地球信息科学学报 ›› 2011, Vol. 13 ›› Issue (4): 534-538.doi: 10.3724/SP.J.1047.2011.00534

• 地理模型与模拟应用 • 上一篇    下一篇

多年平均气温空间化BP神经网络模型的模拟分析

张赛1,2, 廖顺宝1   

  1. 1. 中国科学院地理科学与资源研究所 资源与环境信息系统国家重点实验室,北京 100101;
    2. 中国科学院研究生院,北京 100049
  • 收稿日期:2010-02-19 修回日期:2011-07-15 出版日期:2011-08-25 发布日期:2011-08-23
  • 通讯作者: 廖顺宝(1966-),男,博士,副研究员,研究方向为遥感与GIS应用。E-mail:liaosb@igsnrr.ac.cn E-mail:liaosb@igsnrr.ac.cn
  • 基金资助:

    中科院"十一五"信息化专项——人地系统主题数据库建设与服务(INF0-115-C01-SDB3-02); 资源与环境信息系统国家重点实验室自主研究课题——地球科学数据质量评价研究(088RA106SA)。

Simulation and Analysis of Spatialization of Mean Annual Air Temperature Based on BP Neural Network

ZHANG Sai1,2, LIAO Shunbao1   

  1. 1. State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, CAS, Beijing 100101, China;
    2. Graduate University of Chinese Academy of Sciences, Beijing 100049, China
  • Received:2010-02-19 Revised:2011-07-15 Online:2011-08-25 Published:2011-08-23

摘要: 气温数据空间化是插补无站地区温度、使气温数据便于综合分析的重要技术手段。理想情况下,气温的空间化分布受经度、纬度和海拔高度的影响,呈现规律性的空间分布态势。但是,各种微观因子如坡度、坡向、地形起伏、地表覆被等的存在,在一定程度上扰乱并弱化了这种规律性的分布态势。本文基于Matlab平台,利用BP神经网络研究了多年平均气温数据空间化的新方法。结果表明,与传统的IDW插值、Kriging插值、样条插值和趋势面插值相比,BP神经网络的绝对误差仅为0.51℃,具有较高的空间化精度,同时它更加准确地反映了诸如阿尔泰山、天山、昆仑山、喜马拉雅山等山区低温带的气温分布规律。本研究不仅丰富了气温数据空间化的理论、技术和方法,为相关研究提供了重要的基础数据;而且也为降雨、蒸发等模型因果关系不十分明确的气候/气象要素的空间化提供了一定的参考和借鉴。

关键词: 多年平均气温, 空间化, ANN, BP神经网络

Abstract: Air temperature is one of the main influential factors of ecosystem. Ideally, air temperature is mainly affected by longitude, latitude, altitude and the distance from the ocean, so its spatial distribution should show a regular tendency. However, to some extent, the existence of various microcosmic topographical factors (such as slope, aspect, topographic relief,terrain shade land cover, etc.) disturbs its fundamental distribution tendency, even strongly in certain areas, and thus complicates the research and estimation on air temperature. Artificial Neural Network (ANN), which has adaptive capability, high-efficient computing power and powerful nonlinearity approach capability, can effectively improve prediction precision and have generalization capacity. BP neural network is one of the easily understood and most effective methods of ANN. By applying BP neural network and Matlab platform, the spatialization of mean annual air temperature was carried out in this paper, and the spatialization result was compared with those by previous researchers. The comparison confirmed the advantage of the spatialization method. The results from the comparison indicate that BP neural network has higher accuracy with a mean absolute error of 0.51℃ than other spatialization methods, including IDW, Kriging, Spline and Trend. Furthermore, the results of the spatialization are able to describe the distribution of low air temperature in mountain areas, such as the Altai, the Tianshan Mountains, the Kunlun Mountains and the Himalayas in more details. This study not only complements theories, technologies and methods of air temperature spatialization, but also provides an important data product for relevant researches. It also provides a reference to spatialization of other climate data, such as rainfall, evaporation, and so on.

Key words: mean annual air temperature, spatialization, ANN, BP neural network