Doctoral Dissertation Research: Parameter and Metric Space Investigation: Towards "Honesty in Modeling"

2000年，一群科学家和研究人员聚集在加拿大班夫，参加第四届国际地理信息系统与环境建模会议，就如何消除模型与地理信息系统集成方面取得进一步进展的一些主要障碍发表了一份简短但有力的声明。在“班夫声明”中总结了会议讨论的主题的七项建议中包括：(3)模型应具有元数据，应评估通过模型传播误差的情况，并将其传达给模型预测的用户；(6)“诚实建模”[应考虑到]预测中的不确定性和误差。这一博士论文研究项目将通过调查广泛使用的城市元胞自动机模型SLEUTH中的误差传播来源来寻求在地理建模中推进更大的真实性。这位博士生认为，在SLEUTH的参数空间中，以及由此延伸到所有细胞自动机模型中的某些区域，参数交互可能会产生不寻常和不稳定的行为，将错误传播到模型输出中。此外，学生希望证明，在模型的参数空间内，有多组令人满意的城市系统描述参数，无论使用哪种适配度来比较模型的性能与控制数据。虽然城市自动机建模的前期工作主要集中在模型的应用上，但最近的研究已开始将模型校准视为建模过程中的关键阶段。先前的研究将更进一步，检查一个模型的“螺母和螺栓”，并确定参数相互作用是否会导致模型误差的传播。在理解这些广泛使用的模型如何在内部运行和行为方面，该方法是向前迈出的重要一步。侦探城市增长模型的参数空间将通过一系列实验来检验，这些实验使用基本的几何、理论和现实世界的城市数据。侦探模型将通过将参数空间解析成100个单位范围中的5个块来进行详尽和反复的重新校准，从而产生4,084,101个初始参数集。这些重复校准将被用来使用模型中当前使用的14个拟合度量来寻找最符合数据的度量。对于模型的新度量m，将增加参数空间的元胞无序。该度量将计算整个参数空间的稳定性，并将允许识别具有不一致行为的区域。然后，将使用自组织映射图对三个数据集的穷举校准的度量结果进行分析和可视化。这将允许确定用于匹配测量和模型行为的度量之间的联系。使用三个不同的数据集提供了一定程度的置信度，即结果不是数据的残差，而是模型行为的残差。该项目将针对模型参数值的极大组合校准模型。这类模型在过去十年中引起了人们的极大兴趣，希望可以用细胞自动机模型来模拟这种空间特征和土地利用或土地覆盖的变化，并复制现实。许多学者对这些模型持怀疑态度，认为任何空间变化模式都可以复制，只要模型有足够数量的参数，这些参数的值已经用当前地理计算模型的能力进行了校准。这个项目通过演示参数空间是否合理稳定以及参数值是否合理一致来处理这些问题，即使在给定完全不同的初始状态配置的情况下。这项工作的更广泛的影响包括：开发了一种新的方法来研究在城市自动机模型框架内发生的参数空间和相互作用。作为博士论文研究改进奖，该奖项还将提供支持，使有前途的学生建立一个强大的独立研究生涯。

A group of scientists and researchers who met in Banff, Canada, in 2000 for the 4th International Conference on Integrating GIS and Environmental Modeling issued a brief but powerful statement regarding how to eliminate some of the major barriers to further progress in integrating modeling and GIS. Among the seven recommendations that appeared in "the Banff Statement" that summarized topics discussed at the meeting were: (3) Models should have metadata, and the propagation of error through the model should be assessed and communicated to the users of model predictions; and (6) "Honesty in Modeling" [should account] for uncertainty and error in predictions. This doctoral dissertation research project will seek to advance greater honesty in geographic modeling by investigating the sources of error propagation in a widely used urban cellular automata model, SLEUTH. The doctoral student argues that there are areas within the parameter space of SLEUTH and, by extension, all cellular automata models, where parameter interactions can create unusual and unstable behavior that propagates error into model outputs. Additionally the student expects to demonstrate that within the parameter space of a model, there are multiple sets of satisfactory urban system description parameters, no matter which measure of fit are used to compare the model's performance to control data. While prior work in urban automata modeling has been focused on the application of models, recent studies have begun to look at the model calibration as a critical stage in the modeling process. Prior research will be taken a step further by examining the "nuts and bolts" of one model and by determining if the parameter interactions can lead to the propagation of model error. The approach is a significant step forward in understanding how these widely used models internally function and behave. The parameter space of the SLEUTH urban growth model will be examined through a series of experiments with basic geometric, theoretical, and real-world urban data. The SLEUTH model will be exhaustively and repeatedly recalibrated by parsing the parameter space into blocks of 5 out of the range of 100 units, resulting in 4,084,101 initial parameter sets. These repeat calibrations will be used to find those that best fit the data using the fourteen metrics of fit that are currently used in the model. A new measure for the model, m, the cellular disorder of the parameter space will be added. This metric will calculate the stability throughout the parameter space and will allow for the recognition of areas that have inconsistent behavior. Metric results from the exhaustive calibration of the three datasets will then be analyzed and visualized using self-organizing maps. This will allow for the determination of links between the metrics used to measures of fit and model behavior. Using the three different datasets provides some degree of confidence that the results are not residuals of the data, but of the model's behavior.The project will calibrate the model for an extremely large combination of the model parameter values. Such models have captured major interest in the past ten years in the hope that such spatial characteristics and land-use or land-cover change might be simulated and reality replicated with cellular automata models. Many scholars view these models with a great deal of skepticism, believing that any spatial pattern of change can be replicated given a model with a sufficient number of parameters whose values have been calibrated with the power of current geocomputational models. This project deals with these concerns by demonstrating whether the parameter space is reasonably stable and if the parameter values are reasonably consistent even given quite different initial state configurations. The broader impacts of this work include: the development of a new method for investigating the parameter space and interactions taking place within the framework of urban automata models. As a Doctoral Dissertation Research Improvement award, this award also will provide support to enable a promising student to establish a strong independent research career.