An Investigation of Nonlinear Forecasting for Improved Stock Projections: Understanding Variability in Fish Populations
Link to NOAA Strategic Plan: NOAA's Mission Goal 1: Protect, Restore, and Manage the Use of Coastal and Ocean Resources Through an Ecosystem Approach to Management
RESEARCH OBJECTIVES AND SPECIFIC PLANS TO ACHIEVE THEM
1) The project objective of understanding sources of variability (both anthropogenic and environmental) will be addressed by using historical icthyoplankton data for the CALCOFI domain. These are unique data that can be used to answer one of the classic questions in marine fishery management: whether fishing itself will increase or dampen the population variability of targeted fish species.
2) The project objective using nonlinear methods to improve prediction directly addresses the overall NOAA Fisheries mission. It is essential information for setting harvest targets of fished species in the CALCOFI domain.
3) The nonlinear methods used here directly address the mission of ecosystem-based management. This is done explicitly in the state space embedding procedure, in addressing the dimensionality question, and by identifying mutually co-predictive subsystems that may yield models with higher predictive skill. In the latter case we are determining the boundaries of the ecosystem view that is required for prediction (how complex does the best model need to be?).
4) Demonstrating the applicability of the forecasting technology and refining the methods to apply specifically to data of this kind (as must certainly be done) is of general utility to its future applications to other fisheries.
Specific questions and problems we address are as follows:
1) How predictable are fish larvae data in each ecological domain (taken as a composite over all species in an ecological region as defined by the expert systems habitat classification).
2) Are exploited species inherently less variable than unexploited species? Are exploited species inherently less predictable than unexploited ones? Are there any systematic differences here? Can fishing affect dynamic stability? Is there a generic theoretical result? What empirical evidence is there for any effects of fishing predicted by models?
3) Is there much cross-predictability between regions? For example, does the model built for oceanic species, also capture the composite dynamics of the coastal species? If this is true it suggests that the environmental signal in spatially averaged coastal data produces an integrated average of the climatic changes that may also be occurring in the open ocean. This is important for next generation modeling efforts that may incorporate physical forcing variables.
4) What is the effective dimensionality of each of these problems? That is, how many variables are required to model the problem, in order to obtain a given level of predictability?
5) How does generation time affect scales of predictability?
6) Can cross-predictability between species be used to discern dynamically coupled subsystems? Such subsystems would represent natural communities, or functional assemblages. This can be determined by a hierarchical search that produces a matrix of mutual predictability between species and species groups. If so, can one build better forecast models by focusing on each of these more tightly connected ecological subsystems?
RESEARCH ACCOMPLISHMENTS
We have accomplished items 1-5 above as follows:
1) Found fishing increases boom and bust variability of exploited populations. This is a classical question in fisheries science that we were able to answer generally and empirically for the first time. The important implication of this work is that the destabilization of the population is a consequence of common fisheries practices that target the larger older individuals. This work was published in Nature October 2006, and was mentioned as an important finding by NOAA Administrator Vice Admiral Lautenbacher in a speech at the National Academies of Sciences later that fall.
2) Found nonlinear forecast methods are effective for fisheries. These methods work best when the time series composite is constrained by habitat type.
3) Found low dimensional nonlinearity in the population dynamics of both exploited and unexploited populations.
4) Confirmed that fishing results in a truncated age and size structure for the population, and further related this to destabilization of exploited populations.
