Estimating missing hourly climatic data using artificial neural network for energy balance based ET mapping applications

Abstract

Remote sensing based evapotranspiration (ET) mapping has become an important tool for water resources management at a regional scale. Accurate hourly climatic data and alfalfa-reference ET (ETr) are crucial inputs for successfully implementing remote sensing based ET models such as Mapping Evapotranspiration at High Resolution with Internalized Calibration (METRIC) and Surface Energy Balance Algorithm for Land (SEBAL). In Turkey, hourly climatic data may not be available at all locations, either due to cost constraints or due to equipment malfunctioning. In this study, the artificial neural network (ANN) technique was used to estimate missing and unmeasured hourly climatic data and ETr for the agriculturally important semi-humid Bafra plains located in northern Turkey. Modelled and actual climatic data were used to derive ET maps from two Landsat 5 Thematic Mapper images acquired on 2 September 2009 and 4 August 2010. The METRIC algorithm was used to generate ET maps. Accuracy assessment of the METRIC-derived ET maps indicated that climatic data and ETr estimated through ANN could be used for accurately mapping ET, where hourly climatic data are missing or not measured.