一种基于神经网络的气象要素插值方法与分析

为解决三维气象要素场的插值问题,利用神经网络的多维非线性映射功能,结合气象要素的特点,提出了一种方便灵活、精度好的插值方法.并利用NCEP再分析气象要素格点资料,对插值模型在空间域的插值精度进行了分析,对其中的一些数据现象,根据气象要素的分布特性,用信息量的观点做出了解释.得到了插值精度受迭代次数和样本点选取方法影响,不同气象要素由于空间变化率的不同而具有不同的最佳训练次数和最佳训练点选取方法的结论.根据分析结论给出了提高插值精度的方法.

Method and analysis of meteorological record interpolation based on neural network

To solve the meteo rolog ical r ecord interpo lativ e pro blem, a conv enient and precise interpo lation model w as developed based o n the non-liner r ef lected ability of neural netw ork and the feature of the meteorolo gical reco rd. T he spat ial interpolat io n precisio n w as analy zed w ith the reanalyzed NCEP meteoro logical grid reco rds. According to the characteristics of the meteo rolog ical record dist ribut ion, some ex planatio ns of data phenomena w ere g iven. The result show s that iterative t imes and the select io n of t raining sample influence inter polat ion precision and that differ ent meteoro logical elements hav e dif ferent best iterativ e t ime and best of t raining sample select ion. Some methods to enhance the interpolat ion pr ecision based on the result s w ere prov ided.