پیش بینی دمای ماهانه با استفاده از روش‌های جدید مبتنی بر یادگیری ماشین در اقلیم‌های مختلف ایران

دمای هوا از اجزای اصلی مطالعات هیدروکلیماتولوژی، و متغیری کاربردی برای مدل های برنامه ریزی و بهره برداری از منابع آب است. دستیابی به داده های قابل اعتماد برای پیش بینی تغییرات دمایی در سال های آتی به منظور کاربرد آن در مدل های هواشناسی و هیدرولوژیکی از مهم ترین چالش ها است. طی سالیان اخیر، مدلهای پیش بینی مختلفی توانسته اند به‌عنوان یک راهکار قابل اطمینان مورد توجه قرار گیرند. اکثر این مدل ها بر اساس داده های تاریخی و با بهره گیری از تکنیک های هوش مصنوعی عمل می کنند. در تحقیق حاضر با استفاده از روشهای مبتنی بر هوش مصنوعی مانند روشهای هیبرید ga-ann ، ica-ann، pso-ann و مدلهای elm، orelm سعی شد بهترین مدل برای پیش بینی داده های دمای ماهانه در اقلیم های مختلف ایران (ایستگاه های اهواز، کرمانشاه، مشهد و رشت) در یک دورۀ آماری 58 ساله معرفی گردد، تا بتوان با استفاده از آن به نتایج قابل اعتمادی دست پیدا کرد. نتایج نشان داد که خروجی مدل  orelmدارای بهترین برازش با داده های مشاهداتی با ضریب همبستگی 99/0 بوده، و همچنین دارای بهترین و نزدیک ترین پراکندگی نقاط در اطراف خط 45 درجه می باشد که از این نظر دقیق ترین مدل محسوب می شود. برای اطمینان از صحت انتخاب مدل برتر از نمودار تیلور نیز استفاده شد. نتایج نشان داد که نزدیک ترین نقطه به نقطه مرجع مربوط به روش orelm می باشد. لذا برای پیش بینی دمای ماهانه در اقلیم های مختلف می توان با اطمینان از مدل orelm  استفاده کرد. این رویکرد کمک زیادی به محققین بخش آب و هواشناسی می‌کند تا با استفاده از هوش مصنوعی، تغییرات دمایی را با دقت بالاتری در سال های آتی پیش‌بینی نموده و با اطمینان در مدل های برنامه ریزی منابع آب استفاده نمایند.

forecasting monthly temperature using new methods based on machine learning in different climates of iran

introductionair temperature is one of the main components of hydroclimatology studies and a useful variable for planning and exploitation models of water resources. obtaining reliable data to predict temperature changes in the coming years to use in meteorological and hydrological models is one of the most important challenges.  methodologythe studied areas are ahvaz, kermanshah, mashhad, and rasht stations, located in different regions of iran. the statistical period of the monthly temperature of the selected stations for the use of artificial intelligence models was from 1963 to 2020 (58 years), and all the available statistics were used to select the best model at each station.the aim of this research was to use artificial intelligence as an efficient tool to predict monthly temperatures. in this regard, ga-ann, ica-ann, and pso-ann hybrid models and elm and orelm models were used to predict monthly temperatures. then, considering 80% of the data as training data and 20% of the data as test data, the best model structure with a different number of inputs with the lowest error rate and the highest correlation coefficient with the observational data was obtained. rmse, nrmse, nash, and r statistical indices were used to select the best model. finally, taylor’s diagram was used to ensure the selection of the best model. this diagram introduces the best model with the lowest simulation error based on three indicators: standard deviation, correlation coefficient, and rmse value.  results among the models, the orelm model showed the value of nrmse in the test phase in the stations of ahvaz, kermanshah, mashhad, and rasht, respectively, 0.0378, 0.0461, 0.0523, and 0.0605, which has the highest accuracy. taylor’s diagram also confirmed this result by using more error criteria, and the orelm model is confidently introduced as the best artificial intelligence model for predicting temperature changes in these 4 stations that represent 4 different climates. based on the results, the orelm model is more accurate than other models in the training and test stages with regard to all evaluation indicators. after that, the elm model ranks second in terms of prediction accuracy.  discussionthe possibility of predicting air temperature changes for a long-term period based on a low amount of information compared to atmospheric numerical models and using only monthly temperature data is one of the most important achievements of this research. in this study, monthly temperature changes are predicted based on artificial intelligence methods without the need for complex atmospheric parameters, without the need for complex atmospheric analysis maps and software, and without spending a lot of time and money calibrating and validating mathematical models. considering the importance of knowing temperature changes as one of the most important parameters of hydroclimatology balance, artificial intelligence models used in this research can be recommended, especially for areas without basic statistics or in situations where it is not possible to use mathematical models. based on the results obtained, the models developed in this research can be proposed for other study areas with different climates.  conclusionobtaining reliable data to predict temperature changes in the coming years to use in meteorological and hydrological models is one of the most important challenges. in recent years, various prediction models have been able to be considered reliable solutions. most of these models operate based on historical data and use artificial intelligence techniques. in this research, by using methods based on artificial intelligence such as hybrid methods, ga-ann, ica-ann, pso-ann, and elm and orelm models, the best model for predicting monthly temperature data in different climates in iran was tried. ahvaz, kermanshah, mashhad, and rasht stations) should be introduced over a statistical period of 58 years so that reliable results can be obtained by using them. the results showed that the output of the orelm model has the best fit with the observational data with a correlation coefficient of 0.99 and also has the best and closest distribution of points around the 45-degree line, which is considered the most accurate model in this regard. taylor’s diagram was also used to ensure the accuracy of selecting the best model. the results showed that the closest point to the reference point is related to the orelm method. therefore, the orelm model can be reliably used to predict the monthly temperature in different climates. this approach greatly helps the researchers in water and climatology use artificial intelligence to predict temperature changes with higher accuracy in the coming years and use them confidently in water resource planning models.