Developing an Effective Approach for Identifying and Mapping Vernal Pools in Michigan

Transcription

1 Developing an Effective Approach for Identifying and Mapping Vernal Pools in Michigan Yu Man Lee and Helen Enander, Michigan Natural Features Inventory Michael Battaglia, Michigan Tech Research Institute The Stewardship Network Conference January 24, 2015

2 Acknowledgements Funding provided by Michigan Department of Environmental Quality (MDEQ) through U.S. EPA Wetland Program Development Grant, and MNFI. MDEQ Wetlands Program Amy Lounds, Chad Fizzell, Anne Garwood MNFI Vernal Pools Project Team Peter Badra, Suzan Campbell, Joshua Cohen, Helen Enander, Phyllis Higman, Daria Hyde, Brian Klatt, Michael Kost, Jo Latimore, Michael Monfils, Michael Penskar, Rebecca Rogers, Edward Schools; field assistants John Fody, Joann Jeplawy, Rebecca Norris; and administrative support Sue Ridge, Nancy Toben. Michigan Tech Research Institute (MTRI) and Michigan Technological University Laura Bourgeau-Chavez, Kirk Scarborough, Michael Battaglia, Sarah Endres, and Zach Laubach. Herpetological Resources and Management, LLC David Mifsud Partners Michigan Department of Natural Resources, Michigan Nature Association, Michigan Chapter of The Nature Conservancy, Huron River Watershed Council, MSUE Oakland County, Huron Pines, Montmorency County Conservation Club, Montmorency County Conservation District, Headwaters Land Conservancy, Huron-Manistee National Forest, Ottawa National Forest, and numerous volunteers.

3 What are vernal pools? Naturally occurring, temporary pools/wetlands in shallow depressions in forested landscapes

4 Size Key Characteristics < 1 ha / 2.5 ac, most < 0.2 ha / 0.5 ac Hydrology Seasonally flooded, semi-permanent No permanent inlet/outlet or persistent surface connection to permanent water Geomorphology Isolated/confined basin; also can be part of or connected to other wetlands Fish Absent / temporary

5 June 23, 2013 May 2, 2013

6 Key Characteristics Presence of plants and animals that can withstand flooding & drought Fairy Shrimp Blue-spotted Salamander Spotted Salamander Wood Frog

7 Vernal Pool Significance Critical habitat for wildlife Invertebrates and amphibians Other wildlife species Rare species Water flea Fingernail clam Damselfly larvae Copperbelly Water Snake LT, E Blanding s Turtle - SC Wood Duck

8 Vernal Pool Significance Ecosystem services Nutrient & energy source Water storage & infiltration Groundwater / aquifer recharge Flood control May help improve water quality

9 Vernal Pool Status and Threats Limited information and protection in Michigan Hard to identify Loss and degradation Drained/filled for development Excavated - detention ponds Converted to permanent ponds Extensive loss of forests Vulnerable to climate change

10 Vernal Pools in Michigan MDEQ, MDNR, and others are interested in protecting and managing vernal pools in MI. Michigan s Wildlife Action Plan has identified ephemeral wetlands as critical and imperiled habitat. Forest Certification Standards (FSC, SFI) and Michigan s Sustainable Soil & Water Quality Practices protect vernal pools. Protecting vernal pools contributes to forest resilience

11 MNFI Vernal Pool Project Develop an effective & efficient approach for identifying, mapping, assessing and monitoring vernal pools Foundation for a statewide vernal pool mapping and monitoring program in MI

12 Project Components Evaluated different methods for identifying and mapping vernal pools (VPs) Developed initial framework for classifying, assessing and monitoring VPs Developed pilot volunteer VP mapping and monitoring program

13 Project Components Developed statewide database to track and monitor VPs Developed amphibian and reptile best management practices manual HRM Convened Vernal Pools Work Group

14 Detecting and Mapping VPs Evaluated different methods for detecting and mapping VPs Air photo interpretation Radar / Lidar GIS modeling Field sampling

15 Vernal Pool Study Areas UP Ottawa National Forest State Forest Baraga, Crystal Falls FMUs NLP Huron National Forest State Forest Atlanta, Grayling FMUs SLP Waterloo-Pinckney SRA Proud Lake SRA Highland SRA

16 Air Photo Interpretation 1, 344 Potential Vernal Pools (PVPs) Identified and Mapped

17 Sampling Design 1 ha (2.5 ac) test cells, randomly ordered

18 VP Sampling Results Total number of test cells surveyed in the field SLP 125 NLP 168 UP 109 Total 402

19 Results Air Photo # Surveyed test cells with PVPs SLP 40 NLP 40 UP - 30 # Surveyed test cells w/o PVPs SLP 85 NLP 128 UP - 79 # Surveyed test cells - with VPs in the field Accuracy / True Positives SLP 29/40 (73%) NLP 34/40 (85%) UP 18/30 (60%) Omission / False Negatives SLP 10/85 (12%) NLP 4/128 (3%) UP 20/79 (25%) # Surveyed test cells with no VPs / not VPs in the field Commission / False Positives SLP 11/40 (27%) NLP 6/40 (15%) UP 12/30 (40%) Accuracy / True Negatives SLP 75/85 (88%) NLP 124/128 (97%) UP 59/79 (75%) *Lathrop et al NJ 88% accuracy rate, 12% commission, 30% omission

20 Pilot Radar Study - MTRI Develop methods to detect vernal pools without a priori information using satellite radar imagery (PALSAR) Develop methods integrating PALSAR with other data sources Evaluated use of SAR, LiDAR (light detection and ranging using laser; Pinckney SRA only) and 10 m DEM (digital elevation model) to identify vernal pools Compare mapping results against field data

21 Japanese ALOS PALSAR Satellite L-band 24 cm wavelength Penetrates a forest canopy 10 m resolution FPS product Two season ALOS PALSAR data used spring, summer (spring summer) Assume that spring image will exhibit bright spots where vernal pools are located due to characteristic RADAR double bounce while that effect will be diminished in summer.

22 PALSAR May and August

23 LiDAR Intensity Spring 2009 Intensity is a measure of return strength of the laser pulse that generated the point 23

24 Random Forest LiDAR and DEM Producers Accuracy=36% Users Accuracy=26.9% Input Layers DEM Difference TPI LiDAR Intensity

25 Random Forest PALSAR and 10 m DEM Producers Accuracy=76.4% Users Accuracy=72.7% Input Layers PALSAR PALSAR Change DEM Difference TPI

26 Pinckney Vernal Pools PALSAR- DEM

27 Summary & Conclusions PALSAR spring and summer imagery provides a method for finding potential vernal pools across a 70 x 70 km region for further investigation with Air Photos or field data sampling Advantage freely available, independent of cloud cover or canopy cover LiDAR alone provides indications of where vernal pools are with leaf-off imagery Advantage high resolution provides boundaries of vernal ponds Disadvantage high omission and commission errors in Random Forests, high cost and not widely available PALSAR with 10 m DEM (or LiDAR) Provides improved mapping capability when coupled with field verified vernal pool sample data 10 m DEM available statewide most cost-effective Need to explore different classification methods Random Forests, MCDA, etc.

28 Air Photo vs. Radar - SLP # Surveyed test cells with PVPs Air Photo 40 Radar 65 # Surveyed test cells w/o PVPs Air Photo 60 Radar 35 # Surveyed test cells - with VPs in the field Accuracy / True Positives Air Photo 29/40 (73%) Radar 32/65 (49%) Omission / False Negatives Air Photo 9/60 (15%) Radar 5/35 (14%) # Surveyed test cells with no VPs / not VPs in the field Commission / False Positives Air Photo 11/40 (27%) Radar 33/65 (51%) Accuracy / True Negatives Air Photo 51/60 (85%) Radar 30/35 (86%)

29 GIS Distribution Modeling of VPs Statistical modeling to predict suitability of vernal pool occurrence. MaxEnt maximum entropy approach - compares known locations to random background points of a set of environmental conditions to predict or identify areas of relative habitat suitability for VPs. Identify environmental drivers for VP distribution Used 75% field data to train or develop the model and 25% to test the model

30 Sturgeon Incised Terrain Sample Modeling Sites Extents NLP Tunneled Uplands Southeastern Interlobate Core

31 Environmental Variables Name Scale/resolution Source Quarternary geology 1:500,000 Farrand and Bell (1982a, 1982b) Glacial landsystems 1:250,000 RS&GIS, Michigan State University (Lusch et al. 2005). Geology Estimated groundwater recharge PLSS section RS&GIS, Michigan State University (Lusch et al. 2005). Glacial Deposits - Estimated Transmissivity 1000 m raster RS&GIS, Michigan State University (Lusch et al. 2005). Drift thickness 500 m raster RS&GIS, Michigan State University (Lusch et al. 2005). Bedrock surface elevation 500 m raster RS&GIS, Michigan State University (Lusch et al. 2005). Elevation 10 m raster USGS (2009a) Elevation 30 m raster USGS (2009b) Groundwater Topography Local elevation difference a 30 m raster Derived from 30m Digital Elevation Slope (degrees) 30 m raster Derived from 30m Digital Elevation Topographic position index (continuous) 30 m raster Derived from 30m Digital Elevation and Weiss (2001), Jenness (2013) Topographic position (6 classes) 30 m raster Derived from 30m Soils Digital Elevation and Weiss (2001), Jenness (2013) Bolstad concavity-convexity index 30 m raster Derived from 30m Digital Elevation and Bolstad et al. (1998) Compound topographic index (CTI) 30 m raster Derived from 30m Digital Elevation and Moore et al. (1993) Surface curvature 30 m raster Derived from 30m Digital Elevation and ArcMap Spatial Analyst Curvature tool Soil series parent material 1:15,840 SSURGO Soils data, Natural Soil Drainage Index b 1:15,840 SSURGO Soils data, and Schaetzl et al. (2009) Clay in all horizons 1:15,840 SSURGO Soils data, Clay in uppermost mineral horizon 1:15,840 SSURGO Soils data, Texture of uppermost mineral horizon 1:15,840 SSURGO Soils data, Texture of lowermost mineral horizon 1:15,840 SSURGO Soils data, Parent material graveliness 1:15,840 SSURGO Soils data, Anderson Level I and II cover types (2006) 30 m raster NOAA (2008) Local positive landcover c 30 m raster Derived from CCAP 2006 land cover Landcover 28

32 Variable % + Landcover (300 km 2 ) 39.7 Ground water recharge 32.6 Elevation (30m) 7.7 Drift thickness 7.0 Surficial geology class 5.9 Depth to 1 st water table 3.6 TPI 2.2 Clay in all horizons 1.3 Xx % habitat SLP Model 7.7% of study area suitable

33 Model Evaluation AUC (area under the curve) probability that a random presence location is ranked higher than a random background point AICc (Akaike information criterion) with correction for finite sample size Omission % on test and training samples Comission % on PVP field verified as not VP

34 Glacial outwash sand and gravel and postglacial alluvium Variable Response Curves Ice-contact outwash sand and gravel Ground water recharge Geology Positive land cover Elevation

35 GIS Model Results by Site Site Variables Parameters AUC Train AUC Test AICc SLP NLP UP Site FPA Omission Test Omission Train Commission SLP 7.73% 0% 3.4% 78.7% NLP 4.30% 0% 4.00% 52.9% UP 14.8% 0% 9.50% 55.6%

36 Proposed Approach for Mapping VPs Start with GIS model to identify areas statewide or regionally with greater likelihood for vernal pools Esp. in the SLP and NLP Start with or follow up with radar (where spring & summer imagery available) Follow up with targeted aerial photo interpretation Verify vernal pools in the field

37 Next Steps Continuing to map vernal pools Verso Paper Corp, Michigan Forest Products Council, MDNR, and MNFI W. UP Hiawatha National Forest E. UP Ongoing - Volunteer vernal pool mapping & monitoring Develop and implement statewide vernal pool mapping and monitoring program Identify and collaborate with additional partners to identify and map vernal pools MI VP Partnership Additional research on VP mapping methods UP needs more work/analysis Radar and GIS modelling more work

38 Questions?