JoH paper:基于并行计算和遥感产品的模型分布式校准策略研究

我们基于并行计算技术,提出了一种基于遥感产品和径流观测的模型分布式校准方法,并将该方法与四个传统的方法进行了比较。我们发现,新的方法不仅能提升模型的模拟精度,而且计算效率明显更高(传统方法校准需要几个小时,而我们只需要几十分钟)。论文地址: https://www.sciencedirect.com/science/article/abs/pii/S0022169421004066

我的方法选择SWAT模型进行了测试,代码地址如下,欢迎大家使用。
https://github.com/HydroRS/ParallelSWAT

Satellite-based products can provide valuable information for model calibrations and evaluations. However, effectively and efficiently constraining hydrological models with both satellite-based information and streamflow measurements remains a challenge. Here, a parallel computing-based and spatially stepwise strategy that enables separate optimizations of different objective functions in a spatially distributed and parallel manner was proposed for model calibration with streamflow observations and satellite-based evapotranspiration (ET). The new calibration strategy (M5) was tested with the Soil and Water Assessment Tool (SWAT) model in the upper Heihe River basin of China, a mountainous watershed on the northeastern Qinghai-Tibet Plateau. The performance of M5 was evaluated and compared with two single-variable calibration strategies, i.e., streamflow-only calibration and ET-only calibration, and with two multivariate calibration methods, i.e., joint calibration and stepwise calibration. Results indicate that M5 achieves the best model performance among the five calibration strategies in reproducing temporal variations of streamflow and ET. Moreover, M5 improves the simulation of the spatial pattern of ET and attains a higher spatial efficiency than the other calibration strategies. M5 also exhibits a higher computational efficiency, with a magnitude up to 2 times greater than the other calibration strategies, due to the application of parallel computing. Further analysis demonstrates that the new calibration strategy can lead to a synergic relationship between the simulation accuracy of streamflow and ET, underscoring the added value of satellite-based products for model calibrations.

Leave a Reply

  Subscribe  
Notify of