A Framework for Improving Methods for Inverse Modeling of Streambed Fluxes and Thermal Diffusivity Using Temperature Time Series
Abstract
Recent decades have seen significant progress in methods to estimate streambed fluxes and parameters, and increasingly complex studies demand further improvements. One of the challenges to improving methods has been in communicating exactly how methods differ and how those differences matter. We propose a systematic framework describing the steps used in applying these methods and improving them: 1) identifying model structure, 2) performing signal analysis, 3) estimating parameters and their uncertainty, and 4) conducting experiments. The first three steps define how methods differ, and the fourth is how we discern among them and show how each individual component affects results. Model structure comprises the partial differential equation (PDE) and boundary conditions (BCs). Signal processing is used to obtain relevant information from the time series and filter out unwanted contributions (e.g., noise). Parameter estimation has traditionally applied single values of amplitude and/or phase to estimate single velocity and/or diffusivity values, and there is a need to apply concepts of sampling and overdetermination to improve estimates and provide uncertainty derived from measurements. Experiments target such information as is needed to improve components. Existing data analysis packages comprise a mix of different components, and this framework clarifies the unique combinations of steps in terms the different underlying boundary condition, signal processing, and estimation procedures. This can highlight strengths and weaknesses of different approaches. For example, it is commonly assumed that non-sinusoidal temperature signals cause errors, but that depends primarily on the signal processing methods applied and is not a feature of the linear time invariant models commonly used. Similarly, vertical bed heterogeneity is not inherently an issue with these general methods, but choices in BCs and PDE are needed to appropriately extract the information. Some combinations of measurement and signal processing also provide opportunity for multiple estimates of parameters at a location, allowing for estimates of uncertainty. Understanding how the different components work individually and together will help frame experiments to test models and advance the field more rapidly.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.H33E..06L