nse, kge, kge_prime and
log_kge return their conventional statistic, where 1 is a
perfect fit and higher is better. calib_aeme() and run_and_fit()
instead minimise the values returned by FUN_list entries, so these
_loss companions return -1 * the corresponding statistic (lower is
better, -1 = perfect fit) and are what you pass in FUN_list for
calibration:
Usage
nse_loss(obs, sim, na.rm = TRUE)
kge_loss(obs, sim, na.rm = TRUE)
kge_prime_loss(obs, sim, na.rm = TRUE)
log_kge_loss(obs, sim, na.rm = TRUE)
Arguments
- obs
Numeric vector of observed values.
- sim
Numeric vector of simulated (modelled) values.
- na.rm
Logical; should missing values (NA) in either vector be removed
pairwise before computation? Default is TRUE.
Value
numeric; -1 * the corresponding statistic (-1 = perfect fit,
higher = worse fit).
Details
FUN_list <- list(HYD_temp = kge_loss, LKE_lvlwtr = rmse)
mae, rmse and pbias are already
0-is-best, minimise-oriented, so they have no _loss companion - use
them directly.
Examples
obs <- c(1, 2, 3, 4)
sim <- c(1.1, 2.1, 2.9, 4.2)
nse_loss(obs, sim)
#> [1] -0.986
kge_loss(obs, sim)
#> [1] -0.9663051
kge_prime_loss(obs, sim)
#> [1] -0.9661881
log_kge_loss(obs, sim)
#> [1] -0.9622493
