Modeling some factors of affecting rill erosions using fuzzy inference system

Rill erosion is causes soil degradation, loss of rangeland potential and unfavorable conditions for plant establishment. So understanding, modeling and predicting this type of erosion on rangeland and natural resource management is important. One of the major challenges in planning and management of natural ecosystems such as rangelandsis lack of understanding the complexities and dynamics of these ecosystemsinconventional models. That based on basic idea of fuzzy theory, this limit can be entered in the model. The main aim of this study is to design and evaluate the performance of rill erosion prediction model using fuzzy inference systems on a part of Ahmadabad- Mashhad rangelands.
In this study by establishing 50-m transects in each different slope length, the number of rills and effective parameters in shaping rills, including canopy cover, ground cover, gravel, sand, silt, clay, slope and mutual effects between length and degree of slope have been measured. By measuring, width and depth of any rill and their geometric section, cross-section of rills have been calculated. Then by using Gamma test, the optimal combination of variables "percentage of gravel surface", "mutual effects between length and degree of slope" and "amount of silt available in soil" as inputs factors and "cross-section of rills per unit length" as variable Output (Function) were chosen for the fuzzy modeling. Some (9) different combinations of membership functions to fuzzification of input and output variables with 5 defuzzification approach combined maximum - minimum of fuzzy were trial and error. Model is chosen so that the addition of small amounts of IPE, the generalizability also is appropriate.
Results showed that the defuzzification of smallest maximum (SOM) has the lowest IPE compared with the other methods and then the mean of maximum method (MOM) with the lowest IPE was used in second priority. In all methods of defuzzification (except the mean of maximum method), models such as “Gaussian – trapezoidal” “Gaussian 2 – trapezoidal” and “Gaussian 2” selected models are known. According to the results obtained, in the erosion modeling using the softer functions such as Gaussian, will lead to better results than the sharp and simple functions such as trapezoidal and triangular. The combination of two trapezoids and triangles will have better results than using each alone, if only the simplicity of the model has to be considered.
Gaussian membership function with the defuzzification of smallest maximum is the optimal model for predicting rill erosion in Ahmad-Abad area of Mashhad.