The contributions of Daren M. Carlisle, James Falcone, David M. Wolock and Michael R. Meador were prepared as part of their official duties as US Government employees.
Predicting the natural flow regime: models for assessing hydrological alteration in streams †
Article first published online: 3 MAR 2009
Copyright © 2009 John Wiley & Sons, Ltd.
River Research and Applications
Volume 26, Issue 2, pages 118–136, February 2010
How to Cite
Carlisle, D. M., Falcone, J., Wolock, D. M., Meador, M. R. and Norris, R. H. (2010), Predicting the natural flow regime: models for assessing hydrological alteration in streams . River Res. Applic., 26: 118–136. doi: 10.1002/rra.1247
- Issue published online: 26 JAN 2010
- Article first published online: 3 MAR 2009
- Manuscript Accepted: 15 JAN 2009
- Manuscript Revised: 24 NOV 2008
- Manuscript Received: 25 AUG 2008
- natural flow regime;
- predictive models;
- random forests;
- hydrologic modification
Understanding the extent to which natural streamflow characteristics have been altered is an important consideration for ecological assessments of streams. Assessing hydrologic condition requires that we quantify the attributes of the flow regime that would be expected in the absence of anthropogenic modifications. The objective of this study was to evaluate whether selected streamflow characteristics could be predicted at regional and national scales using geospatial data. Long-term, gaged river basins distributed throughout the contiguous US that had streamflow characteristics representing least disturbed or near pristine conditions were identified. Thirteen metrics of the magnitude, frequency, duration, timing and rate of change of streamflow were calculated using a 20–50 year period of record for each site. We used random forests (RF), a robust statistical modelling approach, to develop models that predicted the value for each streamflow metric using natural watershed characteristics. We compared the performance (i.e. bias and precision) of national- and regional-scale predictive models to that of models based on landscape classifications, including major river basins, ecoregions and hydrologic landscape regions (HLR). For all hydrologic metrics, landscape stratification models produced estimates that were less biased and more precise than a null model that accounted for no natural variability. Predictive models at the national and regional scale performed equally well, and substantially improved predictions of all hydrologic metrics relative to landscape stratification models. Prediction error rates ranged from 15 to 40%, but were ≤25% for most metrics. We selected three gaged, non-reference sites to illustrate how predictive models could be used to assess hydrologic condition. These examples show how the models accurately estimate pre-disturbance conditions and are sensitive to changes in streamflow variability associated with long-term land-use change. We also demonstrate how the models can be applied to predict expected natural flow characteristics at ungaged sites. Copyright © 2009 John Wiley & Sons, Ltd.