ABSTRACT

Mitigation of adverse effects of global warming relies on accurate flow projections under climate change. These projections usually focus on changes in hydrological signatures, such as 100-year floods, which are estimated through statistical analyses of simulated flows under baseline and future conditions. However, models used for these simulations are traditionally calibrated to reproduce entire flow series, rather than statistics of hydrological signatures. Here, we consider this dichotomy by testing whether performance indicators (e.g. Nash-Sutcliffe coefficient) are informative about model ability to reproduce distributions and trends in the signatures. Results of streamflow simulations in 50 high-latitude catchments with the 3DNet-Catch model show that high model performances according to traditional indicators do not provide assurance that distributions or trends in hydrological signatures are well reproduced. We therefore suggest that performance in reproducing distributions and trends in hydrological signatures should be included in the process of model selection for climate change impact studies.

摘要

减轻全球变暖的不利影响依赖于气候变化下的准确流量预测。这些预测通常侧重于水文特征的变化，例如 100 年洪水，这些变化是通过对基线和未来条件下模拟流量的统计分析来估计的。然而，用于这些模拟的模型传统上被校准以再现整个流量序列，而不是水文特征的统计数据。在这里，我们通过测试性能指标（例如 Nash-Sutcliffe 系数）是否能提供有关模型重现签名分布和趋势的能力的信息来考虑这种二分法。使用 3DNet-Catch 模型对 50 个高纬度流域的河流流量模拟结果表明，根据传统指标的高模型性能并不能保证水文特征的分布或趋势得到很好的再现。因此，我们建议在气候变化影响研究的模型选择过程中应包括再现水文特征分布和趋势的性能。