Primary Period assumptions in unmarked dataframes

[Joshua Armstrong]

Hello All,

I am a PhD student who, for the past couple months, has been using the "unmarked" package with colext function to model occupancy dynamics of an invasive aquatic plant. The system I work in is highly dynamics and communities are regularly disturbed causing colonization and extinction events randomly throughout a season or year. Due to the characteristics of the system, we are not interested in detection. I have ~10 surveys/year across 4 year. The data was collected at uneven intervals leading to uneven sampling periods. My first questions is has anyone used this framework for occupancy dynamics with uneven primary periods?

Due to one of the model assumptions being that the occupancy state does not change within a primary period (i.e. the system is closed), I have treated each survey as its own primary period (leading to 40 columns of presence/absence data), all associated with Julian days and year. I did this because within a season/year (typical primary periods), the occupancy state can change due to disturbance. Below is the summary of the unmarked dataframe, with continuous variables standardized:

unmarkedFrame Object

373 sites

Maximum number of observations per site: 40

Mean number of observations per site: 4.79

Number of primary survey periods: 40

Number of secondary survey periods: 1

Sites with a least one detection: 93

Tabulation of y observations:

0              1             <NA>

1602        183        13135

Site-level covariates:

Cont_var1                      Cont_var2                             Cont_var3

Min.        : -1.1261         Min.        :  -0.58705            Min.         :  -1.9761

1st Qu.   : -0.6162         1st Qu.   :  -0.51851            1st Qu.    :  -0.7871

Median  : -0.2738         Median   :  -0.36529            Median   :  -0.1772

Mean     :  0.0000         Mean      :   0.0000               Mean      :   0.0000

3rd Qu.  :  0.2824         3rd Qu.   :   0.02568             3rd Qu.   :   0.5356

Max.      :  8.1446         Max.       :  10.12976            Max.       :   3.2778

Cont_var4                       Cont_var5

Min.        : -1.2842         Min.        :  -1.1428            

1st Qu.   : -0.7880         1st Qu.   :  -0.7517         

Median  : -0.3377         Median   :  -0.2949          

Mean     :  0.0000         Mean      :   0.0000               

3rd Qu.  :  0.7696         3rd Qu.   :   0.5091            

Max.      :  2.5453         Max.       :   3.9603            

Yearly-site-level

year                      date

2017 : 3730        Min.         :  -0.3548

2018 : 3730        1st Qu.    :  -0.3548

2019 : 3730        Median   :  -0.3548

2020 : 3730        Mean      :   0.0000

                             3rd Qu.   :  -0.3548

                             Max.       :   4.0376

Everything works fine when running the code and modeling. My advisors and I just wanted to make sure our handling of primary periods makes sense?  Thank you for any help.

Thanks,

Josh Armstrong

[gcsadoti]

Hi Josh,

The multi-season model was developed explicitly to address imperfect detection (building off the single-season model). I suspect that with only one survey per period, your detection estimates look funky (large SEs, etc.). There are times when one must use a post-hoc approach to identifying periods of closure (which may vary in length), but there's an overarching assumption of this model that at least some primary periods will have more than one survey in time (or sometimes space; e.g. the Nichols et al. multi-scale/multi-method model).

That said, if ...

1. you don't have *any* primary periods in which the closure assumption holds and

2. you are certain that detection of your spp is (or is essentially) perfect

... then I think you are better off using an approach such as a generalized linear (and maybe mixed) model.

For example, here's one paper: https://www.sciencedirect.com/science/article/abs/pii/S0006320710000054 (this paper also addresses thresholds, it's that's a separate topic). In this glm approach, transitions between surveys can be modeled such as when spp in a site remained 1) absent vs. colonized and 2) persistent vs. extirpated.

Cheers

Giancarlo

[Joshua Armstrong]

Hello Giancarlo,

Thank you for the quick reply. I will look into some of your suggestions.

Thanks,

Josh