Bayesian inference and assessment for rare-event bycatch in marine fisheries: a drift gillnet fishery case study

Comparing different models to forecast the number of mass shootings in the United States: An application of forecasting rare event time series data

The number of mass shootings in the United States has increased in the recent decades. Understanding the future risk of the mass shootings is critical for designing strategies to mitigate the risk of mass shootings, and part of understanding the future risk is to forecast the frequency or number of mass shootings in the future. Despite the increasing trend in mass shootings, they thankfully remain rare events with fewer than 10 mass shootings occurring in a single year. Limited historical data with substantial annual variability poses challenges to accurately forecasting rare events such as the number of mass shootings in the United States. Different forecasting models can be deployed to tackle this challenge. This article compares three forecasting models, a change-point model, a time series model, and a hybrid of a time series model with an artificial neural network model. Each model is applied to forecast the frequency of mass shootings. Comparing among results from these models reveals advantages and disadvantages of each model when forecasting rare events such as mass shootings. The hybrid ARIMA-ANN model can be tuned to follow variation in the data, but the pattern of the variation may not continue into the future. The mean of the change-point model and the ARIMA model exhibit much more less annual variation and are not influenced as much by the inclusion of a single data point. The insights generated from the comparison are beneficial for selecting the best model and accurately estimating the risk of mass shootings in the United States.

Disentanglement network data to characterize leatherback sea turtle Dermochelys coriacea bycatch in fixed-gear fisheries

To characterize sea turtle bycatch in fixed-gear fisheries in Massachusetts, USA, we analyzed a 15 yr dataset of entanglement reports and detailed documentation from disentanglement operations. Almost all (272) of the 280 confirmed entanglements involved leatherback turtles Dermochelys coriacea. The majority of turtles were entangled in actively fished (96%), commercial (94%) pot/trap gear with unbroken/untriggered weak links, specifically the buoy lines marking lobster, whelk, and fish traps. Most reports came from recreational boaters (62%) and other sources (26%), rather than commercial fishers (12%). Leatherback entanglements occurred from May to November, with peak reporting in August, and included adult males, adult females, and subadults. All entanglements involved the turtle’s neck and/or front flippers, with varying degrees of visible injuries; 47 entangled leatherbacks were dead in gear, 224 were alive at first sighting, and 1 case was unknown. Post-release monitoring suggested turtles can survive for days to years after disentanglement, but data were limited. While the observed entanglements in our study are low relative to global bycatch, these numbers should be considered a minimum. Our findings are comparable to observed numbers of leatherbacks taken in Canadian fixed-gear fisheries, and represent just one of multiple, cumulative threats in the North Atlantic. Managers should focus on strategies to reduce the co-occurrence of sea turtles and fixed-fishing gear, including reductions in the number of buoy lines allowed (e.g. replace single sets with trawls), seasonal and area closures targeted to reduce sea turtle-gear interaction, and encourage the development of emerging technologies such as ‘ropeless’ fishing.

Ensemble Random Forests as a tool for modeling rare occurrences

Relative to target species, priority conservation species occur rarely in fishery interactions, resulting in imbalanced, overdispersed data. We present Ensemble Random Forests (ERFs) as an intuitive extension of the Random Forest algorithm to handle rare event bias. Each Random Forest receives individual stratified randomly sampled training/test sets, then down-samples the majority class for each decision tree. Results are averaged across Random Forests to generate an ensemble prediction. Through simulation, we show that ERFs outperform Random Forest with and without down-sampling, as well as with the synthetic minority over-sampling technique, for highly class imbalanced to balanced datasets. Spatial covariance greatly impacts ERFs’ perceived performance, as shown through simulation and case studies. In case studies from the Hawaii deep-set longline fishery, giant manta ray Mobula birostris syn. Manta birostris and scalloped hammerhead Sphyrna lewini presence had high spatial covariance and high model test performance, while false killer whale Pseudorca crassidens had low spatial covariance and low model test performance. Overall, we find ERFs have 4 advantages: (1) reduced successive partitioning effects; (2) prediction uncertainty propagation; (3) better accounting for interacting covariates through balancing; and (4) minimization of false positives, as the majority of Random Forests within the ensemble vote correctly. As ERFs can readily mitigate rare event bias without requiring large presence sample sizes or imparting considerable balancing bias, they are likely to be a valuable tool in bycatch and species distribution modeling, as well as spatial conservation planning, especially for protected species where presence can be rare.

Fit to predict? Eco-informatics for predicting the catchability of a pelagic fish in near real time

The ocean is a dynamic environment inhabited by a diverse array of highly migratory species, many of which are under direct exploitation in targeted fisheries. The timescales of variability in the marine realm coupled with the extreme mobility of ocean-wandering species such as tuna and billfish complicates fisheries management. Developing eco-informatics solutions that allow for near real-time prediction of the distributions of highly mobile marine species is an important step towards the maturation of dynamic ocean management and ecological forecasting. Using 25 yr (1990-2014) of NOAA fisheries' observer data from the California drift gillnet fishery, we model relative probability of occurrence (presence-absence) and catchability (total catch per gillnet set) of broadbill swordfish Xiphias gladius in the California Current System. Using freely available environmental data sets and open source software, we explore the physical drivers of regional swordfish distribution. Comparing models built upon remotely sensed data sets with those built upon a data-assimilative configuration of the Regional Ocean Modelling System (ROMS), we explore trade-offs in model construction, and address how physical data can affect predictive performance and operational capacity. Swordfish catchability was found to be highest in deeper waters (>1,500 m) with surface temperatures in the 14-20°C range, isothermal layer depth (ILD) of 20-40 m, positive sea surface height (SSH) anomalies, and during the new moon (<20% lunar illumination). We observed a greater influence of mesoscale variability (SSH, wind speed, isothermal layer depth, eddy kinetic energy) in driving swordfish catchability (total catch) than was evident in predicting the relative probability of presence (presence-absence), confirming the utility of generating spatiotemporally dynamic predictions. Data-assimilative ROMS circumvent the limitations of satellite remote sensing in providing physical data fields for species distribution models (e.g., cloud cover, variable resolution, subsurface data), and facilitate broad-scale prediction of dynamic species distributions in near real time.

Estimating Limit Reference Points for Western Pacific Leatherback Turtles (Dermochelys coriacea) in the U.S. West Coast EEZ

Biological limit reference points (LRPs) for fisheries catch represent upper bounds that avoid undesirable population states. LRPs can support consistent management evaluation among species and regions, and can advance ecosystem-based fisheries management. For transboundary species, LRPs prorated by local abundance can inform local management decisions when international coordination is lacking. We estimated LRPs for western Pacific leatherbacks in the U.S. West Coast Exclusive Economic Zone (WCEEZ) using three approaches with different types of information on local abundance. For the current application, the best-informed LRP used a local abundance estimate derived from nest counts, vital rate information, satellite tag data, and fishery observer data, and was calculated with a Potential Biological Removal estimator. Management strategy evaluation was used to set tuning parameters of the LRP estimators to satisfy risk tolerances for falling below population thresholds, and to evaluate sensitivity of population outcomes to bias in key inputs. We estimated local LRPs consistent with three hypothetical management objectives: allowing the population to rebuild to its maximum net productivity level (4.7 turtles per five years), limiting delay of population rebuilding (0.8 turtles per five years), or only preventing further decline (7.7 turtles per five years). These LRPs pertain to all human-caused removals and represent the WCEEZ contribution to meeting population management objectives within a broader international cooperative framework. We present multi-year estimates, because at low LRP values, annual assessments are prone to substantial error that can lead to volatile and costly management without providing further conservation benefit. The novel approach and the performance criteria used here are not a direct expression of the “jeopardy” standard of the U.S. Endangered Species Act, but they provide useful assessment information and could help guide international management frameworks. Given the range of abundance data scenarios addressed, LRPs should be estimable for many other areas, populations, and taxa.