Calibration of Hydrologic Models Using Multiobjectives and Visualization Techiques

9418147 Sorooshian随着水文学家和气候学家研究新出现的问题和开发新的数据来源，对陆地表面水文模型作为水文气候学研究工具的依赖正在增加。然而，水文模型只能与模型假设、输入和参数估计一样可靠。现场测量、先验信息和校准是参数估计中使用的三种技术。虽然使用实地测量越来越重要，但经验表明，作为一个实际问题，几乎所有模型都需要至少校准某些参数。然而，大量的模型校准经验表明，通常不可能找到首选（最佳）解决方案。也就是说，对于给定的模型和校准数据，通常有相当大的参数空间区域似乎给出大致等效的结果。这可能会引起对模型的可靠性和现实性以及对其预测的信心的质疑。这项研究的主要目标是开发校准水文模型的技术，以改善寻找优选解决方案的前景。主要目标是：1。认识到水文模型校准问题固有的多目标性质，并在多目标框架下提出校准程序。2. 探索使用多标准、数据子集（强调不同水文过程或模型性能的不同方面）、信息内容度量和全局搜索算法来识别非劣解空间和优选解的创新方法。3. 在测试用例上，确定是否可以识别非劣解空间的一个令人满意且可靠的估计，从中可以选择一个首选的参数估计（或估计集）。考虑到非劣解空间和数据误差，开发估算参数和模拟变量置信区间的技术。深入了解数据错误、模型错误和参数交互在产生非劣解中所起的作用。4. 对于测试用例，确定如何通过系统的基于非劣解空间的计算机探索可靠、有效和高效地获得首选解。实现了一个系统的基于工作站的校准策略，并与现有的单目标策略进行了比较。本研究将重点关注被认为是可行的陆地表面土壤-植被-大气转移方案（SVATS）的新一代陆地表面水文模型；这些模型受到了水文学家和水文气象学家的广泛关注，但围绕它们的校准问题尚未得到系统的关注。预计所提出的技术将帮助水文学家建立更可靠和更现实的模型。在一般科学知识方面，预计该研究将导致对水文模型识别可校准问题的见解，特别是在可靠性和唯一性方面。

9418147 Sorooshian The reliance on land surface hydrologic models as tools in the study of hydroclimatology is increasing as hydrologists and climatologists examine emerging problems and exploit new data sources. However, hydrologic models can only be as reliable as model assumptions, inputs, and parameter estimates. Field measurements, prior information, and calibration are three techniques used in parameter estimation. While the use of field measurements is gaining in importance, experience has shown that as a practical matter, virtually all models require calibration of at least some parameters. However, extensive experience with model calibration has indicated that it is very often not possible to find a preferred (best) solution. That is, for a given model and calibration data, there are usually sizable regions of the parameter space that appear to give roughly equivalent results. This can gives rise to questions about the reliability and realism of the model, and the confidence in its predictions. The primary objective of this research is to develop techniques for calibrating hydrologic models that improve the prospects for finding preferred solutions. Major objectives are to: 1. Recognize the inherently multi-objective nature of the hydrologic model calibration problem and pose the calibration procedure in the framework of multiple objectives. 2. Explore innovative ways of using multi-criteria, data sub-sets (that emphasize different hydrologic processes or different aspect of model performance), measures of information content and global search algorithms in identifying the non-inferior solution space and preferred solutions. 3. On test cases, determine if a satisfactory and reliable estimate of the non-inferior solution space can be identified, from which a preferred parameter estimate (or set of estimates) can be selected. Develop techniques for estimating confidence intervals on paramete rs and simulated variable, taking into account the non-inferior solution space and data errors. Gain insights into the roles that data error, model error, and parameter interaction play in producing non-inferior solutions. 4. Determine, for test cases, how preferred solutions can be reliably, effectively and efficiently obtained through a systematic computer based exploration of the non-inferior solution space. Implement a systematic workstation based calibration stratigy and compare it with existing single-objective strategies. This research will focus its attention on the emerging generation of land surface hydrologic models being considered as viable land-surface soil-vegetation-atmosphere transfer schemes (SVATS); such models are receiving widespread attention among hydrologists and hydrolmeteorologists but the calibration issues surrounding them have yet to receive systematic attention. It is expected that the proposed techniques will help hydrologists build more reliable and more realistic models. In terms of general scientific knowledge, it is expected that the research will result in insights into problems of hydrologic model identification can calibration, particularly in the area of reliability and uniqueness.