摘要
It is well known that sufficiently long and continuous streamflow data are required for accurate estimations and informed decisions in water-resources planning, design, and management. Although streamflow data are measured and available at most river basins, streamflow records often suffer from insufficient length or missing data. In this work, artificial neural networks (ANNs) are applied to extend daily streamflow records at Lilin station located in Gaoping River basin, southern Taiwan. Two ANNs, including feed forward back propagation (FFBP) and radial basis function (RBF) networks, associated with various time-lagged streamflow and rainfall inputs of nearby long-record stations are employed to extend short daily streamflow records. Performances of ANNs are evaluated by root-mean-square error (RMSE), coefficient of efficiency (CE), and histogram-matching dissimilarity (HMD). Inconsistency among these evaluation measures is solved by the technique for order performance by similarity to ideal solution (TOPSIS), a widely used multi-criteria decision-making approach, to find an optimal model. The results indicate that RBF-E1 (entire-year data training with Q t and Q t−1 inputs) has the minimum RMSE of 104.4 m3/s, second highest CE of 0.956, and third lowest HMD of 0.0096, which outperforms other ANNs and provide the most accurate reconstruction of daily streamflow records at Lilin station. Keywords Data extension Artificial neural networks Performance criteria Multi-criteria decision-making