Prädiktion der Ozeantemperatur im räumlichen und zeitlichen Verlauf mit Hilfe dynamischer linearer Modelle

Mehr zum Buch

Atmospheric and oceanic processes, key components of the global climate system, exhibit variability across space and time. This analysis employs statistical approaches rather than deterministic modeling due to the challenges posed by large, incomplete data sets that often display complex interactive relationships. Traditional space-time methods are limited because direct specification of the joint space-time covariance structure is often unfeasible, given spatial non-stationarities and nonseparable space-time interactions. This paper develops dynamic linear (state-space) models that capture the temporally dynamic structure within an autoregressive framework while incorporating a spatially descriptive component. To manage extensive observational areas, dimension reduction of the spatial field is achieved using empirical orthogonal functions. The methodology is applied to sea surface temperature measurements from the Northwest European Shelf during 1983-1992, predicting observed point measurements to a grid of approximately 20 km (1/3° east-west and 1/5° north-south) using the Kalman filter. Unlike other spatiotemporal state-space models, this approach does not require fixed measurement locations and allows for the dynamic integration of a large-scale trend component, along with efficient parameter estimation.