Evaluation of Linear and Non-Linear Downscaling Methods in Terms of Daily Variability and Climate Indices: Surface Temperature in Southern Ontario and Quebec,Canada

We downscaled atmospheric reanalysis data using linear regression and Bayesian neural network (BNN) ensembles to obtain daily maximum and minimum temperatures at ten weather stations in southern Quebec and Ontario, Canada. Performance of the linear and non-linear downscaling models was evaluated using four different sets of predictors, not only in terms of their ability to reproduce the magnitude of day-to-day variability (i.e., “weather,” mean absolute error between the daily values of the predictand(s) and the downscaled data) but also in terms of their ability to reproduce longer time scale variability (i.e., “climate,” indices of agreement between the predictand's observed annual climate indices and the corresponding downscaled values). The climate indices used were the 90th percentile of the daily maximum temperature, 10th percentile of the daily minimum temperature, number of frost days, heat wave duration, growing season length, and intra-annual temperature range.

Our results show that the non-linear models usually outperform their linear counterparts in the magnitude of daily variability and, to a greater extent, in annual climate variability. In particular, the best model simulating weather and climate was a BNN ensemble using stepwise selection from 20 reanalysis predictors, followed by a BNN ensemble using the three leading principal components from the aforementioned predictors. Finally, we showed that, on average, the first three indices presented higher skills than the growing season length, number of frost days, and the heat wave duration.     

1960~2010年西北干旱区极端降水特征研究

基于中国西北干旱区1960~2010 年70个台站逐日降水资料,采用欧盟组织(STARDEX)定义的降水极值指数,分析结果表明: ① 研究区近51 a来强降水发生频次、湿期平均长度表现出增多和增长趋势,干日数和干期平均长度表现出减少和变短趋势; ② 单次强水的强度在增加,表现为：湿日数减少,而降水的总量却显著增加。③ 湿日数减少主要是0~6 mm强度的降水日数减少,12~24 mm强度的降水日数显著增加,后者对降水总量的增加贡献较大。④ 绝大部分站点强降水（>12 mm）的雨日（量）都以上升趋势为主,表现为下降趋势的站点极少。