You are here:
USING FISHER INFORMATION TO DETECT GRADUAL AND RAPID ECOSYSTEM REGIME SHIFTS
Citation:
Mayer**, A, C. W. Pawlowski**, AND H C. Cabezas*. USING FISHER INFORMATION TO DETECT GRADUAL AND RAPID ECOSYSTEM REGIME SHIFTS. Presented at Joint annual mtg. of the Ecological Society of American & Inter. Society of Ecological Modeling, Savannah, GA, August 03 - 08, 2003.
Impact/Purpose:
To inform the public
Description:
As ecosystems experience perturbations of varying regularity and intensity, they may either remain within the state space neighborhood of the current regime, or "flip" into the neighborhood of a regime with different characteristics. Although the possibility of such regime shifts needs to be incorporated into informed environmental monitoring and management efforts, the identification of regime shifts typically requires an intimate knowledge of the ecosystem dynamics as well as extensive supporting data. We demonstrate the utility of a dynamic information index that lowers the threshold of knowledge required to detect shifts in ecosystem regime. Information theory can be used to quantify the order inherent in systems in spite of imperfect observations or ?signals' from the source system. We use the Fisher Information index to develop a measure of the variation of a dynamic steady state based on a generated probability density function. Periodic systems that spend more time in a few segments of their trajectory have a high Fisher information, whereas periodic systems spending an equal amount of time in all the segments of their trajectory have zero information. This form of the Fisher Information is a measure of system order. A transition from one periodic regime to another appears as a change in Fisher information. We calculate changes in Fisher Information from time series data for ecosystems at several scales (Lake Erie, the Pacific Ocean, and the global climate system), and demonstrate that the differences in Fisher Information behavior are correlated with regime shifts. Fisher Information changes more gradually for systems undergoing a gradual shift, whereas the index changes rapidly (either positively or negatively) for rapid shifts. Fisher Information can be calculated for an unlimited number of variables, although including spurious variables that are not associated with ecosystems dynamics decreases the sensitivity of the index to regime shifts.