Enumeration QualityIndicatorMethod

Quality indicator methods for comparing radar rain predictions against gauge observations.

Each method quantifies prediction quality differently. All methods can be normalized to a 0-100 scale (0=worst, 100=best) via SpeedMatrix.NormalizeQualityIndicator.

Default method: KLING_GUPTA

Enumeration Members

DELTA RATIO SUCCESS_RATE RMSE MAPE NASH_SUTCLIFFE KLING_GUPTA

Enumeration Members

DELTA

DELTA: "delta"

Average absolute difference between rain prediction and gauge observation.

Formula: mean(|rain - gauge|)

  • Range: 0 to +Infinity
  • Perfect score: 0
  • Normalized: inverted using reference max (default 10 mm/h)

RATIO

RATIO: "ratio"

Average of min/max ratio between rain and gauge values.

Formula: mean(min(rain, gauge) / max(rain, gauge))

  • Range: 0 to 1
  • Perfect score: 1
  • Handles both over- and under-prediction symmetrically
  • Skips points where both values are zero

SUCCESS_RATE

SUCCESS_RATE: "success_rate"

Percentage of points where the min/max ratio meets a quality threshold.

Formula: count(ratio >= threshold) / total * 100

  • Range: 0 to 100
  • Perfect score: 100
  • Default threshold: 0.8 (configurable via successThreshold)
  • Both-zero points count as successful (correct "no rain" prediction)

RMSE

RMSE: "rmse"

Root Mean Square Error — penalizes large errors more than small ones.

Formula: sqrt(mean((rain - gauge)^2))

  • Range: 0 to +Infinity
  • Perfect score: 0
  • More sensitive to outliers than DELTA
  • Normalized: inverted using reference max (default 10 mm/h)

MAPE

MAPE: "mape"

Mean Absolute Percentage Error — error relative to observed value.

Formula: mean(|rain - gauge| / gauge) * 100

  • Range: 0 to +Infinity (commonly 0-100+)
  • Perfect score: 0
  • Skips zero gauge values to avoid division by zero
  • Normalized: inverted using reference max (default 100%)

NASH_SUTCLIFFE

NASH_SUTCLIFFE: "nash_sutcliffe"

Nash-Sutcliffe Efficiency — hydrology standard comparing prediction to mean observation.

Formula: 1 - sum((rain - gauge)^2) / sum((gauge - meanGauge)^2)

  • Range: -Infinity to 1
  • Perfect score: 1
  • 0 means model is as good as predicting the mean
  • Negative values mean model is worse than predicting the mean

See

KLING_GUPTA for an improved alternative that decomposes performance

KLING_GUPTA

KLING_GUPTA: "kling_gupta"

Kling-Gupta Efficiency — decomposes performance into correlation, variability bias, and volume bias.

Formula: 1 - sqrt((r - 1)^2 + (alpha - 1)^2 + (beta - 1)^2)

Where:

  • r = Pearson correlation coefficient (timing/pattern accuracy)

  • alpha = stdDev(rain) / stdDev(gauge) (variability ratio)

  • beta = mean(rain) / mean(gauge) (volume bias ratio)

  • Range: -Infinity to 1

  • Perfect score: 1 (r=1, alpha=1, beta=1)

  • Stricter than NSE on systematic bias

  • KGE > -0.41 is considered better than mean prediction

This is the default quality indicator method.

See

Gupta et al. (2009) "Decomposition of the mean squared error and NSE performance criteria"

Settings

Member Visibility

Theme

On This Page

DELTARATIOSUCCESS_RATERMSEMAPENASH_SUTCLIFFEKLING_GUPTA