تخمين خشكسالي با استفاده از شبكه هاي هوشمند

Drought Estimate Using Artificial Network Estimation Drought Using Intelligent Networks

خشكسالي يكي از پديده‌هاي آب و هوايي است كه در همه ي شرايط اقليمي و در همه ي مناطق كره ي زمين به وقوع مي‌پيوندد. پيش‌بيني خشك‌سالي نقش مهمي در طراحي و مديريت منابع طبيعي، سيستم‌هاي منابع آب، تعيين نياز آبي گياه ايفا مي‌نمايد. در اين پژوهش جهت تخمين شاخص بارش استاندارد 12 ماهه ي چهار ايستگاه باران سنجي دلفان، سلسله، دورود و بروجرد واقع در استان لرستان از مدل شبكه ي عصبي موجك استفاده شد و نتايج آن با ساير روش هاي هوشمند از جمله شبكه ي عصبي مصنوعي مقايسه گرديد. براي اين منظور از پارامتر بارش در مقياس زماني ماهانه در طي دوره ي آماري (1372-1392) به عنوان ورودي و شاخص بارش استاندارد به عنوان پارامتر خروجي مدل ها انتخاب گرديد. معيارهاي ضريب همبستگي، ريشه ي ميانگين مربعات خطا و ميانگين قدر مطلق خطا براي ارزيابي و عملكرد مدل ها مورد استفاده قرار گرفت. نتايج نشان داد هر دو مدل قابليت خوبي در تخمين شاخص بارش استاندارد دارند، ليكن از لحاظ دقت، مدل شبكه ي عصبي موجك عملكرد بهتري نسبت به شبكه ي عصبي مصنوعي از خود نشان داده است. در مجموع نتايج نشان داد استفاده از مدل شبكه ي عصبي موجك مي‏تواند در زمينه تخمين خشكسالي موثر باشد.

Background and Objective Drought is one of the phenomena of climate that occurs in all climatic conditions and in all parts of the planet. Drought prediction has an important role in designing and managing natural resources, water resource systems, and determining the plantʹs water requirement. For estimating drought, various approaches have been introduced in hydrology that artificial models are the most important ones. In this study for evaluating the accuracy of the models in estimating the 12-month standard rainfall index, monthly data from four weather stations in Boroujerd, Dorood, Selseleh and Dolphan in Lorestan province have been used. For modeling of drought in these stations utilized wavelet neural network and artificial neural network models and the results were compared to each other for the accuracy of the studied models. In a few studies, each of the models presented in the drought estimation has been studied. But the purpose of this research is simultaneous analysis of these models at four stations for estimating the standard rainfall index. Methods In this study, Boroujerd, Dorood, Selseleh and Dolphan that located in Lorestan province have been selected as the study area During the statistical period, the precipitation parameter was used at monthly time scale (1962-1372) for input and standard rainfall index as the output parameter of the models. For this purpose, at first 80% of the data (1372-1382) were selected for calibration of the models and 20% of the data (2012-2013) were used to validate the models. The wavelet neural network, which has a very good fit with the sinusoidal equations by separating the signal into high and low frequencies, can greatly increase the accuracy of the model and reduce noise. Artificial neural networks are inspired by the brain information processing system that ability to approximate patterns of a model has increased the scope of these networks. Correlation coefficient, root mean square error and mean absolute error value were used for evaluation and performance of the models. Results The results showed that both models have good performance in estimating the standard rainfall index in the four stations studied. Also, according to the evaluation criteria, the wavelet neural network model was found to have the highest accuracy and low error rate compared to the artificial neural network model. Conclusions In total, the results showed that the use of wavelet neural network model can be effective in estimating the standard rainfall index. also It can be useful in facilitating the development and implementation of management strategies to prevent drought and is a step in making managerial decisions to improve water resources.