Modelling vegetation dynamics for future climates in Australian catchments: Comparison of a conceptual eco-hydrological modelling approach with a deep learning alternative

Abstract

Dynamically simulating leaf area index assists in modelling the feedbacks between eco-hydrologic and climatic processes. The particular challenge for Australia is the prevalence of arid and semi-arid ecosystems where water availability plays a crucial role in vegetation productivity. To understand whether existing LAI models can capture plant dynamics under changing climates, we tested two competing models across Australia's different climate zones: a conceptual eco-hydrologic model that applies water use efficiency term to relate LAI to water uptake, and a deep learning approach. An initial virtual catchment experiment with deep learning showed that it only uses information from potential evapotranspiration. For future climates, the conceptual model captured a negative trend and increasing variance in LAI, which is plausible given projected rainfall changes, while deep learning did not. Our study demonstrated an example of ‘right answer for the wrong reasons’, and the importance of incorporating knowledge of water-carbon coupling for appropriate scenarios.