Data and code: catchment similarity and parameter transfer for ungauged catchments

# Data and code: catchment similarity and parameter transfer for ungauged catchments This package contains the code and the derived data for the study "How should catchment similarity be measured and model parameters be transferred in regionalisation for ungauged catchments?". The study uses a modified Xinanjiang model with degree-day snow in 671 CAMELS catchments in the contiguous United States. It compares process-wise parameter mixing (MWSA), similarity based on within-catchment distribution descriptors, conventional regionalisation methods, and learned donor regionalisation (LDR), which learns donor selection from the KGE actually achieved when parameters are transferred between gauged catchments. ## Contents | Folder / file | Contents | | --- | --- | | `code/` | Python scripts for calibration, descriptors, all regionalisation methods, the evaluations and the figures. See `code/README.md` | | `data/` | Catchment lists, calibrated parameters, descriptors, trained models and all results reported in the paper. See `data/README.md` | | `README.md` | This file | ## Catchments and periods - 463 CAMELS catchments used by Shen et al. (2022), split into 305 training and 158 testing catchments (stratified by HUC02 region and aridity tercile, seed 2026). - 208 further CAMELS catchments used only as independent catchments. - Calibration period 1980-10-01 to 1995-09-30, validation period 1995-10-01 to 2010-09-30, extended validation period 2010-10-01 to 2014-12-31. ## External data The package does not redistribute third-party data. The scripts read the following public data sets; `code/README.md` lists the exact files and where to put them. - CAMELS: forcing, streamflow, attributes, catchment boundaries and SAC-SMA benchmark output (Newman et al., 2015; Addor et al., 2017). - HydroSHEDS 15″ DEM and flow accumulation (Lehner et al., 2008). - MODIS MOD13C1 0.05° NDVI (Didan, 2015). - POLARIS soil properties, 0–5 cm (Chaney et al., 2019). With the derived data in `data/`, every table and figure of the paper can be recomputed without rerunning the simulations. This needs only the CAMELS attribute tables (a few MB). Fig. 1 also needs the CAMELS catchment boundaries and US state boundaries. Rerunning the simulations needs the CAMELS forcing and streamflow, and recomputing the descriptors needs the gridded data. ## Quick start ``` cd code pip install -r requirements.txt set CAMELS_DIR=<folder with the CAMELS attribute tables> (Linux/macOS: export CAMELS_DIR=...) python make_figures.py ../figures 3 4 5 6 7 8 9 10 # redraws the figures from data/ into figures/ ``` Figure 1 also needs `CAMELS_SHP` and `STATES_SHP`, and Fig. 2 is a schematic drawn without data (see `code/README.md`). ## Licence Code and data are released under the Creative Commons Attribution 4.0 International licence (CC BY 4.0). ## Contact Jun Guo, Email: [e-mail]