PhD thesis:The importance of assessing uncertainty when analyzing climate variability in the Mediterranean


Photo: Josep Llasses Gascón (Autor: Charina Cañas)




Esporles, July 10, 2017. Josep Llasses Gascón defended his doctoral thesis led by Dr. Gabriel Jordà Sánchez, from IMEDEA (CSIC-UIB). The event took place on July 10, 2017 at the University of the Balearic Islands.


The main goal of the present thesis is the characterization of the uncertainties of the Mediterranean Sea observational and simulation products in order to capture climate variability.


Firstly, the capabilities of five marine observational networks to capture temperature and salinity evolution for the last decades of the 20th Century and for the whole 21st Century, as well as the characterization of other local processes such as the deep water formation rate in the western basin, have been tested. Results point out that temperature and salinity annual basin averages and long term trends are properly captured by all networks. Conversely, deep water formation rate in the western basin and other regional processes are largely over or underestimated by all networks.


Secondly, the uncertainties of an ensemble of regional climate simulation hindcasts, spanning for the last decades of the 20th century, have been studied. Specifically, we quantified and analyzed the uncertainties linked to the model reproduction of a key climate process such as the redistribution of heat and salt within the basin. From the characterization of those uncertainties, it has been pointed out that the spread between models is much larger than the ensemble average for the vertical salt transfer and for the heat transfer between 0-150 m and 150-600 m. At lower layers there is a set of models showing a good agreement between them, while others are not correlated with any other model. The mechanisms behind the ensemble spread are not straightforward.


Finally, we explored the reliability of uncertainty estimates in climate simulations when the number of available simulations is limited, which hampers the representativity of the spread as a measure of uncertainty. In those cases, the total uncertainty may be strongly underestimated and a correction factor is recommended to obtain more realistic uncertainty estimates. That corrector factor has been deduced in the case of having single or multiple sources of uncertainty and applied to a real case: the Mediterranean Sea level reconstruction from an ensemble of climate projections. It is shown that the reliability on the uncertainty estimates of each source depends on the number of simulations sampling the source and on the relative contribution to the total uncertainty of the source.


This study pretends to draw the attention on the problems associated to uncertainty estimates when the number of simulations is not adequate and give a useful tool to easily correct the uncertainty estimations.




More information: