Assimilation of satellite observations into coastal biogeochemical models

This thesis has investigated the improvement of forecasting temperature in a coastal embayment through the assimilation of sea surface temperature (SST) observations. The research was prompted by the increasing pressures on the coastal marine environment. To better manage the environment, an improved understanding of its future state is necessary. Improving the forecasting of temperature advances our knowledge in this direction. Whilst assimilation of SST is routinely carried out for oceans, its use has been minimal in coastal regions, which is more complicated because of anisotropic covariances and a breakdown of geostrophy in the coastal region. Improvements in computing power, and the introduction of ensemble-based assimilation techniques have made the approach followed in this thesis possible. Previous coastal data assimilation had focused on sea level and tidal prediction. More recently, data assimilation has been undertaken with simple ecological models, but temperature has rarely been the subject of research. Furthermore, most coastal assimilation studies have assimilated point scale in situ observations, rather than satellite derived spatial observations, which are the focus of this thesis.The thesis was conducted using a case study of Port Phillip Bay (PPB), a large embayment in south eastern Australia, where in situ temperature measurements gave an accurate indication of the true state of the temperature of water against which to compare forecasts. Over the long term, the SST observations were found to have negligible bias, however a strong diurnal bias was apparent. The model of PPB replicated the main features of PPB well, although the temperature prediction was warm biased. Existing methods to initialise ensembles and to incorporate forecast error were deemed inappropriate and so a new method for initialising the ensemble was developed based on the singular value decomposition (SVD) of a long model simulation. An appropriate ensemble size was determined by using the system variance explained by the singular values. Forecast error was introduced through the development of a rigorous approach to the generation of perturbed forcing data, which aimed to reduce the introduction of bias into the perturbed forcing data set.Initially, various configurations of ensemble assimilation were tested in a synthetic setting by means of twin experiments. The low correlation between temperature and other variables meant that the multivariate analysis gave poor results, relative to a univariate (temperature) analysis. The use of a heatflux biased model was found to make the analysis suboptimal and to distort the forecast error so that the estimate of the forecast uncertainty was inaccurate, however the assimilation into a biased model gave large improvements over an unassimilated forecast. A comparison between the EnKF and EnSRF exploring the impact of perturbed observations found no significant difference between the methods, although EnSRF maintains the shape of the ensemble anomalies better than the EnKF.The actual assimilation of SST data was contrasted against a climatology forecast of PPB temperature. The assimilation of SST without any specific accounting for the diurnal bias improved the forecast, although errors due to observation bias were noted. Attempts to remove this bias using diurnal correction algorithms failed, owing to a larger than expected cool skin. Conditional merging, which combines spatial and in situ observations, was applied to the SST observations and improved the forecast accuracy by reducing the observation bias. An examination of the assimilation innovations indicated where the forecast accuracy could be improved further.By demonstrating the improvement that the assimilation of satellite derived observations can have on forecasting models, this thesis forms a step towards the development of an operational coastal marine forecasting system. In doing so the work undertaken in this thesis and possible extensions to other biogeochemical processes will become a useful tool for managers to protect the coastal marine environment from the multitude of pressures being placed upon it.