Effects to Model Calibration of Split-Calibration and Reduced Frequency Data

Abstract

In the Feugh at Heugh Head catchment (NE Scotland), we tested the potential of regime-specific split calibration and reduced temporal resolution to improve HBV-light streamflow accuracy and efficiency. Using a long daily hydro- meteorological record 1985–2023 (hydrological year 1 Oct–30 Sep). The model was warm-started (1985–1986), calibrated (1987–2008), and validated (2009–2023). Four families of experiments were undertaken: (i) a conventional daily baseline; (ii) split calibration with winter (high-flow) and summer (low-flow) parameter sets; (iii) reduced data frequency (weekly, monthly); and (iv) combined split calibration at reduced frequency. Performance was assessed with Kling–Gupta Efficiency (KGE), Nash–Sutcliffe Efficiency (NSE), percent bias (PBIAS), annual volume error (ΔQ), and flow-regime volume errors (FHV/FLV), alongside runtime per simulated year; diagnostic hydrographs and seasonal flow-duration curves complemented the metrics. Split calibration improved representation of extremes: relative to the daily baseline (KGE 0.45; NSE 0.51; FHV −43.6%; FLV +15.2%), the Split-High configuration raised KGE to 0.60 and NSE to 0.62 and reduced FHV/FLV to −27.3%/+4.4%, yet the long-standing water-balance shortfall persisted (PBIAS ≈ −20%; ΔQ ≈ +165–167 mm yr⁻¹). Temporal aggregation offered modest efficiency but degraded hydrological credibility: weekly runs were ~35% faster and showed higher seasonal KGE (≈0.63) but large positive PBIAS (≈+27–31%) and substantial ΔQ deficits, while monthly runs performed poorly (winter KGE ≈ 0; NSE < 0) and suppressed peaks. Combining split calibration with reduced frequency failed pre-specified acceptance thresholds (ΔKGE, |PBIAS|, |ΔQ|, |FHV|, |FLV|), although weekly winter appeared superficially strong in KGE. We conclude that decision-grade applications should retain daily resolution with regime-aware calibration; weekly resolution is suitable only for exploratory screening with bias correction, while monthly is not recommended.