Modeling Soil Cation Exchange Capacity in Arid Region of Iran: Application of Novel Hybrid Intelligence Paradigm.

2 The potential of the soil to hold plant nutrients is governed by cation exchange 3 capacity (CEC) of any soil. Estimating soil CEC aids in conventional soil management 4 practices to replenish the soil solution that supports plant growth. In the present study, 5 a multiple model integration scheme driven by hybrid GANN (MM-GANN) was 6 developed and employed to predict the accuracy of soil CEC in Tabriz plain, an arid 7 region of Iran. The standalone models (i.e., artificial neural network (ANN) and 8 extreme learning machine (ELM)) were implemented for incorporating in the MM- 9 GANN. In addition, it was tested to enhance the prediction accuracy of the standalone 10 models. The soil parameters such as clay, silt, pH, carbonate, calcium equivalent 11 (CCE), and soil organic matter (OM) were used as model inputs to predict soil CEC. 12 By the use of several evaluation criteria, the results showed that the MM-GANN 13 model involving the predictions of ELM and ANN models calibrated by considering 14 all the soil parameters (e.g., Clay, OM, pH, Silt, and CCE) as inputs provided superior 15 soil CEC estimates with an NSE = 0.87. The proposed MM-GANN model is a reliable 16 intelligence based approach for the assessment of soil quality parameters intended for 17 sustainability and management prospects. 18


21
Cation exchange capacity (CEC) refers to the extent of the soil's capacity to preserve 22 exchangeable cations, the like of which has a direct bearing on soil fertility triangle 23 (Wolf, 1999). Soil CEC is a sensitive indicator of natural and human-induced 24 perturbations over soil profile and groundwater. Monitoring changes in soil CEC can 25 assist in predicting whether soil quality has degraded, improved, or sustained under 26 diverse agricultural or forestry schemes. In the course of conventional soil 27 management practices to replenish the soil solution that supports plant growth, the 28 negatively charged clay particles and organic substances adsorb and hold on positively 29 charged soil nutrients (e.g. NH4 + , K + , Mg 2+ and Ca 2+ etc.) via electrostatic forces 30 (Ketterings et al., 2007). Depending on the soil structure, CEC clearly demonstrates 31 the shrink-swell potential of any soil; a high CEC value (>40 meq/100g) denotes that 32 a soil structure will recuperate gradually, and at sometimes can show expansive 33 behavior. In contrast, a soil with low CEC value (<10 meq/100g) will have reduced 34 capacity to hold water and end up being acidic rapidly (Thomas et al., 2000). Soil 35 CEC can fluctuate according to clay percentage, soil pH, ionic strength, soil-to-36 solution ratio, clay type and changing organic matter composition. For agriculture, the 37 preferred value of CEC is >10 meq/100g for exchange between plant root hairs and 38 soils (Mengel, 1993). The leaching of contaminants into the underlying aquifer system ). Therefore, in the early stages of agriculture, it is necessary to estimate CEC for 44 determining the supplemental nutrient needs or to remove excess salts which influence 45 over soil structure and agricultural productivity. Soil CEC is a sensitive indicator of 46 natural and human-induced perturbations over soil profile and groundwater. 47 Monitoring changes in soil CEC can assist in predicting whether soil quality has 48 degraded, improved, or sustained under diverse agricultural or forestry schemes. 49 Various methods for direct measurement of soil CEC have been reported extensively 50 over the literature (Delavernhe et al., 2018;Dohrmann, 2006aDohrmann, , 2006b. Multiple 51 comparison of CEC estimation techniques is presented by Conradie and Kotze, 52 (1989). In addition, there were several ancillary approaches such as pedotransfer CEC of agricultural soils found to be directly related with estimated charge of organic 63 carbon and clay in the soil at the actual pH of the soil. Using soil organic and non-64 carbonate clay contents as predictors, Seybold et al. (2004) explained the variation in geometric mean particle-size diameter, the soil particle-size distribution, and soil 69 organic matter content. Several PTF's relating soil CEC with soil's sand, silt or clay 70 fractions, and soil organic carbon content evaluated by (Khodaverdiloo et al.,71 2018).Scholars took into account of calibration dataset size on the prediction accuracy 72 of soil CEC. These classical pedotransfer function-based approaches often suffer from 73 a high degree of inaccuracy due to spatial scale dependence, non-linear relationships 74 among variables and incompetence to handle mixed data (Van Looy et al., 2017). Relatively few studies were accomplished using support vector machine (SVM), 95 random forests (RF), genetic expression programming (GEP), multivariate adaptive 96 regression splines (MARS), and subtractive clustering algorithm based ANFIS for 97 estimating soil CEC using readily measured soil properties as inputs (Akpa et al., 98 2016;Emamgolizadeh et al., 2015;Jafarzadeh et al., 2016;Liao et al., 2014). A hybrid 99 model integrating ant colony optimization (ACO) algorithm with ANFIS improved 100 the prediction accuracy of soil CEC accompanied by optimal choice of input subset 101 which comprised of soil properties (e.g. soil organic matter, clay, silt, pH and bulk 102 density) (Shekofteh et al., 2017). Although there has been a noticeable progress on 103 the AI implementation with the field of geoscience, the enthusiasm of developing and 104 exploring more reliable intelligent predictive models is still ongoing research era. As 105 a result, the inspiration of developing a multiple learning intelligent model is 106 investigated here for the soil CEC.

107
Hybrid soft computing approaches involving evolutionary algorithms coupled with 108 AI techniques facilitate the development of more sophisticated models with higher 109 prediction accuracy. Hence, in the present study, a hybrid approach involving the  (Xk, Yk) ∈ R n x R n , the standard SLFN with L hidden layer nodes can be described 159 as following equation (1).
where i cR  = assigned bias of the i th hidden node, i wR  = assigned input weight 162 connecting the ith hidden and input layer nodes, i β = the weight connecting the ith 163 hidden and output layer nodes,

218
The topography consists of rugged, mountainous rims and the Urmia Lake is   each input combination are presented in Table 3. In this study, the proposed ANN,   Figure 8. The scatter plot of the three highest performed models presented in Figure   268 9 displays the strength, direction, and form of the relationship between the observed 269 and estimated soil CEC by ANN and ELM models. According to the Figure 9, the 270 ELM model outperformed the ANN model although they have very close performance 271 in terms of the statistical indices (Tables 3 and 4). The ELM model is known for its 272 superior learning speed and virtuous generalization performance than the ANN model.  Table 5. Also, the performance statistics of MM-GANN models are 283 presented in Table 6. The MM-GANN models involving the predictions of ELM and    The schematic diagram of genetic algorithm.

Figure 4
The structure of the proposed MM-GANN model for predicting soil cation exchange capacity. Figure 5