مطالعه روند تغییرات فصلی و سالانه بارش با روش رگرسیون چندک (مطالعه موردی: ایستگاه هاشم‌آباد گرگان)

بارش یکی از متغیرهای هواشناسی است که مقدار آن در زمان و مکان از تغییرات زیادی برخوردار است و بر بسیاری از پدیده‌ها و رویدادها در زمینه‌های کشاورزی، محیط‌زیست، منابع ‌طبیعی، فعالیت‌های بشری  موثر است. لذا با توجه به تاثیرپذیری انسان و محیط ‌زیست از بارش، می‌بایست هرگونه تغییر در این عوامل در طول زمان مورد بررسی قرار گیرد. از روش‌های مرسوم جهت بررسی روند در سری داده‌ها، استفاده از آزمون من‌کندال می‌باشد. در صورتی‌که سری داده‌ها متشکل از وقایعی با شدت‌ها و مقادیر مختلف می‌باشد،  لازم است تا دهک‌ها یا صدک‌های سری به منظور تغییر احتمالی مورد بررسی قرارگیرد. بنابراین در این پژوهش، در کنار آزمون من‌کندال و رگرسیون خطی معمولی، از روش رگرسیون چندک نیز  برای بررسی روند فصلی و سالانه بارش در ایستگاه سینوپتیک هاشم‌آباد گرگان در طول سال‌های 1396-1363 (2017-1984 میلادی) استفاده گردید. نتایج آزمون من‌کندال حاکی از وجود روند کاهشی معنی‌دار بارش تنها در فصل بهار می‌باشد. بررسی مقایسه‌ای نتایج روش‌های منکندال و رگرسیون خطی معمولی نشان می‌دهد که هر دو روش جهت تغییرات را به طور مشابه نشان می‌دهند اما بزرگی تغییرات در روش منکندال بیشتر از روش رگرسیون خطی می‌باشد.  نتایج رگرسیون چندک نشان می‌دهد که تمام چندک‌ها در یک سری زمانی، از شیب یکسانی تبعیت نمی‌کند و حتی ممکن است در یک سری، برخی چندک‌ها دارای شیب افزایشی و برخی دارای شیب کاهشی باشند. همچنین نتایج نشان داد که از نظر معنی‌داری شیب، در فصل بهار، چندک‌های میانی شیب کاهشی ولی در فصل‌ تابستان، چندک‌های بالایی شیب افزایشی دارند. در فصل پاییز چندک‌های پایینی دارای روند کاهشی و در فصل زمستان چندک‌های میانی روند کاهشی، اما چندک‌های بالایی روند افزایشی  دارند و همچنین در سری سالانه بارش، چندک‌های پایینی دارای شیب کاهشی بارش می‌باشند که به طور کلی می‌توان اظهار داشت که بر ترسالی‌های شدید و خشکسالی‌های شدید افزوده شده است.

Study of Seasonal and Annual Rainfall Changes with Quantile regression method (Case Study: Gorgan Hashem-Abad Station

Abstract Introduction: Precipitation is an important meteorological variable throughout time and place, and affects many phenomena and events in the contexts of agriculture, environment, natural resources, human activities, and etc. The study of variations in precipitation at different time scales is of great importance. Investigating precipitation’s seasonal time scale can be indicative of the variation pattern of this variable within a year, and its interpretation is beneficial in understanding both the patterns of wet and dry periods within a year and also seasonal hydrological components, while the investigation of annual time scale in a region can significantly be effective in better understanding the changes in the hydrological cycle in the studied area. Therefore, understanding the variability of hydrological processes and their associated statistics is essential for better water resource management.  Also, the changes in when the precipitation begins, can cause variations in the length of wet and dry periods. According to the impacts climate variables have on human and environment, it is necessary to review any changes in these variables over time. The use of the nonparametric MannKendall test to examine the trend in the data series is a common method, which the analysis derived from it can lead to an initial understanding to find out whether the data is random (with no trend) or a trend exists within the data series being studied, but to better understand changes in a variable over time, it is better to examine the changes in different quantiles. The data series consists of events of varying intensities and quantities, and it is therefore necessary to study the deciles or percentiles of the series in order to investigate the aligned or nonaligned probability change. For this purpose, a quantile regression is suggested. In this study seasonal and annual precipitation of the Hashem Abad station in Gorgan was studied. Materials and Methods: The study area is the synoptic meteorological station of Hashem Abad, Gorgan, with an average annual rainfall of 550 mm and an average temperature of 18 ° C, which has a Mediterranean climate. At this station, weather data have been collected since the year 1984. In this study, precipitation data was used by the end of year 2017. After the establishment of seasonal and annual time series, nonparametric MannKendall tests, Sen slope and quantile regression tests were performed and the results were compared.  Results and Discussion: The results of the MannKendall test and ordinary linear regression showed that the precipitation has a significant decreasing trend only in spring, and the remaining seasons as well as the annual precipitation are of no trend. But the quantile regression shows a number of different results, so that not all the quantiles do follow the same slope in a time series, and even in a series, some of the quantiles have an increasing slope while others have a decreasing slope. In the spring, the midrange quantiles have a significant decreasing slope, but in the rest of quantiles, there is a decreasing and nonsignificant slope. In the summer, the upper quantiles are of increasing slope while lower and midrange quantiles have decreasing slope, and these slopes confirm the existence of a significant increasing statistical trend in only extreme upper quantiles. In the autumn, in many quantiles, there is an increasing positive slope, and only in the extreme lower and upper quantiles a decreasing and negative slope is visible. These slopes show a significant decreasing trend in the extreme lower quantiles. In the winter, in the upper quantiles, show an increasing and positive slope, but in the midrange and lower quantiles, demonstrate a decreasing and negative slope. These slopes are statistically significant in many quantiles. On an annual scale, like the winter, the upper quantiles are of increasing slope and the lower and midrange quantiles are of decreasing slope. These slopes are nonsignificant in many cases, and only in extreme lower quantiles a significant decreasing trend exists.  Comparison of the existing significant trends using different regression methods suggests that the quantile regression method is considered to be useful in estimating extreme precipitations trend that cannot be evaluated by the MannKendall and ordinary linear regression methods.  Final conclusion: A comparative study of the results of the MannKendall and ordinary linear regression methods shows that both methods depict the similar variations, but the magnitude of the variations in the MannKendall method is estimated more than linear regression. Also, the results of this study indicate that changes in different quantiles of data may have a significant difference in direction and quantity with the changes in the mean or average of data, and therefore, it is necessary to study and analyze the quantile regression method in order to properly understand the changes in time series of seasonal and annual precipitation data.