摘要
贝叶斯最大熵方法可以对具有一定不确定性的"软数据"和认为没有误差的"硬数据"进行插值。对甘肃省1961—1990年52个气象站点的多年平均降水数据进行空间化研究。通过比较普通克里格、共协克里格、三元回归建模后残差插值以及基于贝叶斯最大熵的3种不同软硬数据参与情况下的插值结果,发现考虑降水30a时间序列不完整性以及辅助变量经验模型不确定性的插值结果的MAE和RMSE,比直接使用多年平均降水数据直接插值的MAE和RMSE小,表明贝叶斯最大熵方法通过对不确定性的考虑可以有效降低预测结果的绝对误差。从降水的空间分布来看,考虑辅助变量DEM的插值结果能相对较好的体现高程对降水的地形影响,尤其分区将辅助变量转换为软数据可以有效体现不同区域高程对降水的不同影响问题。综合误差评价以及降水插值结果的空间分布,认为BME插值过程中可以考虑数据本身以及辅助数据利用的不确定性,使降水空间化的结果更加真实客观,同时为合理利用辅助信息提供了一个新思路。
Bayesian Maximum Entropy （BME） is a spatio-temporal mapping method which can use the soft data with uncertainty and accurate hard data to perform the spatial interpolation. Spatial interpolation of multi-yearly average precipitation was conducted in Gansu Province based on precipitation data from 52 meteorological stations of Gansu Province during 1961-1990 and BME and traditional interpolation methods. Interpolation accuracy of ordinary Kriging interpolation, Cokriging interpolation, residual Kriging interpola- tion after triple regression modeling and three BME interpolation methods incorporating different soft and hard data were compared. Results showed that MAE and RMSE values of interpolations with soft data incor- poration were smaller than those with hard data only. This result indicated BME could effectively reduce the absolute error by taking account of data uncertainty from missing records and relationship model between the interest variable and secondary variable. Based on the fact that the altitude played different roles in different regions, so the DEM was converted into different soft data in different sub regions and the interpolation results showed the effect of elevation on precipitation was better. We can see that BME can perform precipitation interpolation objectively from the error evaluation and spatial distribution of precipitation interpolation results, and provide a new way to integrate the secondary information.
出处
《中国沙漠》
CSCD
北大核心
2012年第5期1408-1416,共9页
Journal of Desert Research
基金
国家基础研究计划"973"项目(2007CB714402)
传染病多维信息集成分析与传播风险预测技术研究项目(2008ZX10004-012)
国家自然科学基金项目(40871161)
遥感科学国家重点实验室自主研究项目资助
关键词
贝叶斯最大熵
地统计学
不确定性
软数据
降水
Bayesian Maximum Entropy
geostatistics
uncertainty
soft data
precipitation
