Runoff predictions in ungauged catchments in southeast Tibetan Plateau

- Rainfall-runoff daily models are good for predicting monthly runoff in TP.
- The spatial proximity approach outperforms the physical similarity approach.
- Predicting skills decline with decrease in mean annual precipitation.
- Incorporation of snowmelt processes into rainfall-runoff models is not helpful.

SummaryThe Tibetan Plateau (TP) plays a key role on both hydrology and climate for southern and eastern Asia. Improving runoff predictions in ungauged catchments in the TP is critical for surface water hydrology and water resources management in this region. However, a detailed runoff prediction study in this region has not been reported yet. To fill the gap, this study evaluates two regionalization approaches, spatial proximity and physical similarity, for predicting runoff using two rainfall-runoff models (SIMHYD and GR4J). These models are driven by meteorological inputs from eight large non-nested catchments (4000-50,000 km2) in the Yarlung Tsangpo River basin located in southeast TP. For each catchment, the two models are calibrated using data from the first two-thirds of the observation period and validated over the remaining period. The calibrated and validated Nash-Sutcliffe Efficiency of monthly runoff (NSE) varies from 0.73 to 0.93 for the SIMHYD model, and are similar to or slightly better than those obtained for the GR4J model. The incorporation of snowfall-snowmelt processes into the rainfall-runoff models does not noticeably improve the runoff predictions in the study area. The main reason is that monthly runoff is dominated by summer precipitation and snowfall in winter accounts for a small percentage (less than 14%). The results from both models show that the spatial proximity approach marginally outperforms the physical similarity approach and both approaches are better than random selection of a donor catchment. This is consistent with recent regionalization studies carried out in Europe and Australia. The study suggests that conceptual rainfall-runoff models are powerful and simple tools for monthly runoff predictions in large catchments in southeast TP, and incorporation of more catchments into regionalization can further improve prediction skills.