Quantifying connectivity uncertainty arising from circulation modelling inaccuracy.

Ocean modelling has become an increasingly important tool to study population connectivity and is our only tool to anticipate changes in dispersal routes in future climates. To estimate the uncertainties in model predictions, a comparison was made between the simulated currents and in situ observations in the Gulf of Lion over the period of 2009-2013. The uncertainties in Eulerian current values were described using several statistical parameters, like the bias, the root mean square (RMSE), the naturalised root mean square (NRMSE), the Hannah and Heinold parameter (HH) and the correlation. Another parameter that was introduced was the correctness, which states the percentage of time the model was deemed “correct”, based on low HH values (<75%) and high correlation (>0.25). So far, the model simulated the flow speed correctly 60-70% of the time and the relative deviation between observed and simulated flow speed was about 10%. Furthermore, ensembles of Lagrangian tracks were simulated accounting for uncertainties in Eulerian flow speed. These uncertainties were either correlated to speed values or chosen according to their statistical distribution. The Lagrangian tracks were analysed to construct connectivity matrices with and without these Eulerian uncertainties. Resulting deviation in retention and larval transfer arising from flow speed uncertainty were quantified.