First global SWAT+ model runs at 2 km but river flow is weak without calibration

Why it matters

A global SWAT+ model could link water quantity, water quality and land management in one high-resolution tool. The open-source workflow lets others rebuild and extend the setup. The results show what must come next: calibration, reservoirs, irrigation and better snow handling.

What they did

The authors built CoSWAT, a Python workflow that downloads and prepares elevation, land use, soil and climate data and writes SWAT+ model files, using parallel processing. They set up the world as regions based on major river basins at 2 km resolution, giving 2.63 million HRUs. The model was run without calibration for 1977–1990 with a 5 year warm-up. They checked evapotranspiration against GLEAM v4 and monthly river discharge against GRDC gauge data.

Key findings

• Evapotranspiration maps from SWAT+ and GLEAM v4 had broadly similar spatial patterns for 1982–1990.
• 78.54 % of sampled points had evapotranspiration differences within ± 100 mm of GLEAM.
• Only 23.02 % of gauging stations had positive Kling–Gupta Efficiency for monthly river discharge.
• 85.31 % of stations showed positive correlation, and the median variability and bias ratios were close to 1.
• Setup took about 12 min for the Save Basin and about 1 h 49 min for the Nile Basin on a 64-core machine.

Limitations

• The model was not calibrated, and it has no reservoirs, irrigation or other water management, which hurt river flow results.
• Too much snow builds up at high latitudes and in mountains, and lakes are not represented, causing evapotranspiration artefacts; the 0.5° climate data is coarse.
• Evaluation of evapotranspiration is limited to one remote-sensing product, since eddy covariance data are scarce, and no direct comparison with other global models was made.

Glossary

• SWAT+: A water and land model that simulates water, sediment and nutrients over units called hydrologic response units.
• HRU (hydrologic response unit): A model unit with a unique mix of land use and soil within a subbasin.
• Kling–Gupta Efficiency (KGE): A score combining correlation, variability ratio and bias ratio between simulated and observed flow; values above 0 are the better range here.
• GLEAM: A dataset of evapotranspiration estimates used here as the reference.