Prediction in ungauged regions with sparse flow duration curves and input-selection ensemble modeling

Streamflow data is crucial for calibrating hydrologic models which quantify the water cycle (Wada et al., 2017) for various purposes ranging from climate modeling (Allen & Ingram, 2002) to climate change impact mitigation (Trabucco et al., 2008), from water sustainability studies to flood forecasting and humanitarian aid (Coughlan de Perez et al., 2016). Scanning over the Global Runoff Data Centre's worldwide map showing where streamflow data has been tracked (GRDC, 2020), one cannot help but notice vast swaths of lands with very few streamflow gauges, e.g., Asia, South America, Oceania, Central America, Africa, and even parts of the southwestern USA (Figure S1 in Supporting Information). In countries like China and Ethiopia, daily observations of streamflow are being recorded, but unfortunately not made openly accessible due to various reasons. In many cases, historical data (prior to the 1990s) are available, but it is difficult to obtain high-quality meteorological forcing data from the same periods. Many regions also typically lack data on physiographic attributes such as soil and aquifer properties.