270 likes | 546 Views
Use of survival data for planted woody stems to refine a vegetation monitoring protocol for restoration sites. Thomas R. Wentworth, Michael T. Lee, Mac Haupt, M. Forbes Boyle, Robert K. Peet 17 November 2010. CVS-EEP Sampling Protocol. Optimized for field efficiency and repeatability.
E N D
Use of survival data for planted woody stems to refine a vegetation monitoring protocol for restoration sites Thomas R. Wentworth, Michael T. Lee, Mac Haupt, M. Forbes Boyle, Robert K. Peet 17 November 2010
CVS-EEP Sampling Protocol • Optimized for field efficiency and repeatability. • Resources include manuals, datasheets, and a data entry and reporting tool. • Scalable to meet future requirements. • Complies with US-FGDC National Vegetation Classification Standard.
Utility of the Collected Data? Stakeholder feedback: What is gained from measurements collected using the CVS-EEP Protocol? • Variables measured are mandated by EEP, not CVS. • EEP initially required multiple types of measurements because it was unclear which ones would be most useful in assessing stem success. • Available data from EEP Monitoring Firms will now allow CVS to assess the utility of each field measurement (e.g., ddh, height, DBH). • Which plant attributes should continue to be measured in the field? • Particular concerns were raised about ddh measurement.
Current Status of CVS-EEP Inventory • Monitoring conducted for 5 years (2006-2010)
Current Status of CVS-EEP Inventory • 785 unique plots monitored 2006-2010 • Range is 3-28 plots/project/year (median = 8)
Overview of Woody Stem Database • As of October 2010, we have 30,544 individual records for planted woody stems. • 166 taxa, 127 species (18 oaks, 6 maples, etc.) • Median is 141 stems/project-year: • height data: 121 stems/project-year • ddh data: 98 stems/project-year • DBH data: 38 stems/project-year • three largest tallies for a project in a given year are 800, 617, and 460 planted stems.
Modeling Rationale • Goal: take a stem and characterize its likelihood of surviving to the next year, • then compare model prediction with reality • among predictive variables available, which are essential and which are extraneous (particular focus on utility of ddh)? • independent variables allow model evaluation with and without ddh-related variables • benefits of such a modeling effort: • evaluating restoration plans, planting lists (including species, source, size, etc.) • being better able to identify projects on good or bad trajectories
Modeling Approach • General approach was logistic regression using GLM, using survival to next year as dependent variable (1=survived, 0=died). • Independent variables incorporated into models: • ddh (ln transformed), RGR of ddh • height (ln transformed), RGR of height • year since planting (1-6) • vigor (1-4) • 1 = not expected to survive • 4 = excellent • source, for example: • ball and burlap (B) • potted (P) • bare root (R) • tubling (T)
Subsetting Database • only planted woody stems • only those with ddh and height • minimum 3 years data (two years for RGR, third year to determine survival from year two) • no pseudoreplication (random selection of one three-year sequence) • withhold 25% of observations for validation (also random, for further work) • 2120 stems, of which 429 (20.2%) died in year three
Discussion • Models tested thus far are in the “fair” range, based on AUC criterion (0.69-0.79). • Height-only (AUC=0.69) and ddh-only (AUC=0.71) models perform similarly. • Combining height and ddh does not much improve model performance (AUC=0.71). • Complex (“everything”) model shows enhanced performance (AUC=0.79) over simple models. • Removing ddh from complex model results in little change in model performance (AUC=0.78). • Categorical-variables-only model performs reasonably well (AUC=0.76).
Conclusions • Given our perspective (predicting stem survival to next year), height and ddh are comparable in utility. • Little benefit to including both variables. • Omitting ddh from complex model has relatively little impact. • In these models, it appears that ddh contributes little to prediction of stem survival, as long as we retain height measurement. However...
Other Considerations • We should exercise caution in discarding variable for which we have such a solid existing data base. • ddh may yet prove to have benefits: • diameter (combined with height) allows for volume computation (d2h) • are there particular subsets of stems where ddh is a critical predictor of success (further work)? • What is the cost in our cost:benefit analysis for this particular variable?
Thank You! http://cvs.bio.unc.edu/