Diagnostics

Completeness

deterministic probabilistic

Because of missing data, the optional use of temporal/conditional, and/or the optional use of bootstrapping, the number of time steps in the period used to compute the evaluation metrics may vary and may be reduced unreasonably. This is why the completeness diagnostic is available: it returns the number of time steps included in each period considered, i.e. once missing data have been discarded, and masking and bootstrapping temporal subsets have been performed.

deterministic-only

The returned shape of the completeness diagnostic is (series, subsets, samples).

Python

>>> res = evalhyd.evald(
...     obs, prd,
...     metrics=["NSE"],
...     diagnostics=["completeness"]
... )

R

> res <- evalhyd::evald(
+     obs, prd,
+     metrics = c("NSE"),
+     diagnostics = c("completeness")
+ )

CLI

$ ./evalhyd evald "obs.csv" "prd.csv" "NSE" --to_file \
> --diagnostics "completeness"

probabilistic-only

The returned shape of the completeness diagnostic is (sites, lead times, subsets, samples).

Python

>>> res = evalhyd.evalp(
...     obs, prd,
...     metrics=["CRPS_FROM_ECDF"],
...     diagnostics=["completeness"]
... )

R

> res <- evalhyd::evalp(
+     obs, prd,
+     metrics = c("CRPS_FROM_ECDF"),
+     diagnostics = c("completeness")
+ )

CLI

$ ./evalhyd evalp "./obs" "./prd" "CRPS_FROM_ECDF" --to_file \
> --diagnostics "completeness"