Evaluating the Effectiveness of Joint Species Distribution Modeling for Fish Communities in the Chesapeake Bay Watershed, USA.

McLaughlin, Paul¹, Kevin Krause², Kelly Maloney³, Taylor Woods⁴, Tyler Wagner⁵, ¹Pennsylvania Coop Fish and Wildlife Research Unit, Pennsylvania State University, University Park, PA, ²Minnesota Department of Natural Resources- Ecological and Water Resources, St. Paul, MN, ³USGS- Eastern Ecological Science Center, Kearneysville, WV, ⁴USGS- Eastern Ecological Science Center, Kearneysville, WV, ⁵USGS, PA Coop. Fish and Wildlife Research Unit, Pennsylvania State University, University Park, PA

Accurately predicting species’ distributions is a critical component of the management and conservation of fish and wildlife populations. While the simplifying assumption is often made that species are distributed independently of one another, recent advances in Joint Species Distribution Models (JSDMs) have made it easier for researchers to include both species-environment relationships and dependence among species within different statistical modeling frameworks. The Chesapeake Bay Watershed (CBW) is one of the largest and heavily studied watersheds within the United States, however few JSDMs have been used to predict fish community distributions watershed-wide. Understanding the drivers and distributions of stream fish communities within the CBW can help inform state fish and wildlife agencies and other water resource agencies that rely on knowledge of stream and river fish communities for assessment programs, many of which have regulatory ramifications and implications for water and fisheries management and aquatic resource use activities. To remedy this, we perform a cross validation study using a JSDM fitted to electrofishing presence absence data from streams and rivers throughout the CBW for sixty different fish species. We highlight species for which the JSDM demonstrated improved conditional prediction (as compared to marginal predictions which ignore species dependencies) and investigate the amount of conditional information required to obtain such improvements. Our findings can help improve species distribution predictions for unsampled areas as well as help make predictions for future species distributions affected by management actions or environmental changes.