Adult Model Covariates

There are multiple covariates used in SR JPE modeling that were developed through separate but related processes. There are currently two documents describing covariate selction and processing for different models: (1) Stock Recruit Covariates and (2) Adult Model Covariates (this document).

The Passage to Spawner (P2S) model relates spawner counts (from redd or holding surveys) to upstream passage counts obtained by video systems. This model is restricted to streams with reliable redd or holding surveys and reliable upstream passage counts. It is also restricted to years where the redd/holding and upstream passage datasets overlap. This vignette describes the process of pulling and processing environmental covariates for use in the P2S model.

Selecting Adult Environmental Covariates

Potential environmental covariates hypothesized to influence prespawn mortality were proposed in meetings with the SR JPE Modeling Advisory Team (MAT). Five initial categories were identified:

  • temperature
  • flow
  • water year type
  • passage timing
  • total passage

There are many ways to summarize each of these categories and initial analyses helped identify collinearity and performance of each potential method by regressing prespawn mortality (calculated as upstream_count / spawner_count) against the environmental variable. When we were using redd counts as spawner_count, our model assumed a 50/50 sex ratio and modified that equation to be upstream_count / (spawner_count * 0.5). Generally, one redd per female is a reasonable assumption although our model left the possibility open for more than one redd per female (source). Note that this covariate analysis and preparation does not include the Sacramento River mainstem as spring run do not spawn on the mainstem Sacramento.

Preparing Covariates

Temperature

Several approaches were considered for summarizing temperature:

  1. Proportion of days where the temperature surpassed a threshold of 20 degrees Celsius (source)
  2. Growing degree days (GDD) with a base temperature of 0 degrees Celsius (source)
  3. Degree Day 20 (DD20), where cumulative degree days are calculated against a threshold of 20 degrees Celsius (source)

Following initial analyses and discussions, we focused on approach 3 because it showed the most consistent relationship with prespawn mortality across streams and accounts for cumulative stress. We calculated the metric for migration months (March - May) in the Sacramento River and holding months (May - August) in each tributary. The resulting dataset is as follows (note that DD less than 0 are set to 0):

year stream gdd_trib gdd_sac gdd_total
1999 deer creek 79.30787 0 79.31
1999 mill creek 24.35044 0 24.35
2000 butte creek 38.89352 0 38.89
2000 deer creek 191.70059 0 191.70
2000 feather river 47.67365 0 47.67
2000 mill creek 97.26157 0 97.26
2000 yuba river 430.34858 0 430.35
2001 butte creek 70.30324 0 70.30
2001 clear creek 17.07813 0 17.08
2001 deer creek 282.67653 0 282.68

The following plot is of the growing degree days above the 20 degree threshold over time for all tributaries:

Flow

Maximum flow more effectively captures the high flow events that support migration speed and passage to upstream holding areas. Additionally, upon inspection of the data source across multiple years average maximum flow over the migratory and holding months (March-May and May-August, respectively) was more representative of the fluctuations in flow over the entire year. The resulting dataset is as follows:

stream year mean_flow max_flow
battle creek 1962 381.2174 1230
battle creek 1963 546.4620 3030
battle creek 1964 299.2228 442
battle creek 1965 531.6793 1590
battle creek 1966 354.5272 664
battle creek 1967 638.0163 1530
battle creek 1968 379.7011 1030
battle creek 1969 666.6467 1270
battle creek 1970 541.0217 1510
battle creek 1971 585.0380 2500

The following plot is of max flow (cfs) over time for all tributaries:

Water Year Type

To account for the influence of large scale interannual climate variability, we investigated water year type as a covariate as described by the California Department of Water Resources and available in the waterYearType package in R.

We used the waterYearType package to pull water year assignments as a categorical covariate. Some streams had very few data points (i.e. for Mill Creek, only seven years were considered dependable), so we simplified all potential categories of water year type into either dry (Dry, Below Normal, Critical) or wet (Wet, Above Normal). The table below summarizes the number of dry and wet years that were included:

Water Year Type Count
dry 60
wet 52

Passage Timing

Passage timing was considered; however, limited data reduced the sample size of the datasets for some tributaries so much as to remove them from candidacy for the model due to lack of statistical power. Passage timing was summarized as the mean, median, and min day of passage. The resulting dataset is as follows:

year stream median_passage_timing mean_passage_timing min_passage_timing
1998 battle creek 23.5 24.03333 22
1999 battle creek 25.0 24.58824 21
2000 battle creek 28.0 28.48684 21
2001 battle creek 21.5 23.33784 19
2002 battle creek 27.0 27.90909 21
2003 battle creek 33.0 32.00000 25
2004 battle creek 24.0 24.34783 22
2005 battle creek 24.0 24.81395 21
2006 battle creek 25.5 25.77500 24
2007 battle creek 22.0 22.95187 19

The following plot is of median passage over time for all tributaries:

Total Passage as Index

We hypothesized that total annual passage might be an indicator of density because more adults in holding/spawning habitat could result in less available habitat and thus influence prespawn mortality.

year stream passage_index
1995 battle creek 66
1995 clear creek 2
1996 battle creek 35
1997 battle creek 107
1998 battle creek 178
1998 clear creek 47
1999 battle creek 73
1999 clear creek 35
2000 battle creek 78
2000 clear creek 9

Combine and Save Covariate Data

Both continuous environmental variables (flow and temperature) were standardized and centered within streams before performing any analyses so that the scale of the data did not affect results. Water year type was coded as a binary variable as 1 for wet (wet, above normal) and 0 for dry (below normal, dry, critical). The resulting dataset is as follows (note that flow data has the longest time series available):

year stream wy_type max_flow_std gdd_std passage_index median_passage_timing_std
1962 battle creek 0 -0.5103739 NA NA NA
1963 battle creek 1 0.6857655 NA NA NA
1964 battle creek 0 -1.0340171 NA NA NA
1965 battle creek 1 -0.2711460 NA NA NA
1966 battle creek 0 -0.8864933 NA NA NA
1967 battle creek 1 -0.3110173 NA NA NA
1968 battle creek 0 -0.6432783 NA NA NA
1969 battle creek 1 -0.4837930 NA NA NA
1970 battle creek 1 -0.3243078 NA NA NA
1971 battle creek 1 0.3335689 NA NA NA

The following plot is of standardized covariates (covariates are colored by type) over time for all tributaries:

Save data object

Data object saved in SRJPEdata as p2s_model_covariates_standard.rds. To access documentation search ?SRJPEdata::p2s_model_covariates_standard.rds.