Premium
Multisite simulation of daily precipitation and temperature in the Rhine Basin by nearest‐neighbor resampling
Author(s) -
Buishand T. Adri,
Brandsma Theo
Publication year - 2001
Publication title -
water resources research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.863
H-Index - 217
eISSN - 1944-7973
pISSN - 0043-1397
DOI - 10.1029/2001wr000291
Subject(s) - precipitation , resampling , standard deviation , quantile , k nearest neighbors algorithm , statistics , climatology , mathematics , environmental science , meteorology , geography , geology , computer science , artificial intelligence
The method of nearest‐neighbor resampling is extended to simultaneous simulation of daily precipitation and temperature at multiple locations over a large area (25 stations in the German part of the Rhine basin). Nearest neighbors refer here to historical days for which the observed weather is closest to that of the simulated weather for a given day. Resampling is done from these nearest neighbors to obtain the weather variables for the next day. The nearest neighbors are defined in terms of a weighted Euclidean distance to a feature vector containing summary statistics of the daily precipitation and temperature fields (spatial averages, fraction of stations with precipitation, and principal components). The inclusion of atmospheric circulation variables in the feature vector is also studied. There is a weak tendency to underestimate the standard deviations and autocorrelation coefficients of daily precipitation and temperature and the standard deviations of the monthly precipitation totals and monthly mean temperatures. However, the underprediction of these second‐order moment statistics is not statistically significant if the number k of nearest neighbors in the resampling procedure is small ( k ≈ 5) and the dimension q of the feature vector is low ( q ≈ 3). A small systematic underprediction is also observed for the quantiles of the distributions of the N ‐day winter maximum precipitation amounts. The spatial dependence of these extremes and the distributions of N ‐day maximum snowmelt are adequately reproduced. Long‐duration simulations show that realistic unprecedented multiday precipitation amounts can be generated.