1 · Data
2 · Flow-regime strata
3 · What-if: perturb the sim
Applies to the simulated series only — watch which KGE term reacts. sim′ = (1+bias)·[μₛ + amp·(sim−μₛ)], then shifted by lag.
Verdict — whole record
Hydrograph — obs vs sim
Dashed lines: the high / low flow thresholds (from observed).
Flow duration curves
Weibull plotting position p = i/(n+1). Shaded band boundaries = your strata cuts.
Stratified scores
Segment NSE uses the segment mean of observed as its benchmark — read it as “skill within this regime”, not comparable to whole-record NSE.
Equations (exactly as computed)
r = Pearson correlation(obs, sim)
α = σ_sim / σ_obs (sample sd, n−1) [2009]
γ = (σ_sim/μ_sim) / (σ_obs/μ_obs) [2012 replaces α]
β = μ_sim / μ_obs
NSE = 1 − Σ(sim−obs)² / Σ(obs−μ_obs)²
PBIAS= 100 · Σ(sim−obs) / Σ(obs)
RMSE = √[ Σ(sim−obs)² / n ]
Reading the decomposition
β off → volume problem: infiltration, %impervious, RDII R-values. α (or γ) off → variability/flashiness: subcatchment width, storage, routing attenuation. r off → timing/shape: width, overland n, conduit roughness, lag. Fix in that order — volume first, timing last (SWMM calibration priority). Guidance ranges follow Moriasi et al. (2007) and Knoben et al. (2019): KGE > 0.75 good, 0.5–0.75 acceptable; note KGE = −0.41 (not 0) is the “mean-flow benchmark” line. Refs: Gupta et al. 2009; Kling et al. 2012.