Integrating social media data with remote sensing technologies for disaster detection involves several technical approaches to address key challenges. Data fusion techniques like multi-modal fusion, which combines outputs from deep learning models for images and text, have proven effective, along with natural language processing (NLP) for structuring unstructured social media data. To enhance geospatial accuracy, advanced geocoding and crowdsourced validation can help align social media geolocations with high-precision satellite imagery. Temporal alignment can be achieved through time series analysis and event detection algorithms that correlate social media posts with satellite acquisition times, while data interpolation methods can estimate gaps in temporal data. Case studies, such as CrisisMappers and NASA’s DISASTERS, illustrate successful applications of these approaches, providing valuable insights for improving situational awareness during disasters.

Google Scholar is your friend. Cal Fire ( https://www.fire.ca.gov/ ) is the research epicenter ( not surprising considering where Silicone Valley is located and the state's susceptibility to every form of human and natural disaster ). I participated in a hackathon sponsored by them and the US DoD on this theme. But one recurrent theme was that all the typical pathologies present in social media become magnified in an emergency situation resulting in a low signal to noise ratio.