Nowadays hydrologic model simulations are widely used to better understand hydrologic processes and to predict extreme events such as floods and droughts. In particular, the spatially distributed LISFLOOD model is currently used for flood forecasting at Pan-European scale, within the European Flood Awareness System (EFAS). Several model parameters can not be directly measured, and they need to be estimated through calibration. In this work we describe how the free software “R” has been used as a single environment to pre-process hydro-meteorological data, to carry out global optimization, and to post-process calibration results in Europe.
Historical daily discharge records were pre-processed for 4062 stream gauges, with different amount and distribution of data in each one of them. The hydroTSM, raster and sp R packages were used to select 700 stations with an adequate spatio-temporal coverage. Selected stations span a wide range of hydro-climatic characteristics. Nine parameters were selected to be calibrated based on previous expert knowledge. Customized R scripts were used to extract observed time series for each catchment and to prepare the input files required to fully set up the calibration thereof. The hydroPSO package was then used to carry out a single-objective global optimization on each selected catchment, by using the Standard Particle Swarm 2011 (SPSO-2011) algorithm. Among the many goodness-of-fit measures available in the hydroGOF package, the Nash-Sutcliffe efficiency was used to drive the optimization. User-defined functions were developed for reading model outputs and passing them to the calibration engine. The long computational time required to finish the calibration at continental scale was partially alleviated by using 4 multi-core machines (with both GNU/Linux and Windows OS) and the “parallel” option available in hydroPSO. Calibration results (not described here) were automatically produced in both text and graphical formats, including a comparison of observed and simulated hydrographs, histograms, boxplots and dotty plots with the parameter values sampled during the optimization. Graphical results allowed a quick assessment of model performance and the identification of individual problems during calibration.
This work illustrates how R proved to be a valuable environment to facilitate modeling, visualization, and data analysis at continental scale in an efficient and reproducible way, without switching to other applications to perform single analyzes. The application discussed here relates to the calibration of a hydrologic model written in pyhton+PCRaster. However, considering the exponentially increasing number of contributed packages, the multi-platform architecture, and the scripting capabilities available, we believe R is a promising environment for hydrology and a similar approach can be applied to a wider class of models requiring parameter optimization.