I am working on CJS models for 2 separate populations. A "South" population was surveyed from 2012-2017, and a "North" population from 2014-2017. However, the "South" population crashed during our study, with only 1 new individual captured in 2016 and again in 2017. Therefore, it seems I should only include years 2012-2015 for this population in MARK.

Given that I have 4 survey periods (years of data) for each of my populations, I have 2 questions:

1) Is there a general rule of thumb about how many survey periods one needs before they can start testing for certain effects on parameters across survey periods using covariates? For example, I want to see if winter temperatures impacted annual survival rates in my populations; therefore, I would include 3 mean temperature values within a column of my design matrix, one for each survival estimate. Is this enough information to adequately test my hypothesis through model ranking?

2) Although I hypothesize that survival could vary based on yearly environmental conditions, I do not have reason to believe that survival would inherently vary over time. Therefore, I do not plan to include any Phi(t) models in my candidate set. However, I know that covariates cannot be included in global models for estimating overdispersion with the median c-hat test in MARK, so if I do move forward with having covariates in some of my candidate models, does it make sense to have fully time-dependent survival in my global model? In order words, is it logical to say that "if my dataset is robust enough that I can estimate different survival rates for each year, then it should be robust enough to time for environmental effects on annual survival rates using covariates" ?

Thank you for your help!