Abstract

Introduction:



Precipitation, as an essential process in the hydrological cycle, plays a vital role in maintaining the balance between fresh and saline water resources globally. It is one of the most important components in hydrology and climatology, directly influencing the hydrological cycle. Accurate precipitation forecasting is essential for properly estimating the cost of water, planning, managing, maintaining, and storing water under adverse weather conditions and droughts, as well as designing flood warning systems. In recent years, researchers have employed models based on artificial intelligence approaches due to the nonlinear and complex nature of hydrological issues. These models are inspired by the characteristics of living organisms and are capable of solving problems that are highly complex and extensive.According to conducted studies, the Support Vector Regression (SVR) model is an efficient tool for estimating precipitation and hydrological issues. Today, to enhance the efficiency and performance of the Support Vector Regression model, combining this model with metaheuristic algorithms is considered a suitable solution for precipitation forecasting. In this research, hybrid models of Support Vector Regression-Particle Swarm Optimization and Support Vector Regression-Whale Optimization were used to estimate precipitation in the coastal areas of the Caspian Sea.

Methodology:



The temperate and humid climate along the southern shores of the Caspian Sea, which is situated as a strip between the Alborz mountain range and the Caspian Sea, is predominantly comprised of low-lying plains. Generally, this area constitutes the smallest climatic zone in Iran and is divided into two regions: the lowland and mountainous areas. The Caspian Sea is located between the longitudes of 38 degrees 46 minutes west and 34 degrees 54 minutes east and the latitudes of 34 degrees 36 minutes south and 33 degrees 47 minutes north, in northern Iran.

The Babolsar synoptic station, located on the shores of the Caspian Sea, is one of the most important meteorological stations in northern Iran. This area, designated as a special coastal region, affects economic investment, aquaculture production, significant commercial ports, shipping, fishing, and its unique geographical location impacts tourism, provincial land use planning, and even national considerations. Therefore, analyzing and examining daily precipitation is essential and necessary.

On the other hand, although Support Vector Regression (SVR) models are widely used for estimating precipitation, research comparing Particle Swarm Optimization and Whale Optimization algorithms in this coastal region has not yet been conducted. Consequently, this study employed optimization algorithms in conjunction with the Support Vector Regression model to estimate precipitation in the coastal areas of the Caspian Sea.



Results and Discussion:



In this study, Support Vector Regression (SVR), combined with wavelet algorithms, Particle Swarm Optimization, and Whale Optimization, was used to model daily precipitation in the coastal areas of the Caspian Sea. The input parameters included relative humidity (RH), maximum temperature (T.max), minimum temperature (T.min), wind speed (WV), and sunshine hours (SSH), while the output parameter was precipitation §. This analysis was conducted over a daily time period from 2013 to 2024 for the Babolsar synoptic station.

The results indicated that hybrid models in the combined scenario, which included all the input parameters, had lower error rates compared to other scenarios. Therefore, increasing the number of effective parameters in hybrid models based on Support Vector Regression led to improved model performance. Additionally, all models that utilized the radial basis kernel function exhibited better accuracy. Furthermore, the SVR-wavelet model demonstrated superior performance compared to the other models examined.

Conclusions:

Estimation of precipitation using hybrid models based on Support Vector Regression (SVR) serves as an efficient tool in the design of climatological and meteorological systems. In the present study, a case study was conducted to evaluate the performance of the hybrid optimization model of SVR for estimating precipitation in the coastal areas of the Caspian Sea, specifically at the Babolsar synoptic station located in Mazandaran Province.

The results of the research, based on the evaluation of scenarios consisting of input parameters, demonstrated that in all examined models, increasing the number of effective parameters leads to better performance in precipitation estimation. Furthermore, the outcomes from the evaluation criteria revealed that the SVR-wavelet model exhibited high accuracy with minimal error. Additionally, according to the examined graphs, the SVR-wavelet model provided precipitation estimates close to their actual values, as evidenced by the Taylor diagram.

In summary, the findings of this research indicate that the use of artificial intelligence models based on the Support Vector Regression approach can be beneficial for estimating precipitation in other regions of the country and could serve as a step towards making appropriate management decisions.