ID:
publications-1577
Type:
PEER REVIEWED ARTICLE
Year:
2017
Authors:
Md Abul Ehsan Bhuiyan , Efthymios I. Nikolopoulos , Emmanouil N. Anagnostou , Pere Quintana-Seguà , Anaïs Barella-Ortiz
Title:
A Nonparametric Statistical Technique for Combining Global Precipitation Datasets: Development and Hydrological Evaluation over the Iberian Peninsula
Venue/Journal:
DOI:
10.5194/hess-2017-268
Research type:
Data Management & Analytics
Water System:
Precipitation & Ecological Systems
Technical Focus:
Abstract:
Abstract. This study investigates the use of a nonparametric, tree-based model, Quantile Regression Forests (QRF), for combining multiple global precipitation datasets and characterizing the uncertainty of the combined product. We used the Iberian Peninsula as the study area, with a study period spanning eleven years (2000–10). Inputs to the QRF model included three satellite precipitation products, CMORPH, PERSIANN, and 3B42 (V7); an atmospheric reanalysis precipitation and air temperature dataset; satellite-derived near-surface daily soil moisture data; and a terrain elevation dataset. We calibrated the QRF model for two seasons and two terrain elevation categories and used it to generate rainfall ensembles for these conditions. We then carried an evaluation based on a high-resolution, ground-reference precipitation dataset (SAFRAN) available at 5 km/1 h resolution and further used generated ensembles to force a distributed hydrological model (the SURFEX land-surface model and the RAPID river routing scheme). To evaluate relative improvements and the overall impact of the combined product in hydrological response, we compared its streamflow simulation results with the results of simulations from the individual global precipitation and reference datasets. We concluded that the proposed technique could generate realizations that successfully encapsulate the reference precipitation and provide significant improvement in streamflow simulations, with reduction in systematic and random error on the order of 20 %–99 % and 44 %–88 %, respectively, when considering the ensemble mean.
Link with Projects:
603608
Link with Tools:
Related policies:
ID:
