Population growth and economic development call for increased food production. Irrigated agriculture is one of the most vital food sources for billions of people worldwide. However, increasing demand for agricultural production and diminished freshwater resources imperil irrigated agriculture's sustainability. Reference crop evapotranspiration (ETo) is the gold standard for farm-level irrigation scheduling. Although reliable, ETo calculations require measurements from different sensors over a reference grass surface.Moreover, the dynamics of ETo and its meteorological driving factors are region- and climate-specific, and change because of climate change. This dissertation aims to leverage the wealth of publicly available data, advances in data science, and state-of-the-art machine learning models to estimate and forecast ETo. California, as one of the most hydrologically altered and agriculturally productive regions of the world, is chosen as the case study of this dissertation. The first project (Chapter 2) of this dissertation uses three feature importance measures to explore the relative influence of meteorological driving forces of ETo in different climatic zones of California. Moreover, this chapter analyzes the trends of ETo and its driving factors in California from 1986 to 2022. The findings of this project suggest that solar radiation and vapor pressure deficit are the most influential driving forces of ETo in California. The trend analysis also demonstrates that California's irrigation-oriented regions are getting hotter and drier, especially during the summer. The second project (Chapter 3) focuses on monthly ETo forecasting. An accurate monthly ETo forecast is essential for larger-scale water resources management. This chapter employs various forecasting models and strategies to investigate their forecasting accuracy, data efficiency, and computational cost. The findings of this project show that statistical forecasting models like Holt-Winters exponential smoothing work as accurately as the cutting-edge deep learning algorithms for monthly ETo forecasting. Moreover, this chapter reveals the lower data efficiency of most deep learning models compared to statistical models. The third project (Chapter 4) develops a generalizable machine learning-based approach to estimate daily ETo with only solar radiation data. This approach, called SolarET, relaxes the need for the reference surface, as solar radiation is the only driving factor of ETo which does not rely on measuring surface characteristics. It is shown in this chapter that SolarET works more accurately than Priestley-Taylor and Hargreaves-Samani in terms of daily ETo estimation. SolarET prediction accuracy is higher in irrigation-oriented regions of California, like the Central Valley than in coastal and desert regions.