Hydrologic Design of Soil and Water Conservation Structures Using Probability Analysis and Machine Learning Techniques in Saurashtra Region of Gujarat, India
P. A. Pandya *
Centre of Excellence on Soil and Water Management, Research Training and Testing Centre, Junagadh Agriculture University, Junagadh, India.
S. H. Bhojani
Directorate of Information Technology, Anand Agricultural University, Anand, Gujarat, India.
S. H. Parmar
Centre of Excellence on Soil and Water Management, Research Training and Testing Centre, Junagadh Agriculture University, Junagadh, India.
G. V. Prajapati
Centre of Excellence on Soil and Water Management, Research Training and Testing Centre, Junagadh Agriculture University, Junagadh, India.
D. D. Vadaliya
Centre of Excellence on Soil and Water Management, Research Training and Testing Centre, Junagadh Agriculture University, Junagadh, India.
*Author to whom correspondence should be addressed.
Abstract
The soil and water conservation structures are constructed to overcome water scarcity as a result of interannual rainfall variability and paucity of the perennial source of water. The present study was aimed to estimate the design runoff for the efficient hydrologic design of various soils and water conservation structures using machine techniques for enabling efficient harvesting of available rainfall with economical design which can support in developing climate resilience for the Saurashtra region of Gujarat, India. The design rainfall at various return periods was predicted by Annual One Day Maximum Rainfall (ADMR) using three technics i.e. Probability Distribution Fitting, Artificial Neural Network (ANN) and Gaussian Process Regression (GPR) for 11 stations. Various goodness of fit tests revealed that ADMR was efficiently predicted by log-logistic (3P) distribution for six stations, generalized extreme value distribution for two stations and lognormal (3P), gamma (3P) and lognormal distribution for one station each. Among ANN and GPR, the performance indicators revealed that GPR has shown a higher capability to predict ADMR as compared to ANN with correlation coefficient ranging from 0.97 to 0.99, mean absolute error from 15 mm to 411 mm and root mean squared error from 40 mm to 494 mm for various stations. The design runoff estimation was demonstrated based on predicted ADMR for return periods suitable for various soil and water conservation structures like field bunding, terrace outlets and vegetative outlets, field diversion, permanent masonry gully control structures, earthen dam, etc. using SCS-Curve Number method for curve number 70 and 85. The study is useful for researchers, planners and engineers to implement the economical, efficient and safe design of various soil and water conservation structures.
Keywords: Annual one-day maximum rainfall, artificial neural network, Gaussian process, probability distribution, runoff estimation