Data Assimilation Seminar
Prof. John Wells (March 19, 2024, 16:00-17:30)
| Affiliation | Ritsumeikan University, Japan / RIKEN, R-CCS |
|---|---|
| Title | Tools to evaluate 3D hydrodynamic simulations of stratified lakes, and to assimilate limited temperature observations |
| Abstract |
Stresses on global freshwater resources will grow acutely with climate change and population growth, and development of ICT tools to better manage water resources will correspondingly increase in importance. To inform scientific understanding, better manage the ecosystem, and possibly to track disastrous pollutant release, this work aims to develop a system to nowcast 3D hydrodynamics in Lake Kinneret (Israel) and Lake Biwa (Japan). In the first part I consider how Principal Component Analysis (PCA) might help validate results from numerical simulations of a stratified lake, and compactly summarize the comparison by extending the well-known Taylor Diagram. Excepting our earlier contribution (Wells et al, SIL Congress 2022), PCA has not been applied to validate simulations in limnology. We compare the first few PCA spatial modes of temperature profiles in the simulation versus observation, together with the variance in each mode. To validate the time-varying weights requires a common basis, taken herein as the spatial modes from the observations. For analysis of temperatures, especially for the short unsteady record measured in spring in Lake Biwa, we also consider whether temperature "fluctuations" at each depth should be defined with respect to the sample mean at that depth, or alternatively to a background "deep, cold" reference temperature. Prior efforts to assimilate observations into 3D lake simulations have been suprisingly limited. In a second part, I will discuss ungoing efforts to assimilate temperature observations at various depths of a single mooring near the center of Lake Biwa. My current implementation simply nudges water temperatures every hour, equally in horizontal planes, in proportion to the difference between the 24-hour boxfiltered temperature observation, and its corresponding simulated value. This suffices to suppress the numerical diffusion of heat that plagues most 3D hydrodynamic simulations, and therefore maintain a fairly realistic profile of buoyancy frequency. |