Application of Artificial Neural Networks to Project Reference Evapotranspiration Under Climate Change Scenarios

Evapotranspiration is sensitive to climate change. The main objective of this study was to examine the response of reference evapotranspiration (ET0) under various climate change scenarios using artificial neural networks and the Canadian Earth System Model Second Generation (CanESM2). The Hargreaves method was used to calculate ET0 for western, central, and eastern parts of Prince Edward Island using their two input parameters: daily maximum temperature (Tmax), and daily minimum temperature (Tmin). The Tmax and Tmin were downscaled with the help of statistical downscaling model (SDSM) for three future periods 2020s (2011-2040), 2050s (2041-2070), and 2080s (2071-2100) under three representative concentration pathways (RCP’s) including RCP 2.6, RCP P4.5, and RCP 8.5. Temporally, there were major changes in Tmax, Tmin, and ET0 for the 2080s under RCP8.5. The temporal variations in ET0 for all RCPs matched the reports in the literature for other similar locations. For RCP8.5, it ranged from 1.63 (2020s) to 2.29 mm/day (2080s). As a next step, a one-dimensional convolutional neural network (1D-CNN), long-short term memory (LSTM), and multilayer perceptron (MLP) were used for estimating ET0. High coefficient of correlation (r > 0.95) values for both calibration and validation periods showed the potential of the artificial neural networks in ET0 estimation. The results of this study will help decision makers and water resource managers in future quantification of the availability of water for the island and to optimize the use of island water resources on a sustainable basis.