Dependence of Simulated High Flows and Flood Events on Meteorological Forcing Products in the Songhua River Basin: A CLM5–CaMa-Flood Assessment

Abstract

Reliable flood simulation in large cold-region basins requires understanding how meteorological forcing differences propagate through runoff generation and river routing. We compared CMFD, GSWP3v1, and CRUNCEPv7 using a controlled, uncalibrated offline CLM5–CaMa-Flood framework for the Songhua River Basin during 1996–2014, with all non-forcing settings fixed. Evaluation included daily and monthly discharge, seasonal hydrographs, annual maximum daily discharge (AMAX), observed Q95/Q99 thresholds, selected 1998 and 2013 warm-season high-flow cases, runoff-process diagnostics, event-window sensitivity tests, 5000 paired year-wise bootstrap resamples, and auxiliary water-level anomalies. CMFD generally produced the highest r, KGE, and daily NSE, but also the largest positive long-term Bias. CRUNCEPv7 systematically underestimated discharge, whereas GSWP3v1 more often yielded the smallest absolute Bias. For both selected events, CMFD reduced peak and volume underestimation, although peaks remained smoothed and delayed. Event-window precipitation differences did not translate proportionally into CLM5 runoff, and the larger CMFD response involved increases in both surface runoff and subsurface drainage. The event-magnitude ordering remained stable across ±30-, ±45-, and ±60-day windows. Bootstrap results showed a robust CMFD advantage over GSWP3v1 for temporal agreement and efficiency, while several CMFD–CRUNCEPv7 comparisons remained sample-dependent. Forcing-product performance was therefore scale-, metric-, and target-dependent and conditional on the fixed model configuration.