HydroAgent Dataset Calibration assets for the HydroAgent LLM agent — everything the agent (and the EF5/CREST hydrologic simulator it controls) needs to calibrate streamflow at a USGS gage: CONUS terrain & default parameter rasters ( basic data/ ) Per gage MRMS hourly precipitation clips ( data mrms clip/ ) Hourly USGS streamflow observations ( gauge/ ) Daily potential ET rasters ( pet/ ) EF5 control file template ( docs/ ) 73 GPT 4o calibration trajectories across 29 gages ( sets for SFT RL/ ) — used as the SFT distillation source for HydroAgent This dataset is the canonical training/eval bundle for the HydroLLM repository. Layout Per gage simulation assets ( gauge , data mrms clip , …) The 14 gages with shipped MRMS clips are the training + evaluation set of the GRPO loop in HydroAgent. All windows are inside MRMS coverage (2017 10 01 → 2019 12 31) and were selected by scripts/audit flood windows.py to contain a clear flood event with rising and receding limbs and a buffer from the window edge. Training gages (10): Gage ID Basin (km²) Lat Lon Window (UTC) : : : 11383500 539 40.0140 121.9483 2018 05 19 → 2018 07 17 11043000 575 33.4798 117.1439 2019 03 15 → 2019 05 13 11152000 632…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy