- Title
- Evaluating different approaches for using the box-cox transformation to model heteroscedasticity in residual errors of hydrological models
- Creator
- McInerney, D.; Thyer, M.; Kavetski, D.; Kuczera, G. A.
- Relation
- Hydrology and Water Resources Symposium 2014. Proceedings of the Hydrology and Water Resources Symposium 2014 (Perth, W.A. 24-27 February, 2014) p. 937-944
- Publisher
- Engineers Australia
- Resource Type
- conference paper
- Date
- 2014
- Description
- The Box-Cox transformation is widely used for normalizing residuals and modelling heteroscedasticity in hydrological models. In this paper we compare the predictive performance of this transformation by applying it to three catchments with distinct climatological conditions. In particular, we apply the Box-Cox transformation with the parameter λ fixed at pre-defined values (common approach), and compare it with λ inferred. We find that when λ is inferred, variance is stabilized and predictive reliability is highest; however, as noted in previous publications, inferring λ can result in unrealistically large predictions (i.e., "explosiveness"). In comparison, we find the log transformation (λ = 0 ) yields a residual error model that performs reasonably well in terms of both reliability and predictive precision for all three catchments, with reduced potential for explosiveness.
- Subject
- box-cox transformation; engineering; hydraulic engineering; environmental engineering
- Identifier
- http://hdl.handle.net/1959.13/1293975
- Identifier
- uon:18710
- Identifier
- ISBN:9781922107190
- Language
- eng
- Reviewed
- Hits: 1870
- Visitors: 1848
- Downloads: 0