From calibration to parameter learning: Harnessing the scaling effects of big data in geoscientific modeling

Abstract

The behaviors and skills of models in many geosciences (e.g., hydrology and ecosystem sciences) strongly depend on spatially-varying parameters that need calibration. A well-calibrated model can reasonably propagate information from observations to unobserved variables via model physics, but traditional calibration is highly inefficient and results in non-unique solutions. Here we propose a novel differentiable parameter learning (dPL) framework that efficiently learns a global mapping between inputs (and optionally responses) and parameters. Crucially, dPL exhibits beneficial scaling curves not previously demonstrated to geoscientists: as training data increases, dPL achieves better performance, more physical coherence, and better generalizability (across space and uncalibrated variables), all with orders-of-magnitude lower computational cost. We demonstrate examples that learned from soil moisture and streamflow, where dPL drastically outperformed existing evolutionary and regionalization methods, or required only ~12.5% of the training data to achieve similar performance. The generic scheme promotes the integration of deep learning and process-based models, without mandating reimplementation.

Deep learning approaches for improving prediction of daily stream temperature in data‐scarce, unmonitored, and dammed basins

“…The data used here represent the best-instrumented sites from USGS, and 415 locations are only a tiny fraction of the millions of river reaches in the United States. In the future, the combination of processbased modelling and machine learning may allow more robust predictions on a global scale, which are already started by other scholars (Jia et al, 2021;Karpatne et al, 2018;Read et al, 2019;Tsai et al, 2021).…”

A Physics-Informed, Machine Learning Emulator of a 2D Surface Water Model: What Temporal Networks and Simulation-Based Inference Can Help Us Learn about Hydrologic Processes

“…This analysis is not intended to be a formal SBI or model calibration; rather, the purpose is to further explore the validity of our emulators. This approach is much more simple than other calibration approaches that might employ an evolution search algorithm [41], gradient-based method to adjust parameters in a series of more limited model simulations [42] or even use ML approaches to replace the calibration routine [43]. Given this proof of concept, future work should include more complex frameworks, including those that loop parameters back to the original physical model simulation or use a more formalized Bayesian framework [44].…”

“…lations [42] or even use ML approaches to replace the calibration routine [43]. Given th proof of concept, future work should include more complex frameworks, including tho that loop parameters back to the original physical model simulation or use a more forma ized Bayesian framework [44].…”

Surrogate Model for Shallow Water Equations Solvers with Deep Learning

“…Such gradient information is extremely useful for solving previously difficult or unsolvable problems. For example, Tsai et al (2021) recently proposed a novel differentiable parameter learning (dPL) framework to integrate big-data DL and differentiable PBMs for parameter calibration. Another example is the use deep learning surrogate model to perform riverine bathymetry inversion (Ghorbanidehno et al, 2021).…”