Potato crops are one of the fundamental productive chains for the economic development of Boyacá region, due to its importance in family nutrition and food security. The objective of this work was to improve the management of the water resources in a potato crop by applying deep learning algorithms to predict the irrigation prescription. Three deep learning models, such as one-dimensional convolutional neural network (1-DCNN), long short-term memory (LSTM), and a hybrid convolutional LSTM were reviewed and evaluated. The climatic variables of the potato crop were collected daily for three years, such as temperature, rain, water content in soil, and evapotranspiration, among others. The measured variables were provided by two meteorological stations located in the USoChicamocha Irrigation District. Deep learning models were trained and validated using Python®, calculating performance metrics. The CNN-LSTM model predicted the potato crop irrigation prescription with better precision in the training and validation dataset with MSE values less than 0,067 and RMSE values less than 0,258. The achievement of the CNN-LSTM algorithm had the highest coefficient of determination, obtaining a score of R² = 0,96. The proven deep learning techniques made it possible to predict the precision irrigation prescription in the studied potato crop, significantly improving the management of water resources. Deep learning techniques have served as a support instrument to assist farmers in decision-making and efficient management of water resources in irrigation planning at the farm level according to the measurement of crop variables.