Parameter calibration for process-driven hydrological models has long predominantly relied on traditional optimization algorithms such as Genetic Algorithms

while relatively fewer previous studies focused on parameter optimization based on the gradient descent methods. This study aims to examine the applicability of gradient descent algorithms in this field and compare their performance systematically against traditional optimization algorithms. We calibrate the parameters of the Hydrologiska Byr?ns Vattenbalansavdelning (HBV) model using six optimization methods－three gradient descent (GD) algorithms of Adam

AMSGrad

and Adadelta

and three traditional optimization algorithms of Covariance Matrix Adaptation Evolution Strategy (CMA-ES)

Adaptive Simulated Annealing (ASA)

and Genetic Algorithm (GA). The results indicate the GD algorithms are better in computational efficiency and simulation stability. They raise runoff fitting accuracy or Nash-Sutcliffe Efficiency (NSE) by roughly 0.01-0.02 compared to traditional algorithms

and reduce Top Peak Error (TPE) by up to 23%. And

they can explore adaptively beyond initial parameter constraints－different from the traditional optimization algorithms that heavily rely on predefined parameter ranges－so that they are effective in guiding parameters toward a more physically reasonable space and significantly reducing dependence on parameter specification. This study has achieved an effective approach for hydrological model parameter optimization

useful for further theoretical or practical studies.