Journal of Earth System Science ( IF 1.3 ) Pub Date : 2021-05-26 , DOI: 10.1007/s12040-021-01612-3 K P Rajindas , A P Shashikala
Abstract
A major portion of the coastline of Kerala is under erosion, primarily due to the action of wind-generated waves. Accurate assessment of the nearshore wave climate is essential for detailed apprehension of the sediment processes that lead to coastal erosion. Numerical wave transformation models set up incorporating high-resolution nearshore bathymetry and nearshore wind data, prove to be sufficient for the purpose. But, running these models for decadal time scales incur huge computational cost. Thus, a Feed Forward Back Propagation ANN is developed to estimate the wave parameters nearshore with training datasets obtained from minimal set of numerical simulations of wave transformation using DELFT3D-WAVE. The numerical model results are validated using Wave Rider Buoy data available for the location. This hybrid methodology is utilized to hindcast nearshore wave climate of a location in north Kerala for a period of 40 years with the ANN model trained with 1-yr data. The model shows good generalization ability when compared to the results of numerical simulation for a period of 10 years. This paper illustrates the data and methodology adopted for the development of the numerical model and the proposed ANN model along with the statistical comparisons of the results obtained.
Research highlights
-
A hybrid methodology, combining numerical modelling and soft computation using ANNs, is developed to obtain long-term nearshore wave hindcast. One years’ numerical model simulation is utilised to train the ANN models.
-
The optimised ANNH, ANNT, ANNθmx and ANNθmy models, with 15, 25, 25 and 30 neurons respectively in their single hidden layer, show good generalization ability when compared to the results of numerical simulation for a period of 10 years. The coefficient of correlation between the numerical model results and the ANNH model is 0.99. Results of ANNT model and the combined result of ANNθmx, ANNθmy models show a coefficient of correlation of 0.97 with the corresponding numerical model results. The new methodology allows for faster reconstruction of long-term time series of nearshore wave parameters.
-
The trained models are used for simulating nearshore wave parameters at a location in North Kerala coast for 40 years. The maximum Hs at the nearshore location from 40 years’ ANN simulation is 3.39 m. Hs exceeds 3 m only for 0.04% of the time. During monsoon, waves feature a narrow range of Tp as well as mean wave direction as opposed to the non-monsoon period.
中文翻译：
混合波变换方法的发展及其在印度喀拉拉邦海岸的应用
摘要
喀拉拉邦海岸线的大部分地区受到侵蚀，这主要是由于风力产生的作用。准确评估近岸波浪气候对于详细了解导致海岸侵蚀的沉积过程至关重要。结合高分辨率近海测深法和近岸风数据建立的数值波变换模型被证明足以满足此目的。但是，将这些模型运行十年的时间尺度会导致巨大的计算成本。因此，开发了前馈传播ANN以使用从使用DELFT3D-WAVE进行波变换的数值模拟的最小集获得的训练数据集估计近岸的波参数。使用可用于该位置的Wave Rider Buoy数据验证了数值模型结果。该混合方法用于以一年数据训练的ANN模型对喀拉拉邦北部某地点的近岸海浪气候进行40年的后预报。与10年的数值模拟结果相比，该模型具有良好的泛化能力。本文说明了用于开发数值模型和拟议的ANN模型的数据和方法，并对获得的结果进行了统计比较。
研究重点
-
开发了一种混合方法，将数值建模和使用人工神经网络的软计算相结合，以获得长期的近岸波后兆。利用一年的数值模型仿真来训练ANN模型。
-
优化的ANN H，ANN T，ANNθmx和ANNθmy模型在其单个隐藏层中分别具有15、25、25和30个神经元，与10年的数值模拟结果相比，它们显示出良好的泛化能力。数值模型结果与ANN H模型之间的相关系数为0.99。ANN T模型的结果以及ANNθmx和ANNθmy模型的组合结果与相应的数值模型结果相关系数为0.97。新方法可以更快地重建近岸波参数的长期时间序列。
-
经过训练的模型用于模拟北喀拉拉邦海岸某处的近岸波浪参数长达40年。经过40年的ANN模拟，近岸位置的最大H s为3.39 m。H s仅在0.04％的时间内超过3 m。在季风，波浪设有T的窄范围的p以及相对于非季风期间平均波方向。
京公网安备 11010802027423号