Semi-Supervised Learning Model for Streamflow Prediction

This repository include the model code for the following paper (under review) Streamflow prediction in data-scarce regions with semi-supervised deep learning (preprint in XXX) In this study, we propose a two-stage semi-supervised learning (SSL) for streamflow prediction in data-scarce regions, based on the Contrastive Predictive Coding (CPC) method. The CPC pre-training code, and fine-tuning code for streamflow prediction are in CPC2HY.zip and Hydro_SSL.zip file, respectively. The model code will be made available after acceptance of this paper. Here, we present the notebooks for model training and testing, and for the subsequent analyses. The CAMELS-DE dataset is freely available and can be downloaded to run the notebooks and recreate the results from this study. The results are not uploaded here, but can be made available upon request.