This paper advances hydrological management by harnessing social media data from Instagram, X, Flickr, and YouTube during Medicane Ianos (September 2020). The dataset included 7,915 texts, 2,949 photos, and 752 videos. Texts were classified into five disaster management categories: Ianos identification, Consequences, Disaster Management, Weather Information, and Emotions/Opinions. Classification used LSTM-RNN and transformers.Photos were categorized as related or unrelated using an ensemble of fine-tuned VGG-19, ResNet101, and EfficientNet models, boosting accuracy. Over 160,000 YouTube and 8,000 Instagram video frames were extracted, analyzed, and assessed via a Relevant Share Video Index (RSVI) to quantify content relevance.Location entity recognition (LER), geoparsing, geocoding, and GIS mapping spatially contextualized the data. Findings reveal that dataset size and classification complexity impact model performance, and custom epoch tuning optimizes accuracy-efficiency balance. LSTM-RNN outperformed transformers on the relatively small corpus. The GR-NLP-Toolkit’s LER performance on medicane texts provides further insights. Lastly, analysis shows most relevant posts appeared during the medicane’s active phase.