Catchment scale runoff time-series generation and validation using statistical models for the Continental United...

Area of Science:

• Hydrology
• Environmental Science
• Statistical Modeling

Background:

• Accurate runoff estimation is crucial for water resource management.
• Existing methods often lack fine-scale spatial and temporal resolution.
• The National Hydrography Dataset Plus Version 2 (NHDPlusV2) provides a valuable framework for catchment-scale analysis.

Purpose of the Study:

• To develop and validate statistical models for generating accurate runoff time-series at the NHDPlusV2 catchment scale across the Continental United States (CONUS).
• To assess the effectiveness of Normalized Difference Vegetation Index (NDVI) based Curve Number (CN) for initial runoff estimation.
• To improve runoff prediction accuracy through statistical model correction.

Main Methods:

• Utilized 17 years of NDVI data to calculate 23 NDVI-based CN (NDVI-CN) values for 2.65 million NHDPlusV2 catchments.
• Employed North American Land Data Assimilation System 2 (NLDAS-2) runoff time-series as reference data for training and validation.
• Developed a spatio-temporal cross-validation framework to ensure robust model evaluation.

Main Results:

• Simple linear regression models effectively corrected NDVI-CN runoff in many CONUS physiographic sections, achieving Nash-Sutcliffe Efficiency values above 0.5.
• The models demonstrated high predictive accuracy for runoff time-series generation.
• Model performance was significantly reduced in physiographic sections with substantial snow accumulation.

Conclusions:

• Statistical models integrating NDVI-CN show strong potential for accurate catchment-scale runoff prediction in CONUS.
• The spatio-temporal cross-validation framework provides reliable model assessment.
• Snow accumulation presents a limitation for the current model's performance, highlighting areas for future research.