Prediction of flow duration curves for ungauged basins
Prediction of flow duration curves for ungauged basins
复制标题
DOI：
10.1016/j.jhydrol.2016.12.048
复制
发表时间：
2017-02-01
影响因子：
6.4
通讯作者：
Gharabaghi, Bahram
中科院分区：
文献类型：
--
作者：
Atieh, Maya;Taylor, Graham;Gharabaghi, Bahram
本研究提出了新的模型，利用人工神经网络（ANN）和基因表达编程（GEP）对北美171个非管制流域和89个管制流域的历史流量记录进行训练和测试，预测非管制流域的流量持续曲线（fdc）。对于89个调节流域，在流量调节前后都产生了fdc。地形、气候和土地利用特征被用来发展这些流域特征与FDC统计分布参数之间的关系：平均值(m)和方差(v)。两个主要假设，即流量调节对平均值(m)的影响可以忽略不计，而方差(v)得到证实。预测平均值（GEP-m）的新GEP模型具有较高的R-2值（0.9）和D值（0.95），RAE值较低，为0.25。预测方差的简单回归模型（REG-v）是均值(m)和流量调节指数(R)的函数。实测性能和不确定度分析表明，ANN-m模型的R-2值为0.97，RAE值为0.21，D值为0.93,95%置信区间最小（+0.22 ~ +3.49）。GEP和ANN模型对流域面积最敏感，其次是年平均降水量、分配熵失序指数和形状因子。(C) 2016 Elsevier B.V.版权所有
This study presents novel models for prediction of flow Duration Curves (FDCs) at ungauged basins using artificial neural networks (ANN) and Gene Expression Programming (GEP) trained and tested using historical flow records from 171 unregulated and 89 regulated basins across North America. For the 89 regulated basins, FDCs were generated for both before and after flow regulation. Topographic, climatic, and land use characteristics are used to develop relationships between these basin characteristics and FDC statistical distribution parameters: mean (m) and variance (v). The two main hypotheses that flow regulation has negligible effect on the mean (m) while it the variance (v) were confirmed. The novel GEP model that predicts the mean (GEP-m) performed very well with high R-2 (0.9) and D (0.95) values and low RAE value of 0.25. The simple regression model that predicts the variance (REG-v) was developed as a function of the mean (m) and a flow regulation index (R). The measured performance and uncertainty analysis indicated that the ANN-m was the best performing model with R-2 (0.97), RAE (0.21), D (0.93) and the lowest 95% confidence prediction error interval (+0.22 to +3.49). Both GEP and ANN models were most sensitive to drainage area followed by mean annual precipitation, apportionment entropy disorder index, and shape factor. (C) 2016 Elsevier B.V. All rights reserved.