تعيين ضريب فرسايش‌پذيري بين‌شياري خاك بر اساس سيستم‌هاي فازي و فازي- ژنتيك در استان آذربايجان‎شرقي

Determination of interrill soil erodibility coefficient based on Fuzzy and Fuzzy-Genetic Systems

در این تحقیق، جهت برآورد ضریب فرسایش‌پذیری بین‌شیاری خاك از روی خصوصیات سهل الوصول از علم منطق فازی و فازی-ژنتیك استفاده شد. بدین ترتیب ابتدا 36 سری خاك با تفرق خصوصیات بالا انتخاب و برخی از خصوصیات آن‌ها از قبیل بافت، ماده آلی، كربنات كلسیم معادل، گچ و ESP، SAR، CEC، EC و pH با روش‌های متداول آزمایشگاهی در سال1387 تعیین گردید. همچنین بعد فراكتالی خاكدانه‌های خاك با استفاده از الك مرطوب و بدون اصلاح شن با استفاده از مدل ریئو و اسپوزیتو محاسبه و ضریب فرسایش‌پذیری بین‎شیاری خاك به كمك دستگاه شبیه‌ساز باران، و اندازه‌گیری شدت تولید رواناب و رسوب تعیین شد. سپس با استفاده از تحلیل آماری متغیرهای درصد شن و بعد فركتالی خاكدانه‌ها به عنوان ورودی مدل‎ها و متغیر ضریب فرسایش‌پذیری بین‎شیاری به عنوان خروجی مدل انتخاب گردید و متغیرهای كلامی خصوصیات فوق و تابع عضویت آن‌ها تعریف گردید. سپس در سیستم استنتاج ممدانی قوانین مدل نوشته شد. در نهایت خروجی مدل با استفاده از روش میانگین وزنی غیرفازی شد. یك بار دیگر توابع عضویت و وزن‌ قوانین با الگوریتم ژنتیك بهینه گردید و بهینه هر وزن، در سیستم فازی وارد شده و توابع فازی بهینه شده به دست آمد. مقادیر R2 و RMSE و GMER و GSDER برای مدل فازی به ترتیب برابر 63/0، 592755، 31/1 و 38/1 و برای مدل فازی-ژنتیك به ترتیب برابر 70/0، 441942، 10/1 و 04/1 به دست آمد، كه حاكی از دقت و كارآیی بالاتر مدل فازی-ژنتیك نسبت به مدل فازی و بیش برآوردی و پخشیدگی نسبتاً زیادتر داده‌های تخمینی مدل فازی نسبت به مدل فازی-ژنتیك می‌باشد.

Introduction: Although the fuzzy logic science has been used successfully in various sudies of hydrology and soil erosion, but in literature review no article was found about its performance for estimating of interrill erodibility. On the other hand, studies indicate that genetic algorithm techniques can be used in fuzzy models and finding the appropriate membership functions for linguistic variables and fuzzy rules. So this study was conducted to develop the fuzzy and fuzzy–genetics models and investigation of their performance in the estimation of soil interrill erodibility factor (Ki). Materials and Methods: For this reason 36 soil samples with different physical and chemical properties were collected from west of Azerbaijan province . soilsamples were also taken from the Ap or A horizon of each soil profile. The samples were air-dried , sieved and Some soil characteristics such as soil texture, organic matter (OM), cation exchange capacity (CEC), sodium adsorption ratio (SAR), EC and pH were determined by the standard laboratory methods. Aggregates size distributions (ASD) were determined by the wet-sieving method and fractal dimension of soil aggregates (Dn) was also calculated. In order to determination of soil interrill erodibility, the flume experiment performed by packing soil a depth of 0.09-m in 0.5 × 1.0 m. soil was saturated from the base and adjusted to 9% slope and was subjected to at least 90 min rainfall . Rainfall intensity treatments were 20, 37 and 47 mm h-1. During each rainfall event, runoff was collected manually in different time intervals, being less than 60 s at the beginning, up to 15 min near the end of the test. At the end of the experiment, the volumes of runoff samples and the mass of sediment load at each time interval were measured. Finally interrill erodibility values were calculated using Kinnell (11) Equation. Then by statistical analyses Dn and sand percent of the soils were selected as input variables and Ki as independent variables for development fuzzy and fuzzy- genetic models. For this reason their linguistic variables were defined and fuzzy models rules were written by Mamdani 's fuzzy inference method. Then, the outputs of model defuzzified by centroid method. Once again, generation of membership functions and fuzzy rules base as well as optimization of fuzzy rule bases was performed by genetic algorithm, and the fuzzy functions were determined by optimized weight of membership functions and fuzzy rules. Results Discussion: Interrill erodibility parameters (Ki) of the examined soils calculated at 3 rainfall rates using are listed in Table 2. The values ranged from 1.03 to 71.79 × 105 kg s m-4, depending on the soil and rainfall intensity. Results showed that the effect of rainfall intensity on Ki turned to be insignificant. This implies that Ki was independent of rainfall intensities. Results showed that the Triangular and Trapezoidal membership functions are better than the other membership functions for linguistic variables which used in this study. The values of R2, RMSE (Root mean square error) and GMER (Geometric mean error ratio) and GSDER (Geometric standard deviation of error ratio) were 0.63, 592755, 1.31 and 1.38 for the fuzzy model, and, 0.70, 441942, 1.10 and 1.044 for the fuzzy- genetic model, respectively. Higher R2 and lower RMSE of the fuzzy – genetic model shows higher accuracy and efficiency of the fuzzy-genetic model. The GSDER criteria shows better matching of the fuzzy- genetic model estimated values with measured values. The GMER criteria shows lower over-estimation of the fuzzy- genetic model than fuzzy model. Conclusion: Fuzzy and fuzzy-genetic models which were designed with two input variables namely aggregates fractal dimensions and soil sand content, capable to predict of interrill erodibility coefficient of soils with reasonable accuracy. So using of these models for predicting of interrill erodibility is recommended.Optimization of fuzzy rule bases and membership functions weight increased model accuracy. Therefore, using genetic algorithm in developing fuzzy models for prediction of soil erosion rate is recommended.