Monthly Streamflow Predictions for Danjiangkou Reservoir Using a Hybrid Model

A hybrid model coupling a simplified Variable Infiltration Capacity hydrological model with a CNN-GRU deep learning network predicts reservoir inflows. The model was applied to forecast 1-, 3-, and 6-month ahead inflows for China's Danjiangkou Reservoir. Results show improved Kling-Gupta efficiency values over standalone models and reduced computational burden for rolling predictions.

Use Cases

• Forecasting reservoir inflows based on bias-corrected GCM predictions and soil moisture data.
• Comparing hybrid model performance against standalone physical or deep learning models.
• Reducing computational load for operational streamflow prediction systems.
• Identifying key predictors for monthly streamflow from multiple data sources.

Strengths

• Model demonstrated improved Kling-Gupta efficiency for lead times up to 6 months.
• Hybrid approach reduced the computation burden of rolling prediction compared to a distributed hydrological model.

Limitations

• Row count and dataset size are unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Data is specific to a single reservoir; geographic generalizability is unverified.

Provenance

Source

Wuhan University

Collection Method

Model outputs from a coupled physically-based hydrological model and deep learning model, driven by bias-corrected GCM predictions.

Geography

Danjiangkou Reservoir, China