A resampling procedure for generating conditioned daily weather sequences

Abstract

[1] A method is introduced to generate conditioned daily precipitation and temperature time series at multiple stations. The method resamples data from the historical record ''nens'' times for the period of interest (nens = number of ensemble members) and reorders the ensemble members to reconstruct the observed spatial (intersite) and temporal correlation statistics. The weather generator model is applied to 2307 stations in the contiguous United States and is shown to reproduce the observed spatial correlation between neighboring stations, the observed correlation between variables (e.g., between precipitation and temperature), and the observed temporal correlation between subsequent days in the generated weather sequence. The weather generator model is extended to produce sequences of weather that are conditioned on climate indices (in this case the Niño 3.4 index). Example illustrations of conditioned weather sequences are provided for a station in Arizona (Petrified Forest, 34.8°N, 109.9°W), where El Niño and La Niña conditions have a strong effect on winter precipitation. The conditioned weather sequences generated using the methods described in this paper are appropriate for use as input to hydrologic models to produce multiseason forecasts of streamflow.

Incorporating circulation statistics in bias correction of GCM ensembles: hydrological application for the Rhine basin

“…Previous studies have included atmospheric variability in methods for model bias correction by conditioning existing methodologies such as quantile regression, quantile mapping and weather generators (resampling) on the governing circulation regime (Wilby et al 1998b;Huth 1999;Clark et al 2004;Friederichs and Hense 2007;Hundecha and Bárdossy 2008;Jagger and Elsner 2009;Bárdossy and Pegram 2011).…”

Translating climate forecasts into agricultural terms: advances and challenges

“…It has also been used successfully for predicting reservoir inflow from seasonal rainfall predictors in Northeast Brazil (DeSouza & Lall 2003). The KNN analog approach can also be applied on shorter time steps to probabilistically sample subsets of past weather observations based on the degree of similarity of current and historical values of a given feature vector that may include atmospheric indicators from SST-forced GCM outputs (Clark et al 2004, Gangopadhyay et al 2005. Appropriate selection criteria can preserve moments of the historical distribution, as well as observed spatial and temporal correlations and correlations among variables.…”

Conditional simulation of remotely sensed rainfall data using a non-Gaussian v-transformed copula

“…A great deal of effort has been made to develop stochastic models for simulation of multivariate rainfall fields (e.g., gauge, radar and satellite). In a number of studies, rainfall fields are directly simulated using different multivariate stochastic techniques (e.g., [40,58,28,37,45,9,6]). Ciach et al [8] developed an operational approach based on empirical investigations of joint samples of radar and ground surface data whereby the radar rainfall uncertainties consist of a systematic distortion function and a stochastic component.…”