Determination of crop coefficient (kc) based on machine learning NDVI time series models

• Autor(a) principal: Duarte, Guilherme Filipe Tomé
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/27012
• topic: NDVI; Kc; Machine learning; Séries temporais -- Time series; Teledetecção -- Remote sensing

This dissertation aims to meet the required challenge of providing reflectance-based crop coefficient models to reduce water consumption in agriculture irrigation. In this work, 6 different models were created for each crop by using normalized difference vegetation index (NDVI) to estimate crop coefficients (Kc) for maize, tomato, potato and sunflower for Lezíria do Tejo region combining different pre-selection methods of time series and mean and k-means to create new time series and use linear and polynomial regression to fit the new generate time series with theoretical Kc curves to use these models to determine Kc in this region. These models’ performance was assessed using the coefficient of determination (R2), root mean square error (RMSE) and a visual inspection of test set predictions. The results show that the Kc-NDVI models created were able to capture the theoretical curves of Kc well, and the use of a pre-selection of time series, mean and k-means for these crops is useful to capture the curves of the crop coefficients since some of the best results obtained were when they were used. The best methodologies depend on each crop; no one is globally better than the others. The results shown are promising and can be seen as potential methods to better determine crop coefficients and the models are suitable for their use at least in the region of this study.