Scientists introduce AutoCF, an automated ecosystem that brings together simulation, testing, and impact analysis for compound coastal flooding. Compound coastal flooding happens when storm surge, rain, and river flow interact to cause floods. AutoCF aims to run the whole workflow in one reproducible pipeline and uses large language model (LLM) assistance to help automate parts of that pipeline.

The authors built AutoCF to handle data harmonization, model construction, observational evaluation, exposure analysis (who and what is flooded), a complete factorial driver attribution (testing combinations of storm surge, precipitation, and river drivers), and execution across different computing platforms. They used the system to run an automated simulation of Hurricane Harvey and compared model results to observations. Across eight monitoring gauges the model had a median root mean square error of 0.147 meters and a correlation of 0.951. The model also correlated 0.942 with 55 high-water marks (locations where flood height was recorded after the storm).

AutoCF was tested on both central processing unit (CPU) and graphics processing unit (GPU) implementations. The two implementations produced very similar maximum water-level fields: for 98.8% of model cells the maximum water-levels agreed within 0.01 meters. This cross-platform agreement supports the reproducibility of the simulations.

The system also supports attribution — figuring out which drivers caused the flooding and its consequences. For Hurricane Harvey, the analysis found that precipitation (rain) was the main driver of building and population exposure. In contrast, coastal forcing (storm surge) mattered more for deep or long-lasting flooding. The paper also introduces a Driver Impact Shift metric that measures whether a driver had a bigger effect on people or buildings than its share of flooded area would suggest.