HESSD Hydrology and Earth System Sciences Discussions HESSD Hydrol. Earth Syst. Sci. Discuss. 1812-2116 Göttingen, Germany 10.5194/hess-2018-309 Estimating water residence time distribution in river networks by boosted regression trees (BRT) model Feng Meili https://orcid.org/0000-0002-2305-6934 1 2 3 Pusch Martin 2 Venohr Markus https://orcid.org/0000-0002-1248-3113 2 School of Geographical Sciences, University of Nottingham Ningbo China, Ningbo, 31500, China Leibniz-Institute of Freshwater Ecology and Inland Fisheries, Berlin, 12587, Germany Department of Civil, Environmental and Mechanical Engineering, University of Trento, Trento, 38123, Italy SMART Joint Doctorate programme “Science for the MAnagement of Rivers and their Tidal systems” 20 09 2018 2018 1 27 Copyright: © 2018 Meili Feng et al. 2018 This work is licensed under the Creative Commons Attribution 4.0 International License. To view a copy of this licence, visit https://creativecommons.org/licenses/by/4.0/ This article is available from https://hess.copernicus.org/preprints/hess-2018-309/ The full text article is available as a PDF file from https://hess.copernicus.org/preprints/hess-2018-309/hess-2018-309.pdf

In-stream water residence time (WRT) in river networks is a crucial driver for biogeochemical processes in riverine ecosystems. Dynamics of the WRT are critical for understanding and modelling nutrient retention in lakes and rivers, in particular during flood events when riparian areas are inundated. This study illustrates the potential utility of integrating spatial landscape analysis with machine learning statistics to understand the effects of hydrology and geomorphology on WRT in river networks, especially at large scales. We applied the Boosted Regression Trees (BRT) approach to estimate water residence, a promising multi-regression spatial distribution model with consistent cross-validation procedure, and identified the crucial factors of influence. Reach-average WRTs were estimated for the annual mean hydrologic conditions as well as the flood and drought month, respectively. Results showed that the three most contributing factors in shaping the WRT distribution are river discharge (57 %), longitudinal slope (21 %), and the drainage area (15 %). This study enables the identification of key controlling factors of the reach-average WRT and estimation of WRT under varying hydrological conditions. The resulting distribution model of WRT is an easy to apply and sound approach helping to improve water quality modelling at larger scales and water management approaches aiming to estimate nutrient fluxes in river systems.