توابع انتقالی برای برآورد میزان رطوبت خاک با به کارگیری پارامترهای فرکتالی در استان اردبیل

اطلاع از میزان رطوبت خاک و ظرفیت نگه‌داشت، در مدیریت اراضی و زراعی مفید می باشد و می‌تواند در پیش بینی میزان و زمان تولید رواناب کاربرد داشته باشد. هدف این پژوهش، ارائه و مقایسه ی توابع انتقالی رگرسیونی و شبکه عصبی مصنوعی برای برآورد رطوبت‌های ظرفیت زراعی (fc)، نقطه پژمردگی دائم (pwp) خاک و بررسی تاثیر استفاده از پارامترهای فرکتالی ذرات اولیه، خاکدانه و منافذ خاک در افزایش دقت این برآوردها بود. برای این منظور، در مجموع 90 نمونه خاک از سه منطقه در استان اردبیل (دشت اردبیل، فندقلو و سرعین) به‌صورت تصادفی برداشته شد. توابع رگرسیونی، برای برآورد fc و pwp یک بار با کاربرد و یک بار بدون کاربرد ابعاد فرکتالی (ابعاد فرکتالی ذرات اولیه خاک، خاکدانه ها و منافذ خاک) به عنوان متغیر مستقل در مدل سازی ایجاد شد. بنابراین، برای برآورد هرکدام از پارامترها (fc و pwp) دو تابع به وجود آمد. هنگامی که ابعاد فرکتالی در ارائه ی توابع انتقالی برای تخمین fc و pwp به کار گرفته شد، سه متغیر (جرم مخصوص‌ظاهری، جرم مخصوص حقیقی و بعد فرکتالی منافذ خاک) به عنوان تخمین‌گر به مدل وارد شد. هنگامی که از این ابعاد در مدل سازی استفاده نشد، تابع انتقالی fc با چهار تخمین‌گر (جرم مخصوص‌ظاهری، جرم مخصوص حقیقی، میانگین هندسی قطر (dg) و انحراف هندسی قطر (σg) ذرات اولیه خاک) ایجاد شد و تابع انتقالی pwp با دو (تخمین گر جرم مخصوص‌ظاهری و جرم مخصوص حقیقی). نتایج نشان داد که در روش شبکه عصبی، استفاده از ابعاد فرکتالی برای تخمین رطوبت pwp و fc به افزایش دقت توابع منجر شد. همچنین استفاده از ابعاد فرکتالی خاکدانه‌ها توانست در افزایش دقت مدل های شبکه عصبی مصنوعی ارائه شده برای تخمین fc نیز موثر باشد، اما دقت مدل‌های ارائه شده را برای تخمین pwp چندان افزایش نداد.

Pedotransfer functions for estimating soil moisture content using fractal parameters in Ardabil province

Extended abstract 1 Introduction Soil moisture curve is an important characteristic of soil and its measurement is necessary for determining soil available water content for plant, evapotranspiration and irrigation planning. Direct measurements of soil moisture coefficients are timeconsuming and costly. But it is possible to estimate these characteristics from readily available soil properties. The purposes of this study were: 1) development of pedotransfer functions (PTFs) for estimating of soil moisture content at field capacity (FC) and permanent wilting point (PWP) conditions by artificial neural networks system (ANN) and multivariate regression method and 2) investigation effects of using soil primary particles, aggregates and porosity fractal dimensions as a predictor for increasing the accuracy and reliability of these PTFs. 2 Methodology For this reason, 90 soil samples from three regions (Agricultural land of the Ardabil plain, Forest of the Fandoglo and Rangelands of the Sareyn, which were located in Ardabil province) were collected in random design sampling method. Then FC and PWP coefficients of these soils were measured using pressure plates apparatus. As well as, some readily available properties of soils such as fractal dimensions (primary particles, aggregates, and soil pores), texture, bulk density and particles density, porosity, organic carbon and calcium carbonate equivalent (CCE) were determined by routine laboratory method. Then data were divided into two datasets randomly: Training dataset (including 72 soil samples) and test dataset (including 18 soil samples). RegressionPTFs for estimating FC and PWP were developed once by using and once without using of the fractal dimension of primary particles (DS), the fractal dimension of aggregates (Df) and fractal dimensions of soil pores (Dy) as independent variables. The predictors of RegressionPTFs once again were used for development of the ANNsPTFs. Therefor two PTFs were developed for predicting each dependent variable (FC and PWP). Statistical and Neurosolution softwares were used for development of the RegressionPTFs and ANNPTFs, respectively. Finally, the accuracy and reliability of PTFs were investigated. 3 Results Discussion Results showed that FC has a positive significant correlation with soil silt (r= 0.52**) and organic carbon content (r= 0.86**), and a negative significant correlation with sand (r= 0.50**), CCE (r= 0.74**), bulk density (r= 0.64**), particles density (r= 0.79**) and Df (r= 0.47**). As well as, there are positive significant correlation between PWP and other soil properties such as soil silt (r= 0.48**) and organic carbon content (r= 0.77**), and negative significant correlation with sand (r= 0.50**), CCE (r= 0.74**), bulk density (r= 0.70**), particles density (r= 0.80**) and Df (r= 0.52**). Results also showed that there is a positive significant correlation between FC and PWP (r= 0.84**). When fractal dimensions used as independent variables for estimating of FC, three variables (bulk density ( rho;b), particles density ( rho;p), and fractal dimension of soil pores (Ds)) included as a predictor in PTFs and these predictors could explain 80% and 98% of variation of FC, at Regression and ANNPTFs, respectively. But when fractal dimensions didn rsquo;t used in modeling, PTFs was developed with four predictors ( rho;b, rho;p, dg and sigma;g) and these predictors could explain 81% and 92% of the variation of FC, at Regression and ANNPTFs, respectively. Results also showed that there were no significant differences between the Regression and ANNPTF which achieved for the estimation of FC values. As well as, RegressionPTF by using fractal dimensions as independent variables for the estimation of PWP was developed with three predictors ( rho;b, rho;p and Ds) and these predictors could explain 76% and 92% of the variation of PWP, at Regression and ANNPTFs, respectively. But when fractal dimensions weren rsquo;t used as independent variables, PTFs was developed with two predictors ( rho;b and rho;p), and these predictors could explain 71% and 85% of the variation of PWP, at Regression and ANNPTFs, respectively. Results of the investigation of accuracy and reliability of the PTFs showed that when fractal dimensions used as independent variables for estimating of PWP, only the accuracy and reliability of the ANNPTF was increased. 4 Conclusions ANNPTFs were more accurate than RegressionPTFs. When fractal dimensions of soil primary particles, aggregates, and pores were used as independent variables in modeling for the prediction of FC and PWP, only the fractal dimension of soil pores included as a predictor and increased the accuracy of ANNPTFs, but it could not increase the accuracy of RegressionPTFs.