Optimization of Irrigation Systems: a case study of Punjab, Pakistan

Abstract

El país árido y agrícola de Pakistán depende en gran medida del Sistema de Riego de la Cuenca Indo, la cual es alimentada principalmente por el río Indo y sus afluentes Jhelum, Chenab, Ravi y Sutlej. Con el recurso hídrico limitado, el sistema agrícola de la provincia de Punjab necesita mejorar su eficiencia y distribución equitativa. Este estudio tiene como objetivo desarrollar un modelo de red flexible que involucra una serie de nodos interconectados para caracterizar el sistema de riego de Punjab, Pakistán. El modelo se calibró utilizando el método Bayesiano de Monte Carlo, donde el conocimiento sobre los parámetros de pérdida/ganancia se actualizó con la información de datos observados. Los resultados del modelo de red propuesto fueron aceptables tanto para la calibración y validación, excepto en algunas colas de presa. Además, los coeficientes de ganancia/pérdida del enrutamiento optimizados explicaron coherentemente las condiciones de campo en el área de estudio. Finalmente, los hallazgos de este estudio implican que las zonas con mayores caudales de descarga son propensas a grandes incertidumbres. El modelo puede ser de utilidad para los administradores de recursos hídricos involucrados en la gestión del agua en la provincia de Punjab.

Water resource is being stressed due to increasing levels of social demands, particularly in emerging and least developed nations. The arid and agro-based country of Pakistan is highly dependent on Indus Basin Irrigation System (IBIS), one of the biggest contiguous irrigation networks in the world, mainly fed by the Indus River, and its Eastern tributaries Jhelum, Chenab, Ravi, and Sutlej. With limited water resources, irrigated agriculture needs to improve its efficiency and equitable distribution. This study aims to develop a flexible network model that involves a series of nodes interconnected via links to characterize the Punjab irrigation within the IBIS system. The model is calibrated by using Bayesian Monte Carlo method, where available knowledge about gain/loss parameters is updated with the information in observed data. After a rigorous calibration and validation, the model performance was assessed through the use of the coefficient of determination (R2), Nash–Sutcliffe Efficiency (NSE), Kling- Gupta Efficiency (KGE), and Percent bias (PBIAS). The overall results of the proposed network model were acceptable for both calibration and validation periods, except in some particular barrage tails. Nevertheless, due to the complexity of the irrigation network and the simultaneous calibration for the entire system, the model behavior is considered satisfactory. The optimized routing gain/loss coefficients coherently explained the field conditions in the study area. The findings in this study imply that zones with higher flow discharges are prone to large uncertainties. In addition, posterior probability distributions of gain/loss coefficients suggest an adequate identification of the parameter uncertainties, with changes in the prior probability distributions. It also demonstrates how a selection of prior distribution bounds and model parameters could improve the reliability and robustness of the model. The model would be helpful to water resource managers involved in agricultural water management in Punjab province. However, the utility of the model should be taken with great care because there are still uncertainties in the modeling results due to the conceptual nature and the quality of input data.