当前小区域的古气候变化研究受模拟资料分辨率和可靠性的严重制约。为了将大区域的气候模拟资料应用到小区域的古气候研究中去,亟待建立有效的降尺度方法。为此以徽鄂地区为例,建立了一个3层BP神经网络拟合模型,利用相关气象要素作为拟合因子,拟合并重建了该地区近千年来1月、7月和年平均的温度和降水序列,通过与观测及模拟资料的对比分析发现,该模型拟合及重建的近千年气候序列有较高的精度和可靠性,能反映小区域气候的年际和年代际变化信号,提高了模拟资料对小区域气候变化的刻画能力。
Researches on the paleoclimatic variabilities in local regions are seriously restricted by the low resolution and uncertainty of the simulated data at present. In order to apply large-scale climate modeling data to paleoclimate research for small region, an effective downscaling method is urgently needed to be built. For this purpose, a triple-layer fitting model of back propagation (BP) neural network was established using relevant meteorological variables as fitting factors. Based on the fitting model, monthly (January and July) and annual mean series of tem- perature and precipitation were reconstructed in Anhui-Hubei region during the last millennium. Comparison of the fitting series with the observed and simulated records indicates that the fitting series have good precision and relia- bility. The signals of climate variation in local region on interannual and interdecadal time scales were captured successfully by the BP neural network model. The results show that this downscaling method improves the capacity of research on paleoclimate variability in local regions using large-scale modeling data.