Generalized additive models (GAMs) were applied to examine the relative influence of various factors on fishery performance, defined as nominal catch‐ per‐unit‐effort (CPUE) of swordfish (Xiphias gladius) and blue shark (Prionace glauca) in the Hawaii‐based swordfish fishery. Commercial fisheries data for the analysis consisted of a 5 year (1991–1995) time series of 27 901 longline sets. Mesoscale relationships were analysed for seven physical variables (latitude, longitude, SST, SST frontal energy, temporal changes in SST (ΔSST), SST frontal energy (ΔSST frontal energy) and bathymetry), all of which may affect the availability of swordfish and blue shark to the fishery, and three variables (number of lightsticks per hook, lunar index, and wind velocity) which may relate to the effectiveness of the fishing gear. Longline CPUE data were analysed in relation to SST data on three spatiotemporal scales (18 km weekly, 1°‐weekly, 1°‐monthly). Depending on the scale of SST data, GAM analysis accounted for 39–42% and 44–45% of the variance in nominal CPUE for swordfish and blue shark, respectively. Stepwise GAM building revealed the relative importance of the variables in explaining the variance in CPUE. For swordfish, by decreasing importance, the variables ranked: (1) latitude, (2) time, (3) longitude, (4) lunar index, (5) lightsticks per hook, (6) SST, (7) ΔSST frontal energy, (8) wind velocity, (9) SST frontal energy, (10) bathymetry, and (11) ΔSST. For blue shark, the variables ranked: (1) latitude, (2) longitude, (3) time, (4) SST, (5) lightsticks per hook, (6) ΔSST, (7) ΔSST frontal energy, (8) SST frontal energy, (9) wind velocity, (10) lunar index, and (11) bathymetry. Swordfish CPUE increased with latitude to peak at 35–40°N and increased in the vicinity of temperature fronts and during the full moon. Shark CPUE also increased with latitude up to 40°N, and increased westward, but declined abruptly at SSTs colder than 16°C. As a comparison with modelling fishery performance in relation to specific environmental and fishery operational effects, fishery performance was also modelled as a function of categorical time (month) and area (2° squares) variables using a generalized linear model (GLM) approach. The variance accounted for by the GLMs was ≈ 1–3% lower than the variance explained by the GAMs. Time series of swordfish and blue shark CPUE standardized for the environmental and operational variables quantified in the GAM and for the time‐area effects in the GLM are presented. For swordfish, both nominal and standardized time series indicate a decline in CPUE, whereas the opposite trend was seen for blue shark.

No abstract

Imaging research laboratories are rapidly creating machine learning systems that achieve expert human performance using opensource methods and tools. These artificial intelligence systems are being developed to improve medical image reconstruction, noise reduction, quality assurance, triage, segmentation, computer-aided detection, computer-aided classification, and radiogenomics. In August 2018, a meeting was held in Bethesda, Maryland, at the National Institutes of Health to discuss the current state of the art and knowledge gaps and to develop a roadmap for future research initiatives. Key research priorities include: 1, new image reconstruction methods that efficiently produce images suitable for human interpretation from source data; 2, automated image labeling and annotation methods, including information extraction from the imaging report, electronic phenotyping, and prospective structured image reporting; 3, new machine learning methods for clinical imaging data, such as tailored, pretrained model architectures, and federated machine learning methods; 4, machine learning methods that can explain the advice they provide to human users (so-called explainable artificial intelligence); and 5, validated methods for image de-identification and data sharing to facilitate wide availability of clinical imaging data sets. This research roadmap is intended to identify and prioritize these needs for academic research laboratories, funding agencies, professional societies, and industry.

Temperature-depth recorders (TDRs) were attached to pelagic longline gear in the Hawaii-based commercial fishery to obtain actual fishing depths and to test the accuracy of catenary algorithms for predicting fishing depths. Swordfish gear was set shallow by typically deploying four hooks between successive floats. The observed depth of the settled deepest hook had a median value of 60 m for 333 swordfish sets. Tuna longline gear deployed more hooks between floats (mean = 26.8), and the observed median depth of the deepest hook was 248 m (n = 266 sets). Maximum gear depth was predicted from estimates of the longline sag ratio and catenary algorithms; however, depth was not predicted for all TDR-monitored sets because estimating sag ratios proved problematic. Swordfish sets had less slack in the main line and correspondingly smaller catenary angles (median = 54.2• ) than tuna sets (median = 63.7• ). Median values of the predicted catenary depth were 123 m for swordfish sets (n = 203) and 307 m for tuna sets (n = 198). Shallow swordfish sets reached only ∼50% of their predicted depth, while deeper tuna sets reached about 70%. These values indicated that capture depths using traditional catenary equations may be biased without the benefit of TDRs affixed to longlines. Generalized linear models (GLMs) and generalized additive models (GAMs) were developed to explain the percentage of longline shoaling as a function of predicted catenary depth and environmental effects of wind stress, surface current velocity, and current shear. The GAM explained 67.2% of the deviance in shoaling for tuna sets and 41.3% for swordfish sets. Predicted catenary depth was always the initial variable included in the stepwise process, and the inclusion of environmental information in the GAM approach explained an additional 10-17% of the deviance compared to the GLMs. The explanatory ability of the environmental data may have been limited by the scale of the observations (1 • in space; weekly or monthly in time) or the geometric (transverse versus in-line) forcing between the environment and longline set. Longline gear models with environmental forcing affecting shoaling may be improved in future studies by incorporating contemporaneous environmental information, although this may restrict analyses to fine-scale experimental longlines. Published by Elsevier B.V.