Modeling marine fish species spatiotemporal distributions utilizing NASA earth system data in a maximum entropy framework (February, 2018)

Date: 
Thursday, February 1, 2018

Fishermen rely on their on-the-water experience to know where and when to find certain species of fish in the Gulf of Maine, and managers use their knowledge of fish and the fishery to design management policies, such as seasonal closures, aimed at ensuring sustainability. Increasingly, climate change is altering historical relationships between the physical environment and our fishery resources, and challenging both fishermen and managers. Conditions in the Gulf of Maine are changing at a relatively rapid pace with warming in this region over the past decade outpacing 99% of the global ocean. The ability to understand where fish are likely to be and when they are likely to be there is critical for the sustainable management of the region's valuable fisheries.

The Maine NASA EPSCoR (Established Program to Stimulate Competitive Research) project is a collaborative project aimed at developing models that link remotely sensed ocean conditions with the spatiotemporal distributions of key fish species. Scientists from the Gulf of Maine Research Institute, Bigelow Laboratory for Ocean Sciences, and NASA Jet Propulsion Laboratory collaborated on using a statistical approach called Maximum Entropy (MaxEnt) to build models relating the occurrence of fish to high-resolution environmental information measured by NASA satellites. This approach was applied to several important fishery species including Atlantic herring, Atlantic mackerel, and butterfish in the Northwest Atlantic Shelf area. Monthly habitat suitability maps were produced (e.g., Figure 1), the relative influence of environmental factors on fish distributions was assessed (Figure 2), and the predictive ability of the MaxEnt models was evaluated using hindcasts of fish distributions in recent years (e.g., Figure 3). Results from the project showed that the MaxEnt models did a good job explaining past changes in fish distributions, showed good predictive ability in hindcasts, and therefore have the potential to provide forecasts of future distributions of these species which serve an important role as prey for commercially important fish (e.g., bluefin tuna). The models also provided insights into what processes drive the distributions of these fish. This MaxEnt modeling framework allowed us to integrate fish records from commercial fisheries and high-resolution environmental data from NASA satellites to describe the spatiotemporal distributions of important fishery species.

Another aim of this work was to build the capacity of scientists in Maine to apply NASA data products in fisheries and marine ecosystem research. Scientists from the Gulf of Maine Research Institute, Bigelow Laboratory for Ocean Sciences, and NASA Jet Propulsion Laboratory worked collaboratively on identifying several NASA data products that are relevant to spatial modeling work in the Maine NASA EPSCoR project. These included the MUR Sea Surface Temperature (SST) product, chlorophyll-a concentration estimates from MODIS on Aqua, and surface winds from the MERRA2 atmospheric reanalysis. A suite of software tools was developed for accessing these NASA data products, and workflows were designed to support different modeling efforts. Data outputs included aggregating data from regional subsets or from satellite grid points contained in the coarser fisheries grid areas, daily and monthly aggregations, calculation of statistical values such as monthly means and percentiles, and calculation of wind speed and direction from vector data.

The Maine NASA EPSCoR project also supported early-career scientist Dr. Lifei Wang to participate in the Satellite Remote Sensing Training Program at the Cornell University. Through this training program, she was able to learn and practice skills to acquire, analyze, and visualize datasets derived from a variety of satellite sensors (e.g., SeaWiFS, MODIS, and MERIS), and apply NASA's SeaDAS software to derive mapped imagery of geophysical parameters (e.g., chlorophyll or CDOM) from raw data obtained through the Ocean Color Web Data Server. The program also provided a good opportunity for scientists in the fields of oceanography and fisheries to get familiar with the underlying principals leading to the measurement from space of sea surface temperature, ocean color, ocean wind speed, and ocean topography.

EPSCoR is a national program designed to foster and build partnerships between government, industry, and academic institutions, to promote regional development of expertise in aerospace and space related research activities. NASA is a federal agency partner in the program.

Figure 1.  Habitat suitability maps of Atlantic herring for January (a), April (b), July (c), and October (d) based on MaxEnt.

Figure 1. Habitat suitability maps of Atlantic herring for January (a), April (b), July (c), and October (d) based on MaxEnt. The color of each ten-minute-square cell indicated the monthly habitat suitability value calculated by MaxEnt. For the scale bar, the color at the top (yellow) indicated highest habitat suitability, and the color at the bottom (dark blue) indicated lowest habitat suitability. Blank (white) pixels occurred where there were missing values for chlorophyll-a concentration estimates from satellite remote sensing because of cloud coverage.

 

Figure 2.  Relative importance of environmental factors in influencing the monthly distributions of Atlantic herring (a), Atlantic mackerel (b), and butterfish (c) based on MaxEnt and the performance of monthly models.

Figure 2.  Relative importance of environmental factors in influencing the monthly distributions of Atlantic herring (a), Atlantic mackerel (b), and butterfish (c) based on MaxEnt and the performance of monthly models.  Abbreviations for environmental factors: SST Sea Surface Temperature, NAO North Atlantic Oscillation, AMO Atlantic Multidecadal Oscillation, GSI Gulf Stream Index.  Performance of monthly MaxEnt models was evaluated by AUC, the Area Under the receiver operator characteristic (ROC) Curve.

 

Figure 3.  Monthly habitat suitability hindcasts for Atlantic herring in the Northwest Atlantic Shelf area for July of 2002-2013.

Figure 3.  Monthly habitat suitability hindcasts for Atlantic herring in the Northwest Atlantic Shelf area for July of 2002-2013.

 

 

 

Written by Lifei Wang (Gulf of Maine Research Institute)