USGS Home
Contact USGS
Search USGS

Influences of availability on parameter estimates from site occupancy models, with application to submersed aquatic vegetation

Gray B.R., M.D. Holland, F. Yi, and L.A.H. Starcevich. 2013. Influences of availability on parameter estimates from site occupancy models, with application to submersed aquatic vegetation. Natural Resource Modeling 26: 526-545

Abstract

River managers are often interested in status and trends of submersed aquatic vegetation (SAV); they may also be interested in the effects of different natural forces and human actions on SAV.  Within the Upper Mississippi River System, rake data derived using the UMRR-EMP/LTRMP monitoring protocol are often used to estimate percent frequency of occurrence (“occurrence”).  As with summaries of data from many monitoring protocols, our estimates of SAV occurrence may often be low.  The source of this problem is that we calculate occurrence as an average of the number of sites at which SAV is detected—and we don’t typically detect SAV at all sites at which it is present.  This study evaluated biases associated with percent frequency of occurrence estimates from LTRMP and EMAP data.  The study demonstrated that biases in typical occurrence estimates from rake data arise not only from false negatives (SAV is present but not detected) but also from incomplete coverage of the site by SAV.  When SAV at a site is sparse, we are less likely to detect it and, hence, we are more likely to erroneously conclude that SAV is absent.  This problem will decrease in importance as coverage within sampling sites increases.  Many factors other than coverage may affect our ability to detect SAV at sampling sites, including species attributes, biomass, substrate and depth.  Thus, we should routinely treat our SAV occurrence estimates as indices, and should be cautious about comparing SAV occurrence estimates across LTRMP strata and across species.  For the same reason, we should be cautious when fitting models to our site detection data—as those models may make inferences across sites with very different detection probabilities and coverages.  The sources of bias addressed in this study may also be important when quantifying differences in SAV levels before and after HREP implementation (however, estimated differences should typically be qualitatively correct).  Preliminary results indicate that the sources of bias described in this study will only rarely affect estimates of trends in SAV occurrence.  This research was supported by the US Army Corps of Engineers’ Upper Mississippi River Restoration program and EPA’s Environmental Monitoring and Assessment Program.

Keywords