The Power of Sampling Method Designs: Detecting Changes in the Abundance of Invasive Species. Sarah E. Johnson Erika L. Mudrak Donald M.

Transcription

1 The Power of Sampling Method Designs: Detecting Changes in the Abundance of Invasive Species Sarah E. Johnson Erika L. Mudrak Donald M. Waller

2 Are Our Monitoring Programs Adequate? Why conservation monitoring usually is, but should not be, a waste of time. (Legg and Nagy 2006) We must evaluate the statistical power and utility of monitoring programs at the outset. (Peterman 1990, Legg and Nagy 2006) Photo: Merel Black

3 Great Lakes Inventory & Monitoring Network

4 Long-term Vegetation Monitoring Changes in Community Composition - Richness - Relative Abundance - Similarity Changes in Community Structure - Distribution of Tree/Sapling Sizes - Shrub Density/Cover -Herb Cover Species-Specific Changes - Abundance (relative frequency) Photo: Merel Black -Size

5 Research Objective To evaluate the efficiency, sensitivity and precision of 3 vegetation sampling methods being considered for use in the GLKN long-term terrestrial vegetation monitoring protocol

6 Subplot 2 Method Subplot 4 Subplot 1 Subplot 3 Extensive legacy data Sites and plots are dispersed evenly on a grid Designed in the 1930 s to estimate timber volume Modified to provide additional data on shrubs, herbs, and CWD Subplot Microplot Herb Quadrats Coarse Course Woody Debris Transects Plot Center (PC) 7.32 m (24.0 ft) radius 2.07 m (6.8 ft) radius 1.0 m 2 area 7.32 m

7 PEL Method ~13 m 30 m Originated with the Plant Ecology Lab of John Curtis and his students in the 1940 s & 50 s for The Vegetation of Wisconsin work. Tree/Herb Transects ~135 m Herb Quadrats 1 m 2 Shrub Quadrats 4 m 2 Extensive legacy data from sites.

8 Transect 1 #1 #2 HYBRID Method Transect 2 #3 #4 Goals: Simple to set up Efficient to sample 50m Adequate to characterize the site Transect 3 #5 #6 50m Balanced in terms of relative effort and quality of data obtained for herbs, shrubs, saplings, and trees Tree Sample Area 50 x 6 m Shrub Circle 2.82 m radius Herb Quadrat 1 m 2 Coarse Course Woody Debris Transect 50 m

9 Research Locations APIS PIRO vs. PEL vs. Hybrid n = 20 sites n = 20 sites

10 Statistical Issue: Ability to Detect True Change (Power) Monitoring Program Detects Change Monitoring Program Detects No Change No Change False-Change Error (Type I) α No Error (1-α) Change No Error (Power) 1-β Missed-Change Error (Type II) β From Elzinga et al. (1998) - Measuring and Monitoring Plant Populations GLKN Network Objective: Detect 20% change α = 0.1 Power 80%

11 Modeling Statistical Power to Detect Change Summer 2005 Vegetation Metrics 20% change Hypothetical Year B Vegetation Metrics The modeled change was normally distributed around 20% change with a given standard deviation Standard deviations based on values estimated from actual 5- year changes at PIRO and in northern WI Each simulation was run 1000 times Wilcoxon tests used to assess significant differences between Summer 2005 and Year B data Estimated power based on frequency of P-values < 0.1

12 Power to Detect 20% changes in Trees at PIRO Hybrid Power (%) Tom Givnish 20 R. Freckman 0 Tree Density Tree BA/ha Fagus grandifolia Density Betula papyrifera Density Tree Subplot Variables Tsuga canadensis Density M. Clayton and Hybrid had equivalent power for tree density and basal area Both methods had low power for detecting change in individual species density

13 Power to Detect 20% Changes in Groundlayer Plants Sampled in 1-m 2 Quadrats at PIRO Hybrid Power (%) Forb Frequency Forb Richness Fern Frequency Woody Plant Frequency Herb Quadrat Variables Maianthemum canadense Frequency The Hybrid method attained higher power for most metrics 80% power was never attained by the method for these variables Both methods had low power for detecting a 20% change in M. canadensis

14 Power to Detect 20% Changes in Groundlayer Plants Sampled in Walk-throughs at PIRO Hybrid 80 Power (%) Groundlayer Richness Forb Richness Walk-through Variable Both methods had > 80% power to detect a 20% change in plant richness The method did not attain 80% power to detect a change in forb richness

15 Number of Sites Required by and Hybrid to Acquire 50% Power to Detect a 20% Change Trait Tree BA (cm 2 /ha) and Tree Density Individual Tree Species Forb Frequency Forb Richness Fern Frequency Maianthemum canadense Frequency Understory Vegetation Richness Forb Richness Level of Analysis Subplot Subplot Herb Quadrats Herb Quadrats Herb Quadrats Herb Quadrats Walkthroughs Walkthroughs # of Sites Needed for 50% Power to Detect 20% Change < 5 LP LP LP LP LP 5 10 HYBRID < 5 LP LP < 5 6

16 Conclusions The method is efficient at sampling trees No method attained adequate power to detect a 20% change in individual tree and herb species The method consistently had low power for detecting changes in understory variables; sampling this strata more extensively is justified Composite indicators are more powerful than species-specific indicators

17 Recommendations to GLKN The GLKN should consider increasing the number of sites sampled at each park If the method is selected, then it should be augmented to sample the understory more extensively Using the Hybrid method could be more cost effective Data should be aggregated by habitat type to reduce amongsite variances and to increase power to detect change Reduce sampling error by hiring skilled botanists!

18 Power to detect changes in Invasive Species? Detecting changes in target species could be tricky, and the power of the sample design should be determined Must establish clear objectives for the monitoring program Must determine the amount of change that is suitable/unsuitable to meet management goals.

19 Acknowledgements GLKN Personnel: Suzanne Sanders, Erik Beever, Bill Route APIS and PIRO Personnel Summer 2005 Field Crew: Dan Olson, Darian Kath, and Joshua Sulman Data Entry: Raffica La Rosa and Aliza Segal UW-Madison Botany Dept. Personnel: Bret Larget, Dave Rogers, Tom Rooney APIS Legacy Data Provided by E. Beals, G. Cottam and other colleagues associated with J.T. Curtis Plant Ecology Laboratory (PEL) Funding: NPS I&M and NRRI Duluth, MN

20 and Hybrid Sampled a Comparable Number of Species per Unit Area up to a Point, but Hybrid Detected More Groundlayer Species Overall Northern Hardwood Stands (n = 9) Pine Stands (n = 8) Number of Species HYBRID Number of Quadrats Number of Quadrats HYBRID Curves created with EstimateS version 7.5 (Colwell

21 and Hybrid Estimates of Tree BA w/ Increasing Number of Individuals was Similar at PIRO 19.4% SE 15.4% SE 13.6% SE 8.6% SE 16.8% SE 12.5% SE 16.8% SE 12.5% SE

22 Consistently had Lower Estimates of Tree BA than PEL a.) b.) Estimated Estimated Mean Mean (±SE) (+/-SE) Tree Tree BA (cm 2 /ha) Upland Mixed Hardwood (n = 7) Conifer-Hardwood (n = 3) 12.3% SE PEL c.) Estimated Tree BA Estimated Estimated Mean (±SE) Tree BA Tree (+/- SE) BA (cm 2 /ha) % 8.00% SE Estimated Tree BA (+/- SE) % SE Hardwood-Conifer (n = 10) 12.1% SE 6.00% SE 600 HYBRID PEL Number of Number Individuals of Individuals Sampled % SE PEL