Snowmelt runoff in global hydrological models

Runoff during the snowmelt period (hereafter SMR) is an important indicator of water availability and snowmelt floods, and a vital input to the water-food-energy nexus problem in mid-to-high latitude regions. However, despite the abundance of large-scale runoff models and data products, little is known about their SMR performance. Furthermore, diagnosing the key processes that can explain the SMR differences among models remains challenging. To address this issue, this study first utilized three key indicators, i.e., total/maximum discharge in snowmelt periods (Qsum/Qmax) and centroid timing of snowmelt (CTQ), as the first-order indices to assess 15 state-of-the-art models and datasets. Further, an innovative "tree-based model complexity scoring" (TBMCS) method was proposed to score the snow accumulation and snowmelt processes of these models, aiming to quantitatively reveal the relation between model mechanism complexity and their SMR performance. Under long-term mean conditions, we found that the models' simulation of CTQ is better than that of Qsum and Qmax. Overall, the proportion of stations with a ±20% PBias in the simulated Qsum and Qmax is below 30%, while the proportion of stations with a ±5 days difference in the simulated CTQ is below 60%. Most models exhibit larger biases in high-altitude or high-latitude regions, such as the western United States, northern Europe, and the Siberian Plain. Runoff data products are almost always superior to their model counterparts, verifying the role of observation constraints in improving SMR. By using TBMCS, we further found that models with more (less) complex mechanisms often performed better (worse) on CTQ, but this does not apply to Qsum and Qmax. Models focusing more on water balance tend to perform better in simulating Qsum and Qmax. By contrast, models with better energy balance processes do not necessarily yield better water quantity simulations, but can yield better CTQ simulations. This study is the first assessment of the SMR performance of state-of-the-art runoff models and data products. It also innovatively introduces the TBMCS method to challenge the traditional paradigm of "complex models are not necessarily better than simple models", laying the foundation for identifying prioritized areas for future model development.