This study explores the potential of Long-Short-Term-Memory (LSTM) networks for rainfall-runoff modelling using the CAMELS dataset, which includes meteorological data and observed discharge for 671 catchments across the contiguous United States. The LSTM, a type of recurrent neural network, is capable of learning long-term dependencies, making it suitable for capturing storage effects in catchments, especially those influenced by snow. The study compares the performance of the LSTM with the Sacramento Soil Moisture Accounting Model (SAC-SMA) coupled with the Snow-17 snow routine. In the first experiment, the LSTM was trained separately for each of the 241 catchments, achieving results comparable to the SAC-SMA + Snow-17 model. The LSTM performed well in snow-influenced catchments and outperformed the benchmark model in arid catchments. In the second experiment, a single LSTM was trained to predict discharge for all catchments within a hydrological unit (HUC), showing improved performance in HUCs with similar hydrological characteristics. However, performance varied when catchments within a HUC exhibited different behaviors. The third experiment tested the effectiveness of fine-tuning the regional model for individual catchments, resulting in improved performance for basins where the regional model initially performed poorly. The results indicate that the LSTM can effectively model rainfall-runoff processes, with the regional model showing improved performance in certain conditions. The study highlights the potential of LSTMs as a regional hydrological model, capable of transferring knowledge from a larger dataset to individual catchments, thereby enhancing model performance. The LSTM outperformed the SAC-SMA + Snow-17 model in several cases, demonstrating its potential for hydrological modelling applications. The study concludes that LSTMs are a promising tool for rainfall-runoff modelling, particularly in scenarios where traditional models may struggle due to computational constraints or data limitations.This study explores the potential of Long-Short-Term-Memory (LSTM) networks for rainfall-runoff modelling using the CAMELS dataset, which includes meteorological data and observed discharge for 671 catchments across the contiguous United States. The LSTM, a type of recurrent neural network, is capable of learning long-term dependencies, making it suitable for capturing storage effects in catchments, especially those influenced by snow. The study compares the performance of the LSTM with the Sacramento Soil Moisture Accounting Model (SAC-SMA) coupled with the Snow-17 snow routine. In the first experiment, the LSTM was trained separately for each of the 241 catchments, achieving results comparable to the SAC-SMA + Snow-17 model. The LSTM performed well in snow-influenced catchments and outperformed the benchmark model in arid catchments. In the second experiment, a single LSTM was trained to predict discharge for all catchments within a hydrological unit (HUC), showing improved performance in HUCs with similar hydrological characteristics. However, performance varied when catchments within a HUC exhibited different behaviors. The third experiment tested the effectiveness of fine-tuning the regional model for individual catchments, resulting in improved performance for basins where the regional model initially performed poorly. The results indicate that the LSTM can effectively model rainfall-runoff processes, with the regional model showing improved performance in certain conditions. The study highlights the potential of LSTMs as a regional hydrological model, capable of transferring knowledge from a larger dataset to individual catchments, thereby enhancing model performance. The LSTM outperformed the SAC-SMA + Snow-17 model in several cases, demonstrating its potential for hydrological modelling applications. The study concludes that LSTMs are a promising tool for rainfall-runoff modelling, particularly in scenarios where traditional models may struggle due to computational constraints or data limitations.