基于克里金插值的SWAT参数率定大规模并行方法

SWAT（Soil and Water Assessment Tool）水文模型是近几年国内外使用最广泛的分布式水文模型，因其良好的精度、广泛的适应度得到了众多水文工作者的欢迎。但由于其参数率定部分过高的计算量，给水文工作者建模调参过程中造成了大量的不便。为了提高SWAT参数率定的运行效率，本文基于克里金插值理论，对于SWAT原有的优化算法进行改进，提出了一种基于随机过程的参数率定方法，通过一种“期望优化”的方式对拟合曲面进行寻优。并通过Intel MKL和一个对等式并行框架加以并行实现。大大提高了SWAT参数寻优的计算效率以及参数精度，并通过天山北坡流域的模拟验证了此并行模型的有效性，说明了克里金优化方法在水文模拟参数寻优的过程中能够达到良好的精度。

A Large Scale Parallel Method for SWAT Calibration based on Kriging Interpolation

SWAT (Soil and Water Assessment Tool) is the most popular distribute hydrologic model in the world because of its high accuracy and fitness. But the large amount of computation in SWAT calibration causes much difficulty to the hydrologists. To improve the performance of SWAT calibration, in this study, we proposed a new calibration method based on kriging interpolation, and the optimization is based on expected improvement. Also, we parallelized this process by the Intel MKL and a peer-to-peer parallel framework. Then the performance of optimization is improved, and we tested the effectiveness of this method by the rivers in the northern slope of Tianshan Mountains. This study drawn the conclusion that the kriging optimization can achieve high precision in the calibration of hydrologic simulation.