پیش‌بینی آنومالی سطح دریا با استفاده از روش تجزیه به توابع حالت‌های ذاتی و شبکه عصبی تابع ‌پایه ‌شعاعی

آنومالی سطح دریا (sla، sea level anomaly) به‌عنوان کمیتی که بیان‌کننده اختلاف ارتفاع سطح ‌لحظه‌ای آب با مقدار متوسط سطح‌ آب در یک بازه زمانی می‌باشد در مطالعه وضعیت سطح آب مناطق مختلف دارای اهمیت چشم‌گیری می‌باشد. منطقه آبی دریاچه خزر به‌عنوان یکی از دو منبع مهم آبی برای کشور ایران از اهمیتی استراتژیک برخوردار است. بدین‌منظور در این پژوهش با استفاده از داده‌های گذر 92 ماموریت‌های ارتفاع‌سنجی‌ماهواره‌ای (توپکس‌پوزیدون، جیسون1، جیسون2 و جیسون3)؛ عبوری از منطقه آبی خزر به مشاهده تغییرات کمیت آنومالی سطح دریا در این منطقه از سال 1993 تا سال 2020 پرداخته شده است. سپس این کمیت با استفاده از روش تجزیه به حالت‌های‌ ذاتی (emd، emperical mode decompsition) به‌عنوان روشی کارا در جداسازی فرکانس‌های تشکیل‌دهنده یک سیگنال مورد آنالیز قرار گرفته‌است و سپس با استفاده از شبکه عصبی توابع پایه شعاعی (rbf، radial basis function) یک شبکه به‌منظور پیش‌بینی آنومالی سطح دریا ایجاد شده است. 9 فرکانس غالب به‌همراه یک ترند نتیجه تجزیه سیگنال مدنظر در این پژوهش می‌باشد که در نهایت منجر به پارامترهای؛ مجذور میانگین خطا به میزان 0.029 متر و 0.034 متر به‌همراه ضریب‌همبستگی 0.99 و 0.97 به‌ترتیب در دو مرحله آموزش و تست شبکه عصبی می‌شود.

Sea level anomaly prediction using Empirical Mode Decomposition and Radial Basis Function Neural Networks

Sea level anomaly as a parameter that expresses the difference between the instantaneous water level height and the average amount of water level in a period of time is of great importance in studying the water level situation in different regions. Predicting a time series requires that the series be static and that seasonal trends and changes be removed from the observations to eliminate the dependence of variance and mean on time. For this purpose, the use of various methods to static a time series has been suggested and used. Using the method of decomposition into the intrinsic modes of a signal that underlies the formation of intrinsic mode functions that include parts of the signal with approximately the same frequency; in order to analyze and isolate the trend and seasonal changes of the signal have been considered. Caspian sea as the largest lake in the world or the socalled largest enclosed water area in the world is located in northern Iran. This important water area has become one of the main sources of income for its peripheral countries. It has important oil and gas resources as well as the main source of sturgeon as one of the most expensive food sources in the world. This strategic region is known as a medium for connecting the East and the West of the world. In addition to the economic and commercial dimension, the Caspian Sea is of great importance from the military point of view, as numerous military maneuvers are held every year by the neighboring countries. For the above reasons; awareness of the water level and its changes has become increasingly important, especially over the past few decades, but despite this importance, not many studies have been conducted to study the water level. Therefore, in this research, using satellite altimeter data, the monitoring of water level changes in this area has been done. In this study a coverage of the sea anomaly parameter and its changes from 1993 to the present has been provided. The Caspian Sea water region as one of the two important water sources for Iran, is strategically important.For this purpose, in this study, using the transit data of 92 satellite altimetric missions passing through the Caspian Sea region, the changes in the sea level anomaly in this region since 1993 have been observed. This quantity is then analyzed using the method of analysis of intrinsic modes as an efficient method in separating the frequencies that make up a signal and then, using a neural network, a network of radial base functions has been created in order to predict sea level anomaly. 9 dominant frequencies along with a trend are the result of signal analysis considered in this study. Finally, it leads to the parameters of the mean square error of 0.029 m and 0.034 m with a correlation coefficient of 0.99 and 0.97, respectively, in the two stages of neural network training and testing.