Global patterns in base flow index and recession based on streamflow observations from 3394 catchments

• Published in: Water resources research Vol. 49; no. 12; pp. 7843 - 7863
• Main Authors: Beck, Hylke E., van Dijk, Albert I. J. M., Miralles, Diego G., de Jeu, Richard A. M., (Sampurno) Bruijnzeel, L. A., McVicar, Tim R., Schellekens, Jaap
• Format: Journal Article
• Language: English
• Published: Washington Blackwell Publishing Ltd 01-12-2013; John Wiley & Sons, Inc
• Subjects: Base flow; Catchment scale; Catchments; Climate change; ensemble; Evaporation; Evaporation rate; Flow characteristics; Flow index; global; Global climate; Hydrology; neural network; Neural networks; recession; Recessions; Regression analysis; Roads & highways; Snow; Snow-water equivalent; Stream discharge; Stream flow; Surface water; ungauged; Water depth
• ISSN: 0043-1397; 1944-7973
• DOI: 10.1002/2013WR013918

Numerous previous studies have constructed models to estimate base flow characteristics from climatic and physiographic characteristics of catchments and applied these to ungauged regions. However, these studies generally used streamflow observations from a relatively small number of catchments (<200) located in small, homogeneous study areas, which may have led to less reliable models with limited applicability elsewhere. Here, we use streamflow observations from a highly heterogeneous set of 3394 catchments (<10,000 km2) worldwide to construct reliable, widely applicable models based on 18 climatic and physiographic characteristics to estimate two important base flow characteristics: (1) the base flow index (BFI), defined as the ratio of long‐term mean base flow to total streamflow; and (2) the base flow recession constant (k), defined as the rate of base flow decay. Regression analysis results revealed that BFI and k were related to several climatic and physiographic characteristics, notably mean annual potential evaporation, mean snow water equivalent depth, and abundance of surface water bodies. Ensembles of artificial neural networks (ANNs; obtained by subsampling the original set of catchments) were trained to estimate the base flow characteristics from climatic and physiographic data. The catchment‐scale estimation of the base flow characteristics demonstrated encouraging performance with R2 values of 0.82 for BFI and 0.72 for k. The connection weights of the trained ANNs indicated that climatic characteristics were more important for estimating k than BFI. Global maps of estimated BFI and k were obtained using global climatic and physiographic data as input to the derived models. The resulting global maps are available for free download at http://www.hydrology-amsterdam.nl.
Key Points
We trained neural networks to estimate baseflow index and recession rate
Streamflow observations from 3394 catchments were used for the training
First global maps of baseflow index and recession rate