Forecasts of Ecological Time Series based on Vector Similarity S-Map

Abstract

S-map is a method based on state space reconstruction provided an efficient method to infer a proxy for the local effect of species interactions and to make reliable forecasts from nonlinear time series, which uses Euclidean distances in the reconstructed state space to determine nearest neighbor points and corresponding weight values. The distance in the state space measures the similarity of the states of two points in the space, which is essential for predicting the evolution of similar states. If the distance measurement is not accurate, it will be problematic to determine adjacent points that correctly reflect the similarity. To solve this problem, in this paper, we propose a S-map that uses vector similarity to select neighboring points and compute weights. The proposed method is first validated on a simulated dataset generated from 12 resource competition models. To further validate our method, we used vector similarity-based S-map for the Time series of sockeye salmon returns from the Fraser River in British Columbia, Canada.