Escalating water scarcity, increasing demand, and rising costs require residents of semi-arid urban areas to make informed decisions on water management. Outdoor landscaping typically represents 40–70% of total water household consumption in some semi-arid areas; thus, outdoor water use is a significant factor that contributes to urban water sustainability. Evapotranspiration (ET) indicates the water and energy exchange between the atmosphere and the soil. This interaction is interdependent on solar radiation, evaporation, transpiration, and other biophysical parameters. ET has become a reference for water management in agriculture (e.g., crop irrigation). However, these water management models do not typically include outdoor vegetation. Advances in Artificial Intelligence (AI) provide promising alternatives for Reference Evapotranspiration (ETo) prediction. This work aims to create an ETo model for small-scale outdoor irrigation to support water and vegetation management in semi-arid urban households. Four Artificial Neural Networks (ANN) models, Multilayer Perceptron (MLP), Long-Short Term Memory (LSTM), Gated Recurrent Unit (GRU), and Convolutional-LSTM (Conv-LSTM), were implemented and trained on ETo using meteorological data calculated with a modified Penman-Monteith equation from 33 stations. The ANN models were compared against four benchmark models: linear regression (LR), XGBRegressor (XGBR), support vector regressor (SVR), and random forest regressor (RFR). All models were evaluated using metrics such as R-squared (R2), mean squared error (MSE), root mean squared error (RMSE), and mean absolute percentage error (MAPE). From the ANN models, the MLP network produced the best data fit with an R2 of 0.9665 on the test data, followed by the Conv-LSTM, LSTM, and GRU with R2 of 0.8412, 0.8202, and 0.8069, respectively. However, the regression models show a slightly better performance than MLP regarding R2.