Using a physics-constrained machine learning model to predict canopy interception

Abstract:

Canopy interception (CI) is a key component of the hydrological cycle which can impacts both water resource management and climate change. We developed a physics-constrained machine learning model (hybrid model) to estimate canopy interception. Our results show that the hybrid model proposed in this study can accurately quantify CI only using fraction of photosynthetically active radiation (fpar), air temperature (Ta), carbon dioxide concentration (Ca), wind speed (WS), relative humidity (RH), photosynthetically active radiation (PAR) and plant function type (PFT) as input, compared to the observations. This method can further provide a good understanding of the drivers of canopy interception. In addition, the fraction of evapotranspiration (ET) due to rain interception is also obtained. More importantly, a global map of canopy interception can be more easily retrieved based on the hybrid model. This kind of hybrid model can successfully reproduce canopy interception – a problem that had long plagued our understanding of land-atmosphere interactions, global water cycle and hydrological modelling.