Abstract
Abstract. Terrestrial water variables are the key to understanding ecosystem processes, feed back on weather and climate, and are a prerequisite for human activities. To provide context for local investigations and to better understand phenomena that only emerge at large spatial scales, reliable information on continental-scale freshwater dynamics is necessary. To date streamflow is among the best-observed variables of terrestrial water systems. However, observation networks have a limited station density and often incomplete temporal coverage, limiting investigations to locations and times with observations. This paper presents a methodology to estimate continental-scale runoff on a 0.5° spatial grid with monthly resolution. The methodology is based on statistical upscaling of observed streamflow from small catchments in Europe and exploits readily available gridded atmospheric forcing data combined with the capability of machine learning techniques. The resulting runoff estimates are validated against (1) runoff from small catchments that were not used for model training, (2) river discharge from nine continental-scale river basins and (3) independent estimates of long-term mean evapotranspiration at the pan-European scale. In addition it is shown that the produced gridded runoff compares on average better to observations than a multi-model ensemble of comprehensive land surface models (LSMs), making it an ideal candidate for model evaluation and model development. In particular, the presented machine learning approach may help determining which factors are most relevant for an efficient modelling of runoff at regional scales. Finally, the resulting data product is used to derive a comprehensive runoff climatology for Europe and its potential for drought monitoring is illustrated.
Highlights
The accuracy of the estimated runoff fields is assessed with respect to data that were not used for model identification and compared to an ensemble of comprehensive land surface models
In a second experiment the focus is on the model’s ability to estimate runoff dynamics at time steps (t) that were not used for model identification
The framework is based on the assumption that runoff at any location in space can be modelled as a function of gridded predictors, including both atmospheric variables and land parameters
Summary
Terrestrial water storages and fluxes are key variables in the Earth system, as they are a primary control for many ecosystem processes (e.g. Ciais et al, 2005; Granier et al, 2007; Reichstein et al, 2013; Guan et al, 2015), influence weather and climate through land–atmosphere interactions (e.g. Koster et al, 2004; Seneviratne et al, 2010) and are the basis for many human activities (e.g Döll et al, 2009; Vörösmarty et al, 2010; Orlowsky et al, 2014). Information of the historical space and time evolution of variables such as evapotranspiration, soil moisture, groundwater and runoff are of great interest Most of these variables are only observed at few locations in space and often with irregular temporal coverage, limiting analysis to the well-monitored regions. Data products providing reliable estimates of the historical space–time evolution of these variables for large, continental-scale regions are of vital importance. Such data products will allow investigating terrestrial water dynamics at locations without observations but more importantly allow for the study of processes and phenomena that emerge on large, continental, scales.
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call