基于时变特征矩阵的相似流域识别方法研究

Research on Similar Basin Recognition Method Based on Time-varying Feature Matrix

为了提升无径流资料区水文预报相似流域识别的可靠性,提出了一种相似流域识别方法,该方法通过引入流域时变特征信息,构建流域时变特征矩阵及其展布方法获取流域特征向量,并利用经典K-means聚类方法实现相似流域识别,并以江西省17个天然闭合流域为例,结合新安江模型,开展特征向量/矩阵构建、相似流域聚类识别、模型参数移用、水文预报结果对比分析等一系列研究。结果表明,提出的时变特征矩阵用于无径流资料区水文预报能取得较好的应用效果,与传统的仅基于时不变下垫面信息的相似流域识别方法相比,预报精度得到提升;基于时变综合要素获得的相似分组取得了最高的预报精度,基于时变气象要素获得的相似分组预报精度优于仅基于时不变下垫面要素获得的相似分组;与基于单一要素相似流域识别方法相比,基于综合要素获得的相似分组改善了参数移用的方向性问题,但尚不能完全消除方向性影响。

A similar basin identification method was proposed to improve the reliability of similar basin identification for hydrological forecasting in ungauged basin.The method introduced the basin time-varying feature information and constructed a time-varying feature matrix(TVFM).A dimension reduction method for the TVFM was proposed to ob-tain the basin feature vector.The classical K-means clustering method was adopted to implement the identification of the similar basins.Seventeen natural catchments in Jiangxi Province were selected and the Xin'anjiang model was applied in this research.A series of studies were carried out which includes feature vector and feature matrix construction,similar basin identification,parameter transfer,and comparison&analysis of hydrological forecasting results.The following conclusions can be drawn.The proposed TVFM can achieve better forecasting results in ungauged basins.Compared with the traditional similar basin identification method based on time-invariant underlying surface information,the prediction accuracy is improved.The similar basin identification based on time-varying comprehensive features generates the best predictions,and the prediction accuracy based on time-varying meteorological features is better than that based only on time-invariant underlying surface features.Comparison with the single factor methods,the directional problem of the pa-rameter transfer can be partially solved by the proposed comprehensive method.Unfortunately,the directional problem cannot be completely overcame until recently.