pub enum Objective {
Nse,
Kge,
LogNse,
Dc,
NegRmse,
}Expand description
率定目标(皆“越大越好“;RMSE 取负)。
Calibration objective. All variants are maximised (RMSE is negated so
that “higher score = better fit” uniformly). Select via Objective::Nse
/ Objective::Kge / Objective::LogNse / Objective::Dc /
Objective::NegRmse when calling calibrate or evaluate.
Variants§
Nse
Nash–Sutcliffe Efficiency, NSE = 1 − Σ(o−s)² / Σ(o−ō)².
Range (−∞, 1]; 1.0 = perfect. Equivalent to the deterministic
coefficient DC of GB/T 22482. The standard calibration default for
flood-forecast models.
Kge
Kling–Gupta Efficiency, KGE = 1 − √( (r−1)² + (α−1)² + (β−1)² ).
Range (−∞, 1]; 1.0 = perfect. Decomposes error into correlation,
variability, and bias — preferred over NSE when bias and peak timing
both matter.
LogNse
Log-space NSE — log_nse; NSE computed on ln(Q + ε).
Range (−∞, 1]. Emphasises low-flow / recession fit by compressing
high flows logarithmally; use for drought or base-flow-dominated
regimes. ε = 1e-6 avoids ln(0).
Dc
Deterministic coefficient DC (GB/T 22482-2008 确定性系数).
Numerically identical to Objective::Nse — provided as a separate
variant so that downstream tooling can distinguish “NSE as optimisation
target” from “DC as the Chinese-standard grade metric” when reporting.
NegRmse
Negative RMSE, -√( (1/N) Σ(o−s)² ).
Range (−∞, 0]; 0.0 = perfect. Maximising this is equivalent to
minimising RMSE. Same units as the discharge series (typically m³/s).
Prefer Objective::Nse / Objective::Kge unless you need
absolute-error (not normalised-error) scoring.