Predicting precipitation δ 18O accurately is crucial for understanding water cycles, paleoclimates, and hydrological applications. Yet, forecasting its spatio-temporal distribution remains challenging due to complex climate interactions and extreme events. We developed a method combining spatio-temporal clustering and deep learning neural networks to improve multi-site, multi-year precipitation δ 18O predictions. Using a comprehensive dataset from 33 German sites (1978-2021), our model considers precipitation δ 18O and its controlling factors, including precipitation and temperature distribution. We applied the K-means++ method for classification and divided data into training and prediction sets. The CNN[1](#fn-0002) model extracted spatial features, while the Bi-LSTM model focused on temporal features. Spatio-temporal clustering using K-means++ improved forecast accuracy and reduced errors. This study highlights the potential of deep learning and clustering techniques for forecasting complex spatio-temporal data and offers insights for future research on isotope distributions.

The combined utilization of spatiotemporal clustering and deep learning neural network models were designed to evaluate the applicability of the multi-year and multi-sites precipitation δ18O forecasting method based on the precipitation isotope data of 10 stations in Germany from 1988 to 2012. In the overall forecasting, the performance of single-site multi-year forecasting is in the order of the Bi-directional Long Short-Term Memory (Bi-LSTM), CNN-Bi-LSTM, and the Convolutional Neural Network (CNN), with CNN-Bi-LSTM being the optimal model for multi-site multi-year forecasts. The seasonal forecasting does not demonstrate a significant improvement compared to the overall forecasting. For forecasting based on spatiotemporal clustering, cluster 1 improved accuracy, and cluster 2 improved error reduction and variance consistency. Nevertheless, the accuracy of forecasts depends solely on the amount of input data when the proportion of forecasting increases to a certain level. Overall, the seasonal forecasting is more appropriate for improving forecasting within a specific season, while spatiotemporal clustering can improve forecasting accuracy to some degree. In addition, optimal solutions exist for the type and number of model clusters. In terms of model types, CNN-Bi-LSTM generally has better forecasting performance than CNN and Bi-LSTM.