| paretoCatchments {hydromad} | R Documentation |
Functions to evaluate model realisations and structures that perform well across simulation periods in multiple catchments
paretoCatchments(stat, csv.name = NA, objectives = "r.squared") getDominatedLong(res.dom)
stat |
data.frame of results, including columns: |
csv.name |
Optional path and prefix in which to write table indicating whether model is dominated across periods ("csv.name_isdominated_models_catchments.csv") and to write a table of non-dominated models (" |
objectives |
Vector of column names containing performance measures to be used by
|
res.dom |
data.frame to be converted to long format, from wide format produced
by |
For each Catchment in the data.frame of results, models that
are dominated across simulation periods are identified using
areModelsDominated. These results are summarised in a number of ways:
The number of dominated model realisations in each catchment indicates how demanding the Pareto Filtering was.
The number of realisations where each model structure is non-dominated suggests the best model structure for each catchment, in terms of its ability to perform well across simulation periods.
Table of non-dominated models, in wide format showing performance in each simulation period. These models could be used as an ensemble of unrejected models.
Minimum performance of all non-dominated models in each period. Unacceptably low performance indicates that the model selection criteria or model structure do not operate well in particular time periods.
Minimum and range of performance of all non-dominated models across each period. A large range of non-dominated performance (RNDP) indicates variation in performance across periods. The minimum value can be used as a threshold to define a feasible set to allow ensemble forecasts. If models at or below the minimum value are rejected, non-dominated model realisations would be rejected.
getDominatedLong is a utility function used by paretoCatchments
No value returned, used for its side-effects. Prints results described
in details. Optionally writes csv files, see csv.name.
Joseph Guillaume
data(YeAl97) paretoCatchments(YeAl97,objectives="E")