Multistate modeling of habitat dynamics

Forum for discussion of general questions related to study design and/or analysis of existing data - software neutral.

Multistate modeling of habitat dynamics

Postby Orien Richmond » Thu Jun 16, 2016 12:18 pm

I'm analyzing a data set consisting of 156 vegetation transects in the Nebraska Sandhills that were monitored from 2003-2015 (2006 and 2010 were missing years). The number of transects monitored in each year when there was monitoring ranged from 30-156 (mean = 98). The transects selected for monitoring in each year were not random (e.g., in year one about half of the transects were monitored; in year two the other half were monitored; subsequent years look to be based on convenience sampling). Each transect in each year was assigned to one of 3 states based on vegetation structure (high, med and low VOR). I have data on management actions for each transect over time (burning, grazing) and rainfall data. I'd like to ask questions like: Are transitions to lower (or higher) VOR states influenced more by precipitation or by management actions?

I'm trying to follow the methods used by Breininger et al. 2010 "Multistate modeling of habitat dynamics: factors affecting Florida scrub transition probabilities" in program MARK. I fixed survival = 1 because a transect can't die. Breininger et al. set recapture prob = 1 because they sampled all locations in every year when locations were sampled. I'm having trouble with this because not all transects in my data set were sampled in every year. My data set also violates the assumption that each transect has an equal probability of capture because the sampling was biased in some years. If a transect was selected for sampling in a given year, its recapture probability is = 1 (once selected for sampling, it is impossible to not find a transect). So it seems recap prob in this instance could/should be reinterpreted as the probability of selecting a site for sampling.

Here are some options I have thought about:
(1) Estimate time-specific recapture probabilities (results in many extra parameters that I'm not interested in); huge number of parameters when fitting models with time- and state-specific recapture probabilities
(2) Fix time-specific recapture probabilities based on the known proportion of transects surveyed in each year--this gets tricky when fitting models with time- and state-specific recapture probabilities

For the transition probabilities, there were very few transitions in the data from Low->High VOR and from High->Low VOR, so I fixed these parameters at a small rate (0.03). This seemed to improve model behavior--does this seem reasonable?

Any advice or insight on the recapture prob issue greatly appreciated. I'm relatively new to MARK and hope to transition to RMark once I have covered the basics.

Thank you
-Orien
Orien Richmond
 
Posts: 6
Joined: Wed Mar 26, 2008 7:56 pm
Location: Denver, CO

Return to analysis & design questions

Who is online

Users browsing this forum: Google [Bot] and 12 guests

cron