Development of Geomorphological Artificial Neural Networks (GANNs) for Modeling Watershed Runoff

9524758 Govindaraju这项建议涉及开发一个地貌人工神经网络(GANN)模型，用于预测流域的非线性降雨-径流关系。Gann代表了一个由大量相互连接的处理单元组成的网络，称为神经元。该网络的特点是权重决定了每个连接的强度，神经元的阈值决定了它们的激活水平。将从分水岭排出水的陆上水流平面、溪流和地下水流路径网络映射到人工神经网络(ANN)的拓扑结构中。该模型将利用分水岭地貌表示的优势，并将能够捕捉分水岭对降水事件的非线性响应。Gann模型将被进一步用来研究时间和空间尺度的影响，并调查各种地表和地下流动路径的意义。江恩模型的性能将通过几种方式进行测试。根据降雨、土壤、温度、地形和其他控制流域内水分运动的变量的数据的大小和可用性，堪萨斯州已确定了四个流域。该网络将使用部分数据进行训练。同时，将使用相同的历史记录来校准基于物理的模型和基于回归的经验模型。然后，这些模型将在预测模式下通过降雨径流记录的剩余部分进行测试。这不仅有助于评估Gann模型在给定降雨、太阳辐射和温度记录时对径流的预测效果，而且还将形成与其他类型模型的性能比较的基础。***

9524758 Govindaraju This proposal deals with the development of a geomorphological artificial neural network (GANN) model for predicting the nonlinear rainfall-runoff relationships of watersheds. GANN represents a massively interconnected network of processing units called neurons. The network is characterized by weights, which determine the strength of each connection, and the threshold values of neurons which dictate their activation level. The network of overland flow planes, streams and subsurface flow pathways, which drains the water from the watershed, will be mapped into the topology of artificial neural networks (ANNs). This model will draw on strengths of the geomorphological representations of the watershed, and will also be capable of capturing the nonlinear response of watersheds to precipitation events. GANN models will be further utilized to study the influence of temporal and spatial scales, and to investigate the significance of various surface and subsurface flow pathways. The performance of GANN models will be tested in several ways. Four watersheds in Kansas have been identified based on size and availability of data for rainfall, soils, temperature, topography, and other variables which govern water movement within the watershed. The network will be trained using a portion of the data. Simultaneously, a physically-based model and a regression-based empirical model will be calibrated using the same historical record. These models will then be tested in prediction mode through the remaining portion of the rainfall-stream flow records. This will not only help evaluate the GANN model in terms of how well it predicts stream flow given rainfall, solar radiation, and temperature records, but will also form a basis for comparing its performance with respect to other types of models. ***