On the validity of first-order prediction limits for conceptual hydrologic models
On the validity of first-order prediction limits for conceptual hydrologic models
复制标题
DOI：
10.1016/0022-1694(88)90136-9
复制
发表时间：
1988-11
影响因子：
6.4
通讯作者：
G. Kuczera
中科院分区：
文献类型：
--
作者：
G. Kuczera
一阶分析是评价参数不确定性通过概念水文模型传播的影响的有力方法。然而，它的有效性取决于一个强有力的假设，即一阶近似在有显著参数不确定性的参数空间区域是有效的。建议采用Beale的非线性测度来检验这一假设。该测量是基于从90%置信度椭球表面随机抽样的参数的实际响应与线性化响应之间的差异。涉及两个非线性概念模型的例子表明，模型非线性非常依赖于应用，突出了在所有模型应用中计算Beale非线性度量的必要性。水文响应的不确定性不仅是由模型中传播的参数不确定性引起的，而且是由模型和测量误差引起的自然不确定性引起的。基于参数和自然不确定性的近似预测极限，在回归环境中开发，它采用与概念水文模型应用中发现的残差特征一致的误差模型。一个涉及八个参数的水流产量模型的例子表明，自然不确定性优于参数不确定性，强调在计算预测极限时需要包括两种形式的不确定性。
First-order analysis is a powerful method for evaluating the effect of parameter uncertainty propagating through a conceptual hydrologic model. However, its validity rests on the strong assumption that a first-order approximation is valid over the region of parameter space where there is significant parameter uncertainty. It is suggested that Beale's nonlinearity measure be used to check this assumption. This measure is based on the discrepancy between actual and linearized response for parameters randomly sampled from the surface of the 90% confidence ellipsoid. Examples involving two nonlinear conceptual models demonstrate that model nonlinearity is very much application-dependent, highlighting the need to compute Beale's nonlinearity measure in all model applications. Uncertainty in hydrologic response is induced not only by parameter uncertainty propagating through the model, but also by natural uncertainty arising from model and measurement error. Approximate prediction limits based on both parameter and natural uncertainty, are developed in a regression context, which employs an error model consistent with the residual characteristics found in conceptual hydrologic model applications. An example involving an eight-parameter streamflow yield model demonstrates dominance of natural over parameter uncertainty, emphasizing the need to include both forms of uncertainty when computing prediction limits.