Skill of modelled runoff against observed runoff at gauges

Per station: the annual error 100 * (mod - obs) / obs and log ratio log(mod / obs), and over the 12 months the Nash-Sutcliffe efficiency of the monthly climatology (mm) and of the monthly shares of the annual total. Summaries by group give the metrics of Chapman et al. (2018) (mean, median and mean absolute error in percent, and the share of stations within +/- 20 %) plus the log-ratio bias and spread.

Usage

wet_flow_validate(x, groups = NULL, min_obs = 10)

Arguments

x

data.frame(station_number, month, obs, mod) in mm, month 0 for the annual total and 1-12 for months. Modelled values below 0 should be floored before they reach here.

groups

Optional data.frame(station_number, <group columns>); each group column gets its own summary.

min_obs

Minimum observed annual runoff (mm) for a station to count.

Value

list(stations, summary). stations: one row per station with obs, mod, err_pct, log_ratio, nse_month, nse_share. summary: one row per group value (and an "all" row) with n, n_low, n_no_mod (stations with no modelled value, left out), mean_err_pct, median_err_pct, mae_pct, within_20, log_bias_median, log_sd, nse_month_median, nse_share_median.

Details

A station whose observed annual runoff is below min_obs mm is left out of every summary and counted in n_low: percentage errors on near-zero runoff are not informative, and a log ratio of zero or negative runoff is undefined. Months need all 12 values in both series; a station without them gets NA monthly metrics.