-fold cross validation creates 10 random divisions of the dataset into 90% learning and 10% test sets using the “Gini” splitting rule. The minimum node size for a parent node was set to 10 cases, and any terminal node must include at least 5 cases. Setting these node size parameters removes many of the over grown trees which were too specific to describe the entire population. The standard error (SE) rule for selection of best tree was set to 1 SE where the estimated error rate is within 1 SE of the minimum. Classification trees with less than 10 nodes were selected to ensure the results captured only the variables that were accurate for the entire population and to simplify ecological interpretation of the results. Once the most appropriate tree was selected, the presence-absence classification trees were evaluated for prediction success. Overall percent accuracy and the relative operating 7 T-10-T-3, Study 237000 characteristic (ROC) were reported to aid in evaluating the classification trees success. Individual classification trees were generated for each of the 5 native species and the results for the top suitable habitats and were mapped in GIS. Results Binary Logistic Regression Model Results Spawning habitat suitability models for lake herring and lake trout (not in Lake Michigan) were statistically significant for 5 of the 6 physical habitat variables, excluding substrate. For lake trout only in Lake Michigan, depth and average surface temperature were included in the model. Habitat suitability models for lake whitefish were statistically significant for depth, slope, CDD, and average surface temperature during spawning. The model for Emerald shiner was significant for depth, slope, and substrate, while the model for Yellow perch was significant for depth, slope, and average surface temperature. The best model selected correctly classified more than 93% of the presence values, and misclassified less than 0.5% of the absence values. Table 2 shows the optimal logistic regression models for each native species and their associated optimal probability values. Table 2.–Binary logistic regression results for select native fish species in open waters of the Great Lakes with associated optimal probability values. Habitat variables include: ‘D’ = depth (m); ‘S’ = slope (degrees); ‘SUB’ = substrate (coded value); ‘DIST’ = distance from major river; ‘CDD’ = cumulative degree-days (days); ‘ST’ = average surface temperature during typical spawning period (°C). Species Complementary log-log logistic regression Probability Emerald shiner 395.68 + -18.36 * D + 93.55 * S + -46.65 * SUB >0.90 Lake herring 33.854508 + -0.155388 * D + 0.484957 * S + -1.358152 * ST + -0.004168 * CDD + -0.021223 * DIST >0.90 Lake trout (no Lake Michigan) 20.53929 + -0.09293 * D + 0.38612 * S + -0.46421 * ST + 0.00409 * CDD + -0.01808 * DIST >0.90 Lake trout (Lake Michigan) 2.0415663 + -0.1360673 * D + 1.3356317 * ST >0.90 Lake whitefish 2.621721 + -0.0749071 * D + 0.4307344 * S + -0.2582083 * ST + 0.0003922 * CDD >0.80 Yellow perch 1.2631 + -0.1097 * D + 0.7101 * S + 0.3195 * ST >0.90 Habitat suitability maps were created for each of the 5 native species utilizing the best logistic regressions determined for each species listed in Table 2. The resulting map of continuous probability values was reclassified into suitable areas, which were areas with a probability higher than the value listed in Table 2, and the remaining values were classified unsuitable. Figures 4 through 8 below show reclassified suitable spawning habitat areas in red for each of the 5 native species of interest. Note that there is a model specific to lake trout in Lake Michigan. Lake trout spawning sites were in Lake Michigan were often found in deeper waters than in the other lakes 8 T-10-T-3, Study 237000 or even for other species. Instead of forcing the lake trout spawning data to fit a model for all Great Lakes, we developed a separate analysis for lake trout in Lake Michigan (Table 2). The suitability map was calculated for Lake Michigan only and mosaicked with the lake trout suitability map for the other four lakes. Table 3 gives statistical values for each of the physical habitat variables in the most suitable areas for each native species. The majority of the suitable habitat area for each physical habitat variable for each native species explained in the following text is defined as the values ±1 standard deviation from the mean. Spawning habitat variables for emerald shiner included: optimal depth ranged from 7.9-13.9 m; slope ranged from 0-1.26 degrees; and substrates included ‘hard’, ‘silt’, and ‘clay’. Lake herring suitable spawning habitat included: a range of 4-42 m in depth; 0-1.37 degrees of slope; 2,503-3,563 days CDD; average surface temperature during spawning between 7-9.4 °C; and 13.5-60.3 km from a river mouth. Lake trout suitable spawning habitat in Lakes Erie, Huron, Ontario, and Superior ranged from 3.8-43.6 m of depth; 0-1.7 degrees of slope; 2,322-3,244 days CDD; average surface temperature during spawning between 4.8-6.8 °C; and 13.3-58.5 km from a river mouth. Lake trout suitable spawning habitat in Lake Michigan ranged from 12.3-62.5 m of depth; and an average surface temperature during spawning between 5.6-8.0 °C. Lake whitefish suitable spawning habitat ranged from 3.2-17.6 m of depth; 0-1.23 degrees of slope; 2,687-3,819 days CDD; and average surface temperature during spawning between 6.1-8.3 °C. Yellow perch suitable spawning habitat ranged from 3.3-16.7 m of depth; 0-1.13 degrees of slope; and an average surface temperature during spawning between 2.9-7.3 °C. 9 T-10-T-3, Study 237000 Table 3.–Statistics describing physical habitat variables of suitable spawning areas for emerald shiner (ES), lake herring (LH), lake trout (LT), lake whitefish, and yellow perch (YP). Variable ES depth mean (s.d.) 7.9 (6.0) min 0 max 146.4 slope mean (s.d.) 0.40 (0.86) min 0 max 29.0 CDD mean (s.d.) – min – max – ST mean (s.d.) – min – max – DIST mean (s.d.) – min – max – hard, silt, Substrate types clay LT in Lake Michigan LW YP 23.0 (19.0) 23.7 (19.9) 0 0 200.9 212.8 37.4 (25.1) 0.1 94.6 10.4 (7.2) 0 203.5 10.0 (6.7) 0 203.5 0.47 (0.90) 0.61 (1.09) 0 0 32.3 32.7 3,033 (530) 2,783 (461) 1,995 1,995 4,251 4,251 8.3 (1.1) 5.8 (1.0) 4.3 2.4 11.6 9.1 – – – 0.38 (0.85) 0 32.7 0.32 (0.81) 0 32.7 – – – 6.8 (1.2) 3.3 9.0 3,253 (566) 1,995 4,251 7.2 (1.1) 3.6 10.4 – – – 5.1 (2.2) 1.2 11.4 36.9 (23.4) 35.9 (22.6) 0 0 141.5 141.5 – – – – – – – – – – – – LH – LT – 10 T-10-T-3, Study 237000 Figure 4.–Emerald shiner suitable habitat area for spawning in the Great Lakes (red). Following the binary logistic model results in this study, there are potentially 29,038 km2 of area suitable for emerald shiner to spawn. Spawning presence locations are shown as a “ + ”. 11 T-10-T-3, Study 237000 Figure 5.–Lake herring suitable habitat area for spawning in the Great Lakes (red). Following the binary logistic model results in this study, there are potentially 74,439 km2 of area suitable for lake herring to spawn. Spawning presence locations are shown as a “ + ”. 12 T-10-T-3, Study 237000 Figure 6–Lake trout suitable habitat area for spawning in the Great Lakes (red). Following the binary logistic model results in this study, there are potentially 43,039 km2 of area suitable for lake trout to spawn in Lakes Superior, Huron, Erie and Ontario with an additional area of 30,488 km2 suitable in Lake Michigan for a combined area of 73,527 km2. Spawning presence locations are shown as a “ + ”. 13 T-10-T-3, Study 237000 Figure 7.–Lake whitefish suitable habitat area for spawning in the Great Lakes (red). Following the binary logistic model results in this study, there are potentially 45,614 km2 of area suitable for lake whitefish to spawn. Spawning presence locations are shown as a “ + ”. 14 T-10-T-3, Study 237000 Figure 8.–Yellow perch suitable habitat area for spawning in the Great Lakes (red). Following the binary logistic model results in this study, there are potentially 43,534 km2 of area suitable for yellow perch to spawn. Spawning presence locations are shown as a “ + ”. 15 T-10-T-3, Study 237000 CART Model Results The classification tree results indicated the presence or absence of suitable spawning habitat for each of the 5 native fish species (Figures 9-18, Table 4). The overall accuracy of all of the classifications trees was above 90% with high ROC training and testing scores. The variables of importance consistently included: depth, cumulative degree-days (CDD), and slope, in varying order. Yellow perch spawning habitat suitability was mapped using the habitat variables in GIS following 2 suitable pathways. Approximately 97% of the presence locations were correctly classified as suitable, and 3.4% of the absence locations were misclassified as unsuitable. All other species were mapped following 3 suitable pathways including emerald shiner (95.3% presence classified, 3.5% absence misclassified); lake herring (92.1% presence classified, 11.0% absence misclassified); lake trout (90.0% presence classified, 10.3% absence misclassified); and lake whitefish (90.8% presence classified, 9.4% absence misclassified). Table 4–Final results of the cross-validated classification tree analysis of suitable spawning habitat for 5 native fish species in the Great Lakes. Species Emerald Shiner Lake Herring Lake Trout Lake Whitefish Yellow Perch No. of nodes 5 8 9 6 5 Relative cost 0.186 0.258 0.200 0.215 0.134 Overall % correct 95.93 91.92 90.78 90.70 94.15 16 ROC (train / test) 0.9761 / 0.9545 0.9398 / 0.9127 0.9336 / 0.9265 0.9394 / 0.9284 0.9486 / 0.9424 Variables of importance (in order) Depth, CDD, Slope Depth, CDD, Slope Depth, CDD, Slope Depth, Slope, CDD Depth, Slope, CDD T-10-T-3, Study 237000 Class Cases % 0 86 50.0 1 86 50.0 BATH <= 9.95 BATH > Class Cases % 0 9 9.7 1 84 90.3 CDD2 > 3563.09 Class Cases % 0 8 25.0 1 24 75.0 Class Cases % 0 1 1.6 1 60 98.4 RVRMTH_KM > 21.44 Class Cases % 0 1 6.3 1 15 93.8 Class Cases % 0 7 43.8 1 9 56.3 SU2 Class Cases % 0 77 97.5 1 2 2.5 CDD2 <= 3563.09 RVRMTH_KM <= 21.44 RVRMTH_KM <= 41.25 RVRMTH_KM > 41.25 Class Cases % 0 6 75.0 1 2 25.0 Class Cases % 0 1 12.5 1 7 87.5 9.95 SU1 SU3 Figure 9.–Final cross-validated classification tree for emerald shiner predicted suitable spawning habitat in the Great Lakes. Splitting variables at each decision node of the tree are: ‘BATH’ = depth (m); ‘CDD2’ = cumulative degree-days; and ‘RVRMTH_KM’ = distance to river mouth (km) with the values for each variable displayed next to the variable name. ‘Class’ represents 0 for absence sites and 1 for presence sites. ‘Cases’ describes the number and percentage of sites that meet the splitting criteria within the node. The mapped suitable habitats are listed below the nodes in order of most suitable (SU1-SU3). 17 T-10-T-3, Study 237000 Figure 10.–Predicted suitable spawning habitat map for emerald shiner in the Great Lakes produced from the classification tree model. Suitable habitat 1 (SU1) is displayed in red, SU2 in purple, and SU3 in orange correspond the SU designations in the classification tree. Spawning presence locations are shown as a “ + ”. The predicted suitable spawning habitat covers a total area of 16,509 km2. 18 T-10-T-3, Study 237000 Class Cases % 0 291 50.1 1 290 49.9 BATH <= 17.45 BATH > 17.45 Class Cases % 0 36 13.2 1 236 86.8 Class Cases % 0 255 82.5 1 54 17.5 SLP <= 0.04 SLP > Class Cases % 0 16 39.0 1 25 61.0 BATH <= 6.50 Class Cases % 0 6 20.7 1 23 79.3 SU2 BATH > 0.04 Class Cases % 0 20 8.7 1 211 91.3 6.50 Class Cases % 0 10 83.3 1 2 16.7 SU1 BATH <= 73.95 BATH > 73.95 Class Cases % 0 88 62.9 1 52 37.1 Class Cases % 0 167 98.8 1 2 1.2 CDD2 <= 2717.13 CDD2 > 2717.13 Class Cases % 0 6 15.4 1 33 84.6 Class Cases % 0 82 81.2 1 19 18.8 SU3 SLP <= 0.93 Class Cases % 0 77 89.5 1 9 10.5 SLP > 0.93 Class Cases % 0 5 33.3 1 10 66.7 BATH <= 37.50 BATH > 37.50 Class Cases % 0 1 10.0 1 9 90.0 Class Cases % 0 4 80.0 1 1 20.0 Figure 11.–Final cross-validated classification tree for Lake herring predicted suitable spawning habitat in the Great Lakes. Splitting variables at each decision node of the tree are: ‘BATH’ = depth (m); ‘CDD2’ = cumulative degree-days; ‘SLP’ = slope (degrees); and ‘RVRMTH_KM ‘= distance to river mouth (km) with the values for each variable displayed next to the variable name. ‘Class’ represents 0 for absence sites and 1 for presence sites, and ‘cases’ describes the number and percentage of sites that meet the splitting criteria within the node. The mapped suitable habitats are listed below the nodes in order of most suitable (SU1-SU3). 19 T-10-T-3, Study 237000 Figure 12.–Predicted suitable spawning habitat map for lake herring in the Great Lakes produced from the classification tree model. Suitable habitat 1 (SU1) is displayed in red, SU2 in purple, and SU3 in orange correspond the SU designations in the classification tree. Spawning presence locations are shown as a “ + ”. The predicted suitable spawning habitat covers a total area of 44,724 km2. 20 T-10-T-3, Study 237000 Class Cases % 0 1084 50.0 1 1085 50.0 BATH <= 39.95 BATH > 39.95 Class Cases % 0 311 24.0 1 986 76.0 Class Cases % 0 773 88.6 1 99 11.4 CDD2 <= 3529.43 CDD2 > 3529.43 SLP <= 3.89 Class Cases % 0 168 14.7 1 974 85.3 Class Cases % 0 143 92.3 1 12 7.7 Class Cases % 0 764 91.6 1 70 8.4 SLP <= 0.24 SLP > Class Cases % 0 102 33.7 1 201 66.3 Class Cases % 0 66 7.9 1 773 92.1 BATH <= 15.80 BATH > 15.80 Class Cases % 0 37 17.5 1 175 82.5 Class Cases % 0 65 71.4 1 26 28.6 SU2 0.24 CDD2 <= 2923.02 CDD2 > 2923.02 Class Cases % 0 8 34.8 1 15 65.2 Class Cases % 0 57 83.8 1 11 16.2 SU1 SLP > BATH <= 59.50 BATH > 59.50 Class Cases % 0 97 68.8 1 44 31.2 Class Cases % 0 667 96.2 1 26 3.8 SLP <= 0.93 Class Cases % 0 89 79.5 1 23 20.5 SLP > 3.89 Class Cases % 0 9 23.7 1 29 76.3 SU3 0.93 Class Cases % 0 8 27.6 1 21 72.4 Figure 1.–Final cross-validated classification tree for lake trout predicted suitable spawning habitat in the Great Lakes. Splitting variables at each decision node of the tree are: ‘BATH’ = depth (m); ‘CDD2’ = cumulative degree-days; ‘SLP’ = slope (degrees); and ‘RVRMTH_KM’ = distance to river mouth (km) with the values for each variable displayed next to the variable name. ‘Class’ represents 0 for absence sites and 1 for presence sites, while ‘cases’ describes the number and percentage of sites that meet the splitting criteria within the node. The mapped suitable habitats are listed below the nodes in order of most suitable (SU1-SU3). 21 T-10-T-3, Study 237000 Figure 14.–Predicted suitable spawning habitat map for lake trout in the Great Lakes produced from the classification tree model. Suitable habitat 1 (SU1) is displayed in red, SU2 in purple, and SU3 in orange correspond the SU designations in the classification tree. Spawning presence locations are shown as a “ + ”. The predicted suitable spawning habitat covers a total area of 44,074 km2. 22 T-10-T-3, Study 237000 Class Cases % 0 650 50.0 1 651 50.0 BATH <= 16.45 BATH > 16.45 Class Cases % 0 74 11.7 1 559 88.3 Class Cases % 0 576 86.2 1 92 13.8 BATH <= 9.45 Class Cases % 0 34 6.7 1 472 93.3 SU1 BATH > 9.45 Class Cases % 0 40 31.5 1 87 68.5 BATH <= 55.20 BATH > 55.20 Class Cases % 0 155 66.0 1 80 34.0 Class Cases % 0 421 97.2 1 12 2.8 CDD2 <= 3430.63 CDD2 > 3430.63 SLP <= 1.15 Class Cases % 0 16 17.0 1 78 83.0 Class Cases % 0 24 72.7 1 9 27.3 Class Cases % 0 144 78.7 1 39 21.3 SU2 SLP > 1.15 Class Cases % 0 11 21.2 1 41 78.8 SU3 Figure 15.–Final cross-validated classification tree for lake whitefish predicted suitable spawning habitat in the Great Lakes. Splitting variables at each decision node of the tree are: ‘BATH’ = depth (m); ‘CDD2’ = cumulative degree-days; and ‘SLP’ = slope (degrees) with the values for each variable displayed next to the variable name. ‘Class’ represents 0 for absence sites and 1 for presence sites, and ‘cases’ describes the number and percentage of sites that meet the splitting criteria within the node. The mapped suitable habitats are listed below the nodes in order of most suitable (SU1-SU3). 23 T-10-T-3, Study 237000 Figure 16.–Predicted suitable spawning habitat map for lake whitefish in the Great Lakes produced from the classification tree model. Suitable habitat 1 (SU1) is displayed in red, SU2 in purple, and SU3 in orange correspond the SU designations in the classification tree. Spawning presence locations are shown as a “ + ”. The predicted suitable spawning habitat covers a total area of 35,077 km2. 24 T-10-T-3, Study 237000 Class Cases % 0 410 50.0 1 410 50.0 BATH <= 11.55 BATH > 11.55 Class Cases % 0 31 7.6 1 376 92.4 Class Cases % 0 379 91.8 1 34 8.2 SU1 BATH <= 32.00 BATH > 32.00 Class Cases % 0 69 71.9 1 27 28.1 Class Cases % 0 310 97.8 1 7 2.2 SLP <= 0.20 SLP > Class Cases % 0 54 94.7 1 3 5.3 0.20 Class Cases % 0 15 38.5 1 24 61.5 CDD2 <= 3242.28 CDD2 > 3242.28 Class Cases % 0 11 78.6 1 3 21.4 Class Cases % 0 4 16.0 1 21 84.0 SU2 Figure 2.–Final cross-validated classification tree for suitable spawning habitat of yellow perch predicted in the Great Lakes. Splitting variables at each decision node of the tree are: ‘BATH’ = depth (m); ‘CDD2’ = cumulative degree-days; and ‘SLP’ = slope (degrees) with the values for each variable displayed next to the variable name. ‘Class’ represents 0 for absence sites and 1 for presence sites, while ‘cases’ describes the number and percentage of sites that meet the splitting criteria within the node. The mapped suitable habitats are listed below the nodes in order of most suitable (SU1 and SU2). 25 T-10-T-3, Study 237000 Figure 18.–Predicted suitable spawning habitat map for yellow perch in the Great Lakes produced from the classification tree model. Suitable habitat 1 (SU1) is displayed in red, and SU2 in purple correspond the SU designations in the classification tree. Spawning presence locations are shown as a “ + ”. The predicted suitable spawning habitat covers a total area of 31,462 km2. 26 T-10-T-3, Study 237000 Information on characteristics of suitable spawning habitat is critical for restoration and conservation of SGCN. First, the information allows prediction of spawning suitability for SGCN over large spatial scales, thus permitting prioritization of spawning sites for conservation or restoration. Second, knowledge of spawning habitat characteristics allows forecasting of anthropogenic and climatic stressors that may influence habitat suitability. Finally, the information allows prediction of suitable habitat for species in areas that haven’t been sampled. Job 3. Title: Develop GIS-based decision support projects.–In FY 12, we enhanced and extended the Lakebed Alteration Decision Support Tool (LADST) from Michigan’s coastal waters to coastal and offshore waters of the Great Lakes. The LADST was developed through funding from Michigan DEQ and US Dept of Energy to assist resource managers in making siting decisions in Michigan’s and US coastal waters for various forms of lakebed alterations, including dredging, drilling, and offshore wind farms, which may impact species of special concern. Users of the LADST can create their own siting suitability maps, based upon criteria of their own choosing, just by visiting a public web site (http://glgis.org/ladst). The LADST is novel in that it makes this type of customized suitability analysis easily accessible to users who have no specialized software and no experience with geographic information systems (GIS). The LADST thus facilitates protection of species of special concern while increasing transparency of the siting and permitting process to resource managers, developers, and concerned citizens. In FY 12, we successfully obtained data and processed many of these data layers. The current version of the LADST for the all waters of the Great Lakes uses data layers with the following list of themes to make suitability maps according to user-supplied criteria: airports, EPA areas of concern, fish spawning sites for several species, islands, national parks, national lakeshores, recommended course lines (i.e. shipping lanes), shorelines, special use airspace, trout refuges, water depth, and wind speed. During the course of the grant period, other possible data layers were investigated, but were determined to be presently unavailable. For example, the U.S. Fish and Wildlife Service has a number of ongoing projects which will produce, in the near future, data sets related to the use of coastal and offshore areas by birds and bats. The LADST is designed to be able to incorporate these new data sets easily, so we anticipate being able to add such data sets soon after they become available. The LADST is now online ((http://glgis.org/ladst), and is being used by the Great Lakes community to inform policy on wind farm sitings in Michigan’s coastal waters (funded by Michigan DEQ coastal zone management program), in US waters of the Great lakes (funded by US Dept. of Energy), and finally for the entire Great Lakes (partially funded by Sierra Club and NOAA). Lacey Mason and Jason Breck played key roles in developing the spatial framework behind Great Lakes Aquatic habitat Framework (‘GLAHF’ PI Catherine Riseng), funded by the Great Lakes Fishery Trust). They developed scrips to efficiently process and attribute different data types to the framework, and figured out how to put the habitat framework and data online. The framework and associated data will permit classification and assessment of Coastal and offshore Great Lakes habitats with added benefits of targeting high priority conservation areas and informing environmental niche and food web models. Lacey’s work on developing habitat layers for the GLAHF project has formed the basis of input data for environmental niche models, which forecast where new invasive species will go in the Great Lakes and will support development of habitat suitability models for native species. During FY 12, we made progress on assessing conditions of the Great Lakes coastal areas by soliciting expert opinions, and through development of the environmental data layers mentioned above. We will use the additional data layers in the GLAHF project to conduct an assessment of the health of Great Lakes coastal waters through collaboration with the National Fish Habitat Partnership Program. 27 T-10-T-3, Study 237000 The value of the LADST and GLAHF for conservation and protection of SGCN habitat is clear. The tool may be used by managers wishing to protect habitats for SCGN spawners and other life stages. Job 4. Title: Develop and deliver educational programs for end-users.–Below is a list of workshops and presentations made by Lacey Mason, Jason Breck, or Ed Rutherford to demonstrate the databases and tools discussed above. GIS programmer J. Breck and GIS analyst L. Mason continued to be available for direct consultation with end-users regarding data and GIS project use. They answered data requests by MDNR employees and other research institutes and resolved technical issues with ArcGIS program use. L. Mason provided assistance to the Marquette Fisheries Research Station in importing images of Lake Superior bottom depth and substrates collected by sidescan sonar, and provided information to the office regarding Fish Division’s ArcGIS 9.2 licenses available through Citrix Server. L. Mason provided assistance to James Breck to map cormorant colonies in Lake Michigan, and to calculate the area for potential feeding near Ludington, MI and Beaver Island, MI. L. Mason assisted the Alpena and Lake St. Clair MDNR fisheries research stations (FRS) in mapping and figure development. D. Fielder (Alpena FRS) requested graphics of Lake Huron and Saginaw Bay for publications and presentations. Fielder also requested help with updating walleye tag return maps for Lake Huron, plotting data from 1981 through 2011. J. Johnson and J. Zellinger (Alpena FRS) requested mapping support for plotting Lake Huron nearshore fish survey and updating the Research Stations’ copies of the Great Lakes GIS database. J. Breck updated a program that processes lake temperature and dissolved oxygen profile data in order to extract several kinds of information that are relevant to management decisions. The new program produces a spreadsheet and a series of graphs that show the temperature and dissolved oxygen profiles for each lake, along with calculated values such as the depths of the metalimnion and hypolimnion and the depth at which dissolved oxygen concentration equals 2 ppm. Based upon these values, the program also classifies the lake following (Schneider, 1975). Managers can use this information to quickly judge the suitability of the lake for various fish species. On 28-29 Nov, 2012, Jason Breck ran the LADST ‘on the fly’ at a Great Lakes Wind Collaborative workshop at NOAA GLERL to help industry, resource managers and non-profits evaluate wind farm impacts on fisheries. Education and outreach of decision support tools, habitat databases, maps, and spawning habitat suitability serve to inform stakeholders about critical spawning and nursery areas for SGCN, and to inform management decisions about conservation opportunities. Job 5. Title: Develop and implement long-term, Internet-based strategies for project maintenance and distribution.–During FY12 we continued to pursue means of releasing the public website for the Great Lakes GIS Online and the LADST including hosting at the Great Lakes Commission. We continued to engage collaborators at the University of Michigan, NOAA-GLERL, Wisconsin Sea Grant, Ohio DNR, Sierra Club and others to develop a Great Lakes Coastal Atlas. Internet distribution of maps, databases and decision support for SGCN will serve to educate the public about critical habitat needs for SGCN and priorities for SGCN conservation. Internet distribution also will greatly facilitate distribution and maintenance of the databases and tools. 28 T-10-T-3, Study 237000 Job 6: Title: Write Annual Report and Journal Publications.–This annual report was submitted. Final analysis and revision is in progress of a journal publication titled: “Spatial and Temporal Variation in Movements of Coho Salmon Inferred from Angler Catch Rate Patterns in Lake Michigan, 1992-2001”. We anticipate submitting the paper to Transactions of the American Fisheries Society in February 2014. We also plan to submit a paper on “Ecological classification of open water habitats in Lake Michigan” for peer review in Journal of Great Lakes Research by June 2013. 29 T-10-T-3, Study 237000 References Beyer, H. L. 2011. Geospatial Modelling Environment: Release 0.5.3 Beta. Spatial Ecology, LLC. Breiman, L., J. Friedman, R. Olshen, and C. Stone. Classification and Regression Trees. Pacific Grove: Wadsworth, 1984.Brewer, S. K., C. F. Rabeni, S. P. Sowa, and G. Annis. 2007. Natural landscape and stream segment attributes influencing the distribution and relative abundance of riverine smallmouth bass in Missouri. North American Journal of Fisheries Management. 27:326-341. De’ath, G., and K. E. Fabricius. 2000. Classification and regression trees: a powerful yet simple technique for ecological data analysis. Ecology 81:3178-3192. De’ath, G. 2002. Multivariate regression trees: a new technique for modeling species-environment relationships. Ecology 83:1105-1117. Esri. 2011. ArcGIS for Desktop: Release 10. Esri, Redlands, CA. Goodyear, C.S., T.A. Edsall, D.M. Ormsby Dempsey, G.D. Moss, and P.E. Polanski. 1982. Atlas of the spawning and nursery areas of Great Lakes fishes. Volumes I-VIII. U.S. Fish and Wildlife Service, Washington, DC FWS/OBS-82/52. Great Lakes Bathymetry [computer file]. 2000-2003. Boulder, CO: National Oceanic and Atmospheric Administration (NOAA), National Geophysical Data Center (NGDC). Available FTP: http://www.ngdc.noaa.gov/mgg/greatlakes/greatlakes.html [September 9, 2011]. Great Lakes Basin Major River Mouths [computer file]. 2011. Ann Arbor, MI: Great Lakes Environmental Assessment and Mapping (GLEAM) Project, University of Michigan School of Natural Resources and Environment. Great Lakes Coastal Forecasting System [computer files]. 2012. Ann Arbor, MI: National Oceanic and Atmospheric Administration (NOAA), Great Lakes Environmental Research Laboratory (GLERL), Great Lakes Coastal Forecasting System. Available FTP: http://www.glerl.noaa.gov/res/glcfs/ [July 16, 2012]. Great Lakes Surface Environmental Analysis [computer files]. 2011. Ann Arbor, MI: National Oceanic and Atmospheric Administration (NOAA), Great Lakes Environmental Research Laboratory (GLERL), Great Lakes CoastWatch Program. Available FTP: http://coastwatch.glerl.noaa.gov/glsea/glsea.html [July 16, 2012]. Haltuch, M.A., and P.A. Berkman. 1999. Guide to Lake Erie Geographic Information System: Bathymetry, Substrates, and Mussels. Ohio Sea Grant, Columbus. Lake St. Clair Substrate [computer file]. Ann Arbor, MI: Institute for Fisheries Research, University of Michigan, Michigan Department of Natural Resources, 2000. Lane, J.A., C.B. Portt, and C.K. Minns. 1996. Spawning Habitat Characteristics of Great Lakes Fishes. Can. MS Report of Fish. Aquat. Sci. 2368: 48p. Department of Fisheries and Oceans, Canada. Powers, C.F. and A. Robertson. 1968. Subdivisions of the Benthic Environment of the Upper Great Lakes, with Emphasis on Lake Michigan. J. Fish. Res. Bd. Canada, 25(6): 1181-1197. R Development Core Team. 2011. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL http://www.R-project.org/. 30 T-10-T-3, Study 237000 Robbins, J. A. 1986. Sediments of Saginaw Bay, Lake Huron: Elemental composition and accumulation rates. Special Report No. 102, Great Lakes Research Division, University of Michigan, Ann Arbor, MI. 103 pp. Salford Systems, Inc. 2011. Salford Systems Predictive Mining Suite, San Diego, California. Schneider, J.C. Typology and fisheries potential of Michigan lakes. Michigan academician 8.1 1975: 59-84. Michigan Academy of Science, Arts, and Letters. Stanfield, L. W., S. F. Gibson, and J. A. Borwick. 2006. Using a landscape approach to identify the distribution and density patterns of salmonids in Lake Ontario tributaries. American Fisheries Society Symposium 48:601-621. Steen, P. J., T. G. Zorn, P. W. Seelbach, and J. S. Scaeffer. 2008. Classification tree models for predicting distributions of Michigan stream fish from landscape variables. Transactions of the American Fisheries Society 137:976-996. Steinberg, D., and P. Colla. CART – Classification and Regression Trees. San Diego, CA: Salford Systems, 1997. Thomas, R.L., A.L. Kemp, and C.I. Dell. 1978. Sediments of Lake Superior. J. Great Lakes Res., 4(3-4), 264-275. Thomas, R.L., A.L. Kemp, and C.F.M. Lewis. 1973. The surficial sediments of Lake Huron. Can J. Earth Sci., 10, 226-271. Thomas, R.L., A.L. Kemp, and C.F.M. Lewis. 1972. Distribution, composition and characteristics of the surficial sediments of Lake Ontario. J. Sed. Petrol., 42:66-84. Wells, L., and A. L. McLain. 1973. Lake Michigan: man’s effects on native fish stocks and other biota. Great Lakes Fish. Comm. Tech. Rep. 20. 55 p. Zaniewski, A.E., A. Lehmann, and J.McC. Overton. 2002. Predicting species spatial distributions using presence-only data: a case study of native New Zealand ferns. Ecol. Model., 157:261-280. Prepared by: Edward Rutherford, Lacey Mason, and Jason Breck Date: December 18, 2012 31