Parameter estimation and sensitivity in Lagrangian oil spill models

Abstract

Lagrangian oil spill models, like many environmental models, are constructed from a large set of submodels, each with their own set of parameters and constants. The submodels belong to different domains of science, such as physical oceanography, petroleum chemistry, sediment and coastal geoscience, biochemistry, and ecotoxicology. Generally, these submodels have been derived and fitted to observations independently and within separate contexts, before being combined in some sequence to become the oil spill model. In this way, oil spill models have a large set of input parameters and constants that affect model output. When oil spill models are compared with observed data, it is challenging to know which of this large set of parameters should be adjusted to improve the model's performance against observations. Given the scale of oil spill models, parameters from different submodels can be correlated in non-trivial ways. It is also generally not known to which parameters the oil spill model displays the most sensitivity. This is further complicated by parameter sensitivity being dependent on environmental input such as wind speeds and ocean currents. Here, we present a sensitivity analysis of an idealised oil spill model, with submodels for oil surface spreading, entrainment of oil by breaking waves, resurfacing of oil, emulsification of surface oil by water uptake, and increase in viscosity of oil due because of weathering and water uptake. We discuss our results in context of improving oil spill models from observations, both laboratory data and sea-truth in the form of past spill events.