Deriving Global Parameter Estimates for the Noah Land Surface Model using FLUXNET and Machine Learning

摘要: With their origins in numerical weather prediction and climate modeling, land surface models aim to accurately partition the energy balance. An overlooked challenge these schemes is role of model parameter uncertainty, particularly at unmonitored sites. This study provides global estimates for Noah using 85 eddy covariance sites FLUXNET network. The at-site parameters are first calibrated a Latin Hypercube based ensemble most sensitive parameters, determined by Sobol method, be minimum stomatal resistance (rs,min), Zilinkitevich empirical constant (Czil), bare soil evaporation exponent (fxexp). Calibration leads an increase mean Kling-Gupta Efficiency (KGE) performance metric from 0.54 0.71. These sets then related local environmental characteristics Extra-Trees machine learning algorithm. fitted used map optimal over globe 5 km spatial resolution. leave-one-out cross-validation mapped suggests that there potential skillfully relate characteristics. results demonstrate use tune parameterizations fluxes provide improved globe.