New Scientific Paper Published on NOAA's Highly Migratory Species Predictive Spatial Modeling Tool

Predictive Spatial Modeling Tool

NOAA Fisheries has published a peer-reviewed scientific article describing the Highly Migratory Species Predictive Spatial Modeling Tool (PRiSM) and its application in assessing the performance of closed areas. PRiSM combines observer data and environmental data to predict where and when fishery interactions may occur. Independent scientists agree that PRiSM is a scientifically accepted and vetted tool. 

As described in the article, the agency used a series of statistical validation approaches to increase confidence in PRiSM results. These approaches compared actual catch data to the PRiSM probability of fishery interaction predictions. The tool then uses that probability to generate metrics that assess the performance of closed areas.

While PRiSM is a valuable tool, it does not make any management decisions. Any potential decisions to modify closed areas would be made through the rulemaking process, which incorporates public comment.

Data-Driven Predictions

PRiSM combines observer data and environmental data to predict where and when fishery interactions may occur. This information can then be used to:

• Determine where fishing or research vessels should be allowed to fish to collect data in the field
• Assess spatial management areas
• Determine essential fish habitat
• Assist in ecosystem-based fisheries management
• Understand the impacts of climate change on fisheries

The fishery observer data is collected by observers on pelagic and bottom longline fishing vessels and the ocean condition data is collected via satellites and survey vessels at the location and time of fishing and includes:

• Water temperature
• Chlorophyll concentrations
• Salinity
• Currents and fronts
• Sea surface height (altimetry)
• Bottom depth

“PRiSM uses the relationships between all environmental and gear information to predict the probability of fishery interactions," says Dan Crear, a NOAA Research Associate and one of the article’s authors. "For example, PRiSM would predict a higher probability of species interaction with fishing gear in areas where water temperature, salinity, current, and other environmental features were previously shown to be ideal for that species.” 

Below is a map of PRiSM’s probability of fishery interaction for the bycatch of billfish (blue marlin, white marlin, roundscale spearfish, and sailfish combined) in the pelagic longline fishery during average conditions in the month of April from 2016–2018.

Image

Closed Area Assessment

There are a number of HMS closed areas that have not been assessed since their implementation in the early 2000s. Since then, ocean conditions, species distributions, and stock statuses have changed and additional regulations have been implemented. “Because fishing vessels cannot fish in these closed areas,” says Crear, “there is little data to determine if the closed areas are still performing as intended. But through PRiSM, NOAA Fisheries generated four metrics to objectively assess the performance of a closed area for one or more species.” 

The scientific article shows that the PRiSM model could be used to provide additional information to fishery managers when there is a lack of fishery-dependent data available. Managers can use PRiSM, in conjunction with other information, to help inform HMS spatial management decisions. 

Validation Approach

PRiSM uses a habitat suitability modeling or species distribution modeling framework. These types of models often relate species occurrence or abundance data with environmental data to understand these ecological relationships. 

PRiSM validation approaches compare model results to actual catch data from observers on fishing vessels. If model results are similar to actual catch data the validation reports the model performed well.

NOAA Fisheries performed three validation approaches for PRiSM. All three validation approaches apply different variations of a method called cross validation. This method divides the data into multiple groups and then chooses all but one of the groups to model (training dataset), then uses that model to predict the probability of fishery interactions for the group left out (testing dataset). Lastly, it generates a metric to see how well the predictions match up with actual fishery interactions from the testing dataset. It then repeats this process with all of the groups. NOAA Fisheries divided the fisheries data up three different ways:

Random

PRiSM divided the data into 10 groups randomly. This approach provides a more general understanding of how well the model does at predicting. 

Spatial

PRiSM divided the data into four or five spatial groups that were systematically layered over the fisheries dataset. This approach ensured the model performed well at predicting in different areas. 

Temporal

PRiSM divided the data into different time period groups. This ensured the model performed well at predicting during recent time periods (2016–2018); the time period NOAA Fisheries is focused on in the scientific article. 

Results from these three validation approaches showed that for the six bycatch species in the study, PRiSM performed well and the predicted probability of fishery interactions are reliable. Those six species were:

Pelagic longline species

• Leatherback sea turtle 
• Shortfin mako 
• Billfish group

Bottom longline species

• Dusky shark 
• Sandbar shark 
• Scalloped hammerhead

Image

PRiSM Closed Area Metrics Explanation 

NOAA Fisheries developed a series of four metrics through PRiSM that managers can use to objectively evaluate closed areas. Fishery managers can use these metrics, in conjunction with other information, to guide data collection, and inform decisions regarding current and future spatial management measures. The metrics were generated from the probability of fishery interactions (i.e., occurrence probabilities) predicted by PRiSM. 

Metric 1

Metric 1 compares the average occurrence probability inside a closed area to the actual average occurrence rates from fisheries data outside the closed area for a given species. These values can then be compared for each month of the closure. If the value inside a closed area is greater than the value outside, then generally the closed area is located in a better location compared to the average location outside the closed area for that species, as shown in the example below.

Image

Metrics 2–4

Metrics 2-4 focus on areas where the highest occurrence probabilities or highest probability of fishery interactions are predicted to occur for a given species. These areas are called high risk areas for that species. The threshold when calculating high risk areas for a species or species group is based on multiple factors, including Endangered Species Act status, overfished/overfishing status, and community importance. 

In addition to single species high risk area, PRiSM can also calculate where multiple species high risk areas overlap.

Metric 2

Metric 2 creates a ratio that compares the median occurrence probability of high risk area inside the closed area to the median occurrence probability of high risk area outside the closed area for each month of the year. A ratio >1 indicates that the closed area may be protecting a higher risk area compared to outside the closed area.

Image

Metric 3

Metric 3 calculates the percent of total high risk area that occurred inside the closed area for each month of the year for a given species. This metric tells us the percent of high risk area the closed area protects. The higher the percent of high risk area inside the closed area, the greater the protection for the given species.

Image

Metric 4

Metric 4 calculates the percent of the closed area that could protect high risk area for each month of the year for a given species. A low percent means there is a small amount of high risk area included in the closed area. A higher percent means more of the closed area consists of high risk area for that species.

Image