No abstract

No abstract

The statistical analysis of environmental monitoring data is an important issue in detecting year-to-year changes in levels and timings of important ecological events. In many cases, this trend detection must explicitly view interannual changes from the context of an evolving seasonal cycle. This study analyses weekly sampling data from a long-term ocean monitoring program near Halifax, Nova Scotia, Canada. A state space model of evolving seasonality with a quadratic trend is fit to these time series to extract the signal from the noisy and irregularly sampled data. The procedure uses Kalman filter innovations to estimate model parameters. A fixed interval smoother is then applied to estimate the system state. The resultant state estimates are subjected to a trend analysis carried out with respect to key ecological events: the level and timing of the peak, trough, spring, and fall abundances. These events are identified using derivative information, followed by a regression-based trend analysis. The analysis found a number of significant linear trends in the biogeochemical variables considered. More generally, the approach used here is suitable for use with monitoring data exhibiting a unimodal seasonal signal with noise and missing values.