Answers are grounded in this paper and cite their sources.
Snow observations can be used for streamflow prediction, while streamflow observations can be used to improve estimates of past snow amounts. For such snow-streamflow inference to be effective, accurate snow representation should lead to accurate streamflow simulations, and vice versa. We define this as mutual snow-streamflow predictability and quantify it as the correlation between snow and streamflow modeling skill across an ensemble of joint snow-streamflow simulations. To improve snow-streamflow predictability, most effort has focused on improving models and observations. However, even in the absence of observation errors and modeling uncertainty, there are intrinsic limits to how much snow and streamflow can inform each other. Using synthetic experiments across 42 Swiss and Western US catchments, we show that these limits depend strongly on modeling choices and catchment characteristics. Inference is most effective between physically consistent properties, such as between daily melt and streamflow dynamics or between seasonal melt and streamflow volumes. Meanwhile, it is substantially less effective between physically distinct properties, or when using certain commonly used snow and streamflow skill metrics. We further show that inference between daily melt and streamflow dynamics becomes less effective with increasing catchment subsurface buffering, while inference between seasonal melt and streamflow volumes becomes less effective with increased rainfall dilution of the snowmelt signal. These findings imply that, while advances in models and data remain valuable, the maximum achievable predictability is fundamentally constrained by catchment properties, hydroclimatic conditions, and the modeler's choice of inference metrics.
No takes yet. Share an insight, caveat, or question.
Wiersma et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: