Canopy temperature emulation in process-based models

Agricultural production relies heavily on the weather, making it especially sensitive to climate change. In the past, high temperatures have had substantial negative effects on crop yields. In a warming climate, these impacts could become even more severe.

Process-based modelling offers a systematic way to examine whether and how future environmental changes may impact crop yields. Many crop models take the 2 m air temperature as an input, allowing the simulated growth and development of the crops to respond directly to that temperature signal. However, depending on the climatic conditions, water status, and the developmental stage of a crop, 2 m air temperatures can be several degrees higher or lower than the actual temperatures at the canopy level. Some crop models therefore compute canopy temperatures based on complex energy balance approaches (EBSC), that have been shown to perform best compared to other approaches. However, these EBSC approaches are computationally expensive and their application in global models can therefore result in considerably higher runtimes. In our work, we developed resource efficient emulators that are based on an EBSC model and can be incorporated in global process-based models without significantly increasing the simulation time.

We applied the emulators in the agricultural modules of the dynamic global vegetation model LPJmL. The validation of daily maximum simulated canopy temperatures shows that LPJmL can reproduce cooling and heating effects of the canopy depending on the water and nitrogen availability of a crop compared to detailed site based observations in different locations throughout the US and Canada. For a global evaluation, we compared our results with skin temperatures from ERA5, which we used as an approximation for canopy temperatures. We show that, on a global scale and for daily maximum values, skin temperatures are significantly better represented by simulated canopy temperatures than by ERA5 2 m air temperatures.

Our results indicate that substituting simulated canopy temperatures for the 2 m air temperatures in processes driven by daily maximum temperatures improves the requirements of modelling heat stress impacts. Particularly, as high temperature processes often follow nonlinear dynamics and are even more affected by small temperature deviations. In a next step, we will use the further developed LPJmL model to analyze such heat stress impacts on crops. For this, we will include high-temperature responses, that will react to the newly implemented canopy temperatures.

The developed emulators can easily be included in other crop models, aiming to improve simulated temperature-dependent process dynamics. With this, we hope to provide a step towards reducing uncertainties in future agricultural yield estimates.