Using ensembles of adaptive neuro-fuzzy inference system and optimization algorithms to predict reference evapotranspiration in subtropical climatic zones

By D. Roy, Rahim Barzegar, John M. Quilty, Jan Adamowski. 2020

75 citations recorded by the source corpus.

Abstract Accurate prediction of reference evapotranspiration (ET0) is essential for efficient planning and management of limited water resources through proper irrigation scheduling. The FAO-56 Penman-Monteith approach to ET0 estimation was adopted to compute ET0 from data obtained during the period 2004–2019 in a subtropical climatic region in Bangladesh. Quantified ET0 values along with the meteorological variables for two other stations located in south Florida, USA were directly obtained from the USGS website. A commonly used machine learning algorithm, Adaptive Neuro Fuzzy Inference System (ANFIS), was employed to predict daily ET0 using regional meteorological data (e.g., daily maximum and minimum temperatures, wind speed, relative humidity, sensible heat flux, latent heat and sunshine duration). Four optimization algorithms were employed to tune ANFIS, resulting in hybrid models: Biogeography-based Optimization (BBO-ANFIS), Firefly Algorithm (FA-ANFIS), Particle Swarm Optimization (PSO-ANFIS), and Teaching-Learning-based Optimization (TLBO-ANFIS). These models’ performances were compared with the standard ANFIS model with parameters tuned using an integrated least squares and backpropagation gradient descent algorithm. Decision theories were applied to determine the reliability of the predictions as well as to rank prediction models based on eight statistical indices. Results indicated that FA-ANFIS resulted in the most accurate ET0 predictions. In the next stage, ensembles of these prediction models were compared to determine whether the ensemble models resulted in more reliable predictions than the standalone models. To develop each ensemble, three decision theories were applied: Shannon’s Entropy, Coefficient of Variation, and Grey Relational Analysis. All three ensemble approaches provided similar results, showing ensemble prediction methods to perform better than most individual models.

Subfields: Artificial Intelligence, Environmental Engineering, Global and Planetary Change

View Paper: https://www.semanticscholar.org/paper/21a78a2a998f4cb4c65956d8815c36b8d83029b1