update analysis with new predictors

[fawda123]

After #2 is done, update multivariate analysis, some ideas:

• Group by resident times
• SEM based on conceptual model?
• Create cond prob models for sites, time periods, combo of variables when blooms occur
• Validate with other Bight surveys

[fawda123]

Try out SEM stuff starting with conceptual model - do for all sites, then subset by location for nuanced stories. The main drive is to develop a narrative about first principles using all data, then describe the different situations as they vary across latitude/space. Very little work comparing long-term trends across locations. The initial analyses support the SEM - use PCA to guide model dev. Also add conditional prob analysis on the back-end - this will place results in a more refined context.