A review of parallel computing applications in calibrating watershed hydrologic models

In recent decades, parallel computing has been increasingly applied to address the computational challenges of calibrating watershed hydrologic models. The purpose of this paper is to review these parallelization studies to summarize their contributions, identify knowledge gaps, and propose future research directions. These studies parallelized models based on either random-sampling-based algorithms or optimization algorithms and demonstrated considerable parallel speedup gain and parallel efficiency. However, the speedup gain/efficiency decreases as the number of parallel processing units increases, particularly after a threshold. In future, various combinations of hydrologic models, optimization algorithms, parallelization strategies, parallelization architectures, and communication modes need to be implemented to systematically evaluate a suite of parallelization scenarios for improving speedup gain, efficiency, and solution quality. A standardized suite of performance evaluation metrics needs to be developed to evaluate these parallelization approaches. Interactive multiobjective optimization algorithms and/or integrated sensitivity analysis and calibration algorithms are potential future research fields, as well.

Automated summary

This paper reviews the application of parallel computing in calibrating watershed hydrologic models and summarizes their contributions, knowledge gaps, and future research directions. The studies parallelized models using random-sampling-based algorithms or optimization algorithms and achieved significant speedup gain and efficiency. However, the speedup gain and efficiency decrease as the number of parallel processing units increases, especially after a certain threshold. Various combinations of hydrologic models, optimization algorithms, parallelization strategies, architectures, and communication modes need to be explored to improve speedup gain, efficiency, and solution quality. A standardized set of performance evaluation metrics should be developed to assess parallelization approaches. Interactive multiobjective optimization algorithms and integrated sensitivity analysis and calibration algorithms can also be potential future research areas.