Performance assessment of the artificial intelligence models for prediction of the infiltration rate in the Surface Soil of Geological Formations (Case Study: Aleshtar Watershed, Lorestan Province)

Introduction

Water repellency is a property that commonly affects the soil surface layer. It results from hydrophobic coatings on soil particles that originate from organic matter. The most significant effect of soil water repellency is a reduction in infiltration rates. The infiltration rate is one of the primary processes of the hydrological cycle. Hydrogeological and subsurface phenomena as infiltration, percolation mainly affect natural or man-made geotechnical soil. Understanding these phenomena are essential for estimation of runoff process, groundwater seepage, erosion, transport substances, evapotranspiration in surface and into groundwater are mainly influenced by precipitation. It is the property of water by which it moves through the soil particles. Infiltration process plays a fundamental role in streamflow, groundwater recharge, subsurface flow, and surface and subsurface water quality and quantity. Also, Soil infiltration is one of the key processes in design of irrigation systems, water resources management and soil protection and soil erosion control in watershed management and good knowledge of the infiltration rate is useful in calculating the natural and artificial groundwater recharge and surface runoff. Therefore, the prpose of this study was Performance assessment of the artificial intelligence models for prediction of the infiltration rate in the surface soil of geological formations in Alashtar watershed, Lorestan province, Iran.



Materials and Methods

The study area is a part of Kashkan watershed, Lorestan province, Iran. So, it was selected as a suitable watershed to Modeling of infiltration rate in different vegetation types by the various soft computing techniques. The study area located between 48°10′28″ - 48°23′29″ N latitudes and 33°45′ 17″ - 33°51′ 23″ E longitudes, and covers an area of 112.54 Km2 approximately. Elevation of watershed varies from 3613 to 1481 m a.s.l. The studied area has a cold and semiarid climate with a mean annual rainfall Less than 570 mm. Most parts of Alashtar watershed are rangeland, while forest, dry farming, and irrigation lands are in considerable quantities and The surface lithology in the Alashtar watersheds are covered by the Eocene, Quaternary, Cretaceous, Miocene, Oligocene, Paleocene, and Pliocene geologic formations. In this study, The double-ring infiltrometer was used to measure the infiltration in the surface soil of some geological formations in the study area. After determining the infiltration rate, Gaussian Process (GP), Classification And Regression Tree (CART), and Random Forest (RF), Multivariate adaptive regression splines (MARS), M5P model tree (M5P) and Reduced Error Pruning Tree(REP Tree) were used to Modeling of infiltration rate in different Surface Soil of Geological Formations. Total data set consists of some physical characteristic of soil out of which 70% data used to train the model and 30% data were used to test the models. Finally, the models’ accuracy was assessed using three statistical parameters, Root Mean Square Error (RMSE), Nash-Sutcliffe model efficiency (NSE), and Coefficient of Correlation (CC), were selected to compare the efficiency of all models. Also for rapid and reliable comparisons, we also used Taylor diagrams. The Taylor diagram displays Root Mean Square Error (RMSE), Coefficient of Correlation (CC) and standard deviation (SD) values with closer positions on the diagram indicating better model performance.



Results and Discussion

The results indicated that the surface soil of OML geological formations had a higher cumulative infiltration and average infiltration rate. In this study, Gaussian Process (GP), Classification And Regression Tree (CART), Random Forest (RF), Multivariate adaptive regression splines (MARS), M5P model tree (M5P) and Reduced Error Pruning Tree(REP Tree) were used for infiltration rate in Alashtar watershed, Lorestan province, Iran. Comparison of these models showed that the M5P model tree (M5P) and Reduced Error Pruning Tree (REP Tree) models, with the combination of time, sand, clay, silt, soil density and soil moisture, could estimate infiltration rate with much less error than the other models. The obtained results suggest that the bagging M5P model tree regression technique in training and testing phase (with CC = 0.99, RMSE = 0.009, NSH = 0.006 and CC = 0.99, RMSE = 0.009, NSH = 0.006 respectivly) is more accurate to estimate the infiltration rate as compare to the GP, CART, RF, MARS and REPTree thegiven study area. Finaly The results showed that M5P model is effective in predicting Infiltration Rate (IR) content in the surface soil of geological formations. Comparison of results suggests that there is no significant difference between conventional and soft-computing based infiltration models. The performance of the developed models was also compared using a Taylor diagram, in which an accurate model is indicated by a reference point, with a correlation coefficient of 1 having the same amplitude of variation as the observations. Thus, M5P was shown to be the most accurate model for cumulative infiltration prediction.

Conclusion

Prediction of the infiltration rate is an essential element of hydrologic design, watershed management, irrigation, and agriculture studies. This investigation identifies the optimal model for predicting Infiltration Rate (IR) using several computing approaches, such as Gaussian Process (GP), Classification And Regression Tree (CART), Random Forest (RF), Multivariate adaptive regression splines (MARS), M5P model tree (M5P) and Reduced Error Pruning Tree(REP Tree) models. In this study, 8 input variables, including time, sand, clay, silt, moisture content, soil bulk density, porosity and infiltration rate, were evaluated using three key performance metrics to assess the efficacy of various predictive models. These metrics comprised the CC, MAE, RMSE. Based on the evaluation results, the soft computing techniques model has a suitable capability to predict the infiltration rate of the soil. Finaly, the results shown that Learning algorithms can be used to quantify the amount of infiltration and also to estimate the amount of runoff in different geological formations. Also, the results shown that these models can be used to quantify the amount of infiltration and estimate the amount of runoff in the Surface Soil of Geological Formations. As well as, the results of this research can be used by the local authority to manage properly, systematically and plan development within their areas.