Estimating Evapotranspiration Using an Extreme Learning Machine Model: Case Study in North Bihar, India

The effective scheduling of irrigation requires knowledge of a crop's consumptive water use according to its metabolic activities. Conversely, to know a crop's consumptive water use, one must know its exact evapotranspiration (ET) rate. Although the United Nations' Food and Agricultural Organization (FAO) has recommended using the standard Penman-Monteith method to determine crop ET (ETcrop), the method's intricacies render it impractical to use in the field in predicting agricultural and irrigation requirement-based water needs. The present study investigated the use of a new approach, extreme learning machines (ELMs), for estimating ETcrop using climatic variables such as temperature, relative humidity, rainfall, sunshine hours, and wind speed. ELM is a single, hidden layer, feed-forward network that provides a unified learning platform with widespread types of feature mappings. It can also be applied in regression. This study compares results obtained using the standard Penman-Monteith method, ELM, artificial neural networks (ANNs), genetic programming (GP), and support vector machines (SVMs). Results suggest that ELM can predict ET more quickly and accurately than all other techniques tested. An ELM with sigmoid transfer function predicted ET with greater accuracy than a hard limit transfer function.