مقایسه کارآیی مدل سنجه رسوب و شبکه عصبی مصنوعی در برآورد بار کف رودخانه‌ها

به دلیل مشکلات نمونه‎برداری و عدم دقّت کافی معادلات تجربی، سنجش و گزینش مناسب‎ترین روش‎های برآورد رسوبات بار کف، اهمّیّت زیادی دارد.هدف پژوهش حاضر، مقایسه کارآیی مدل‎های آماری شبکه عصبی مصنوعی و منحنی سنجه رسوب در برآورد رسوبات بار کف است؛ بدین منظور، ابتدا 5 ایستگاه هیدرومتری دارای بیشترین تعداد نمونه انتخاب شدند؛ سپس منحنی سنجه رسوب و مدل شبکه عصبی مصنوعی با 70% داده‌های آنها ساخته و ارزیابی دقّت برآورد دو مدل با 30% باقیمانده نمونه‎ها انجام شد. نتایج نشان داد که در تمامی ایستگاه‎ها، با افزایش مقادیر دبی جریان، رسوبات بار کف نیز افزایش می‌یابد. میانگین سطح معنی‎داری تفاوت بین مقادیر مشاهداتی و برآوردی مدل شبکه عصبی مصنوعی (0.59) بالاتر از مدل منحنی سنجه رسوب (0.14) است که نشان‎دهنده تفاوت کمتر مقادیر مشاهداتی و برآوردی مدل شبکه عصبی مصنوعی نسبت به مدل منحنی سنجه رسوب و صحّت بیشتر برآوردهای مدل شبکه عصبی مصنوعی است؛ همچنین در تمام ایستگاه‎ها، شاخص مجذور میانگین مربعات خطا برای مدل شبکه عصبی مصنوعی کمتر از مدل منحنی سنجه رسوب است، به طوری که مجموع مجذور میانگین مربعات خطای پنج ایستگاه برای مدل شبکه عصبی مصنوعی و منحنی سنجه رسوب به ترتیب برابر 2505.7 و 5195.3 محاسبه شد. بالاتر بودن ضریب همبستگی بین مقادیر مشاهداتی و برآوردی در هر پنج ایستگاه، با استفاده از مدل شبکه عصبی مصنوعی (0.765) نسبت به مدل منحنی سنجه رسوب (0.5038)، نشان از تخمین‎های دقیق‎تر مدل شبکه عصبی مصنوعی دارد. در نهایت، مدل شبکه عصبی مصنوعی که از دقّت بالاتری نسبت به مدل سنجه بار کف برخوردار است، به عنوان مدل برتر انتخاب شد. با توجّه به مشکلات اندازه‎گیری رسوبات بار کف و اریب زیاد ناشی از محاسبه بار بستر به عنوان درصدی از بار معلّق، نتایج این پژوهش می‎تواند کمک شایانی به برآورد دقیق‎تر بار بستر و نیز بار کلّ رسوبی نماید.

Comparing the Efficiency of Sediment Rating Curve and ANN Models in Estimating River Bedload

Evaluation and selection of the most appropriate methods for bedload estimation is necessary because of the sampling difficulties and inaccurate estimations of the empirical equations. The present study aims to compare the efficacy of ANN and SRC statistical models to estimate the bedload sediments. Collecting bedload measurement data and their respective discharges, 5 stations with the highest number of samples were selected. Then, SRC and ANN models were developed. Finally, the estimations of two models were compared with observed values using correlation coefficient and RMSE indices. The results showed that bedload has been increased by increasing the amount of flow rate in all hydrometric stations. Significant level of difference between observed and estimated values ​​of the ANN model (0.592) is greater than the SRC model (0.144). This means that observed and estimated values ​​of the ANN model are closer together than SRC model, so estimations of ANN model are more accurate. The Root Mean Square Error index (RMSE) for ANN model is also smaller than the SRC model in all stations, so that the sum of five stations RMSE for ANN and SRC models were 2505.7 and 5195.3 respectively. The correlation coefficients of the ANN model are greater than SRC model in all stations. The greater average of correlation coefficients of five stations using ANN model (0.765) than the SRC model (0.503) indicate that ANN model has more accurate estimations. Finally, ANN model was selected as more appropriate model to estimate bedload sediments. Regarding the measurement problems of bedload, our results can lead to making more accurate estimations of bedload and total sediment load. Extended Abstract 1Introduction The sustainable development approach is possible by maintaining and managing triple sources of water, soil and vegetation in the watersheds. The existence of natural factors causing erosion in Iran has made Iran have a high potential for soil erosion and sedimentations. The sediment load of the rivers can be divided into two categories, including suspended load and bedload. Sediment load can be calculated either by direct measurements of sediments or indirectly by sediment transport formulas. Although direct measurements of sediments are more reliable, this is not costeffective for all rivers. In fact, it is particularly more costly and more complex for bedload sediments. On the other hand, estimating the bedload of rivers is very important, because this part of sediment load has a large contribution in total sediment load and also plays a significant role in filling the reservoirs of dams. Due to the complexity of the bedloads transport phenomenon, the relatively precise estimation of bedload sediments is problematic in many rivers, and requires case studies. Therefore, evaluation and selection of the most appropriate methods for bedload estimation are necessary. The present study aims to compare the efficacy of ANN (Artificial Neural Network) and SRC (Sediment Rating Curve) statistical models in five rivers of Iran to estimate the bedload sediments. 2 Materials and Methods Collecting bedload measurements data and their respective flow discharges, 5 hydrometric stations with the highest number of samples were selected. Then, SRC and ANN models were developed using 70% of samples. In order to evaluate the accuracy of the estimations of the two models, the estimated bedload values of the two models for the remaining 30 percent of the flow discharge samples were compared with the corresponding observed values using correlation coefficient (R) and root mean square error (RMSE) indices. Independent ttest was also used to test the significance of the differences between the observed and estimated values of the two models. 3Results and Discussion Although based on the independent Ttest, estimated values of both models are satisfactory, the results of Root Mean Square Error (RMSE) index indicates that there are lower differences between the observed and estimated bedload values for the ANN model in all hydrometric stations. The correlation coefficients between the observed and estimated values of the SRC model are the only significant for Armand station, while correlation coefficients between observed and estimated values of the ANN model are significant (at 5% confidence level) for all stations. These results show that the estimations of ANN model are more accurate than those of SRC model. The result also showed that in all hydrometric stations, bedload has been increased by increasing the amount of flow discharge. Because of the complex relationship between the bedload sediment and flow discharge, it is recommended that Artificial Neural Network model which is well adapted to cope with these complex relationships be used for more accurate bedload estimations. It should be noted that in this research, flow discharge was used as the only input of ANN model to estimate the bedload sediments, while these models are capable of using various parameters affecting bedload discharges as model inputs. 4 Conclusion The purpose of this study was to determine the suitable model to estimate bedload sediments in 5 hydrometric stations located on different rivers of Iran. The results showed that in all hydrometric stations, there is a direct relation between flow discharge and bedload sediments. In other words, bedload discharge has been increased by increasing the amount of flow rate. This study also indicated that bedload estimations of ANN model are more accurate than those of SRC model. Of course, due to the complex relations between the flow and bedload sediment discharges, the suitable model must be determined at each hydrometric station for more accurate estimations of this variable. However, regarding higher accuracy of the Artificial Neural Network estimates, it is recommended that this model be used to estimate bedload sediments in lack of bedload data.