Advancing hydrologic science through process-based evaluation of models

How rapidly advancing climate change will impact the water cycle and its extremes remains poorly understood and is the origin of much uncertainty. This uncertainty limits our ability to build societal and ecosystem resilience – contributing to policy challenges for adaptation to water scarcity and other hydro-climatic risks. In this context, we rely on hydrologic simulation models to provide robust short-term predictions as well as long-term projections of water cycle dynamics across scales. Even though advancements in observational systems and increasingly detailed simulation models enable us to observe and simulate the water cycle at unprecedent resolutions over large domains, intercomparison studies still reveal inconsistent emergent model behavior. These large domain models are difficult to constrain using current observational datasets given that these are often highly imbalanced, while available theory provides only limited guidance regarding which hydrologic processes we can expect to dominate in diverse climates and landscapes. At the same time, the performance of machine learning models improves rapidly, bypassing process knowledge and therefore questioning the basic need for scientific understanding.

In this talk, I argue that process-based evaluation is an important bridge between hydrologic theory, observations and simulation models – even in the presence of high model complexity and significant data imbalances. I will discuss examples of how process-based evaluation can elicit controlling factors on hydrologic processes by utilizing hydrologically relevant gradients in both simulated and observed data. Thus, demonstrating how this approach can provide a pathway towards assessing and ultimately improving the consistency between perceived and simulated hydrologic process controls over large domains.