Abstract. Future global change is likely to give rise to novel combinations of the factors which enhance or inhibit soil erosion by water. Thus there is a need for erosion models, necessarily process-focused, which are able to reliably represent rates and extents of soil erosion under unprecedented circumstances. The process-focused cellular automaton erosion model RillGrow is, given initial soil surface microtopography on a plot-sized area, able to predict the emergent patterns produced by runoff and erosion. This study explores the use of Structure-from-Motion photogrammetry as a means to calibrate and validate this model by capturing detailed, time-lapsed data for soil surface height changes during erosion events. Temporally high-resolution monitoring capabilities (i.e. 3D models of elevation change at 0.1 Hz frequency) permit validation of erosion models in terms of the sequence of formation of erosional features. Here, multi-objective functions, using three different spatio-temporal averaging approaches, are assessed for their suitability in calibrating and evaluating the model's output. We used two sets of data, from field- and laboratory-based rainfall simulation experiments lasting 90 and 30 minutes, respectively. By integrating 10 different calibration metrics, the output of 2000 and 2400 RillGrow runs for the field and laboratory experiments respectively, were analysed. No single model run was able to adequately replicate all aspects of either field and laboratory experiments. The multi-objective approaches highlight different aspects of model performance, indicating that no single objective function can capture the full complexity of erosion processes. They also highlight different strengths and weaknesses of the model. Depending on the focus of the evaluation, an ensemble of objective functions may not always be necessary. These results underscore the need for more nuanced evaluation of erosion models, e.g. by incorporating spatial pattern comparison techniques to provide a deeper understanding of the model’s capabilities. Such evaluations are an essential complement to the development of erosion models which are able to forecast the impacts of future global change.

中文翻译：

---

使用高时空分辨率的 3D 观测来校准和评估以过程为中心的水土流失元胞自动机模型


摘要。未来的全球变化可能会产生增强或抑制水土流失的新因素组合。因此，需要有一定以过程为中心的侵蚀模型，能够可靠地表示前所未有的情况下土壤侵蚀的速率和程度。以过程为中心的元胞自动机侵蚀模型 RillGrow 在给定地块大小区域上的初始土壤表面微地形的情况下，能够预测径流和侵蚀产生的新兴模式。本研究探索使用运动结构摄影测量作为校准和验证该模型的方法，通过捕获侵蚀事件期间土壤表面高度变化的详细、延时数据。时间高分辨率监测功能（即 0.1 Hz 频率下的高程变化 3D 模型）允许根据侵蚀特征的形成顺序来验证侵蚀模型。在这里，使用三种不同的时空平均方法来评估多目标函数在校准和评估模型输出方面的适用性。我们使用了两组数据，分别来自持续 90 分钟和 30 分钟的现场降雨模拟实验和实验室降雨模拟实验。通过整合 10 种不同的校准指标，分别分析了现场和实验室实验的 2000 次和 2400 次 RillGrow 运行的输出。没有任何一个模型运行能够充分复制现场和实验室实验的所有方面。多目标方法突出了模型性能的不同方面，表明没有单一目标函数可以捕获侵蚀过程的全部复杂性。他们还强调了模型的不同优点和缺点。 根据评估的重点，目标函数的集合可能并不总是必要的。这些结果强调需要对侵蚀模型进行更细致的评估，例如通过结合空间模式比较技术来更深入地了解模型的功能。此类评估是对侵蚀模型开发的重要补充，侵蚀模型能够预测未来全球变化的影响。