A new methodology for Standard Evapotranspiration classification over mesoscale regions: Application and evaluation of SOM over remotely sensed data

Standard Evapotranspiration (ETo) is an indicator of water losses given by evaporation and plant transpiration. Its quantification is particularly important for irrigation purposes, however in-situ data is not always accessible. This research aims to develop a methodology for Eto ‘weather’ classification through clustering Eto zones over Ecuador using remotely sensed data and an unsupervised learning algorithm. Thus, we obtained climatological variables from the Weather Research and Forecasting model  corresponding to years 2017, 2018, 2019, 2020, 2021. Following, we pre-processed the raw variables into eight parameters for Eto estimation, as in the Penman-Monteith equation, providing the model input variables for each year of study. Hence, we implemented a Self-Organizing Map (SOM) Artificial Neural Network over each dataset to obtain maps representing Eto clustered classes. Moreover, we tested the methodology's repeatability by applying SOM ten different times over each dataset and by applying the modified Cramers’ V-index to quantify the differences between map comparisons. Accordingly, we selected the SOM parameters that produced a Cramer’s V-index > 0.9  and differences between clustered maps < 0.0001. The outcomes of this research contribute to the classification of Eto ‘weather’ in mesoscale regions with future prospects to Eto ‘climate’ classification over larger temporal and spatial resolutions.