There is growing recognition that climate change is impacting the ocean western boundary current system. In the Pacific, the Kuroshio and its offshore Kuroshio-Oyashio Extension (KOE) play a central role in North Pacific climate and impact the social-ecological dynamics of countries that rely on marine ecosystem services (e.g. fisheries). In the thesis we have used a combination of observations and modelling approaches to understand how past and projected changes in the physical environment of KOE impact social-ecological dynamics linked to the fish industry of Japan and North Pacific more widely. The thesis is articulated in 3 Tasks.
In Task 1, we analyze the climate variability and change of the KOE over the historical and future projection period 1920-2100. We perform this task using Coupled Model Inter comparison Project 5 (CMIP5) models and a large ensemble from the Community Earth System Model (CESM-LE) output runs. The reason for considering also the CESM-LE runs is that they give the possibility to explore how the variance of the KOE in one model (e.g. a fixed set of dynamics) responds to anthropogenic forcing when compared to the range of natural variability of the CESM-LE model.
In Task 2, we have used an Empirical Dynamical Model approach to characterize the joint statistics of the physical and social-ecological environmental system (SEES) that is relevant to climate and fisheries. To define the states of the SEES we use three international fish databases, (1) the Large Marine Ecosystem (LME, 9,000 fish stocks), (2) the NOAA fishery database referred to as Restricted Access Management (RAM, 300 fish stock) and the (3) the Food and Agriculture Organization (FAO, 1400 fish stocks). Among the approaches used to explore the relationship between KOE’s climate and the SEES response, we have developed a Linear Inverse Model (LIM) approach that has been very successful to simulate and predict the KOE physical climate and its relation to large-scale Pacific dynamics such as El Niño Southern Oscillation (ENSO), the Pacific Decadal Oscillation (PDO), and others.
In Task 3, the results from the LIM are compared with a Gated Recurrent Unit neural network that allow us to consider the non-linearities that exist in the relationship between climate and SEES, and the external human pressures on fisheries. Additionally, the machine learning methods can train millions of parameters and are becoming more widely applied in climate science and big data (e.g. large number of fish stocks). However, it has yet to be shown that SEES indicators can be modeled, forecast and diagnosed using deep learning approaches. Also, the sensibility of these algorithms to the changes in physical environment is still object of research.
Friday, November 18, 2022 - 3:00pm
ES&T L1114 & https://us06web.zoom.us/j/81197188079?pwd=YStKbFBsbTJpNzhZL0p0QUpvczdlZz09