| Stochastic weather generator like Climate Generator (CLIGEN) is often used to generate daily weather data when measured data are insufficient or unavailable. It has been used to synthesize daily weather that statistically resembles the present climate or generate daily weather for ungauged areas through spatial interpolation of model parameters from adjacent gauged sites. Meanwhile, CLIGEN is suited to generate daily weather series from monthly projections of general circulation model for assessing crop production and soil erosion under climate change. The objectives were to: (1) evaluate the reproducibility of the latest version (v5.22564) in generating daily, monthly, and yearly precipitation depths at 12 stations, as well as storm patterns including storm duration (D), relative peak intensity (ip) and peak intensity (rp) at 10 stations dispersed across the Loess Plateau; (2) test whether an exponential distribution for generating D and a distribution-free approach for inducing desired rank correlation between precipitation depth and D can improve storm pattern generations. (3) evaluate and improve the ability of the CLIGEN to generate non-precipitation parameters, including dew point tempertature(Tdp) daily maximum (Tmax) and minimum (Tmin) air temperatures, solar radiation (SR), and wind velocity (u) at 12 meteorological stations located in the Loess Plateau of China.the detailed results are as following:(1) Mean absolute relative errors (MARE) for simulating daily, monthly, annual and annual maximum daily precipitation depth across all 12 stations were 3.5%, 1.7%, 1.7% and 5.0% for the means and 5.0%, 4.5%, 13.0% and 13.6% for the standard deviations, respectively. The model reproduced the distributions of monthly and annual precipitation depths very well (P>0.3), but the distribution of daily precipitation depth was less well produced. The first–order, two–state Markov chain algorithm is adequate for generating precipitation occurrence for the Loess Plateau of China. The CLIGEN generated storm patterns poorly. It underpredicted means and standard deviations of D for storms≥10 mm by -60.4% and -72.6%, respectively. Compared with D, ip and rp were slightly better reproduced. MAREs of means and standard deviations were 21.0% and 52.1% for ip, and 31.2% and 55.2% for rp, respectively. When an exponential distribution was used to generate D, MAREs were reduced to 2.6% for the means and 7.8% for the standard deviations. However, ip estimation became much worse, when exponentially generated D even after introduction of proper correlation with precipitation depth was used to compute ip with the MAREs of 128.9% for the means and 241.1% for the standard deviations. Overall, CLIGEN is adequate in generating precipitation depths and occurrence for the region, but the storm pattern generation needs improvement. Exponential distribution reproduced D well, but resulted in overestimation of ip, probably due to the overestimation of rp. For better storm pattern generation for the region, precipitation depth, D, and rp should be generated correlatively using approaches like Copula.(2) CLIGEN reproduced daily Tmax and, Tmin reasonably well. The t- and F-tests showed that neither means nor standard deviations of measured data were significantly different from those of the CLIGEN-generated data at P = 0.01 for all stations. Means and distributions of daily Tdp were reproduced very well; however, standard deviations were less well reproduced with significant differences at P = 0.01 for 4 out of 12 stations for the F-test. Mean and standard deviation for daily SR were much better reproduced by our modified CLIGEN at all stations, though distributions were slightly worsened. Daily u was reproduced well after fixing a unit conversion error with an absolute relative error (RE) of 0.17% for the means and 0.71% for the standard deviation. Mean of same-day temperature range (Tmax1-Tmin1) and one-day lag temperature ranges for both (Tmax1-Tmin2) and (Tmax2-Tmin1) of the CLIGEN-generated data were reproduced well with the absolute RE close to zero. However, compared with the measured data, standard deviations of Tmax1-Tmin1 were consistently underestimated with the MRE of -38.7%, and those of Tmax1-Tmin2 and Tmax2-Tmin1 were consistently overestimated with the respective MREs being 55.8% and 19.6% for all stations. Seasonal serial correlations of SR and cross correlation between temperatures and SR were much better reproduced by the modified model. Specifically, compared to CLIGEN (v5.111), Tmin and Tdp were improved considerably in v5.22564, as well as means of Tmax1-Tmin2 and Tmax2-Tmin1, but standard deviations of Tmax1-Tmin1 were worsened. Standard deviations of Tmax1-Tmin2 and Tmax2-Tmin1 generated by v5.111 and v5.22564 were similar. Furthermore, generation of SR and u was similar in both versions, but was significantly improved in our modified v5.22564. Due to the improvement in SR generation, seasonal serial correlations of SR and cross correlation between temperatures and SR were also improved.(3) Inverse Distance Weighting (IDW), Kriging (KRI), Global Polynomial Interpolation (GPI) and Local Polynomial Interpolation (KPI) satisfied for interpolation of CLIGEN's parameters (precipitation, maximum and minmum temperatures). The paired-samples t test showed that there were no significant differences between meansured data and interpolated data at P=0.05 level. However, there were significant differences among results from each interpolation methods. For means, standard deviations and skewness coefficients of precipitation considered, KPI provided the best or nearly best, cross-validated results, in comparison to three other methods, followed by KRI. GPI and IDW were worse. LPI and IDW were better than GPI and IDW for interpolating precipitation occurrence. Different menthods have their advantages for maximum and minimum temperatures'interpolation, but LPI was better than others as a whole. Overall, LPI is one of the best methods for interpolating the parameters of precipitation, maximum and minimum temperatures for CLIGEN.
|
PDF Full Text Request