NARX model identification for analysing Amazon vegetation under climate change

The Amazon rainforest is a critical landscape and harbours a wide range of biodiversity. This is considered to be one of the largest sink for anthropogenic carbon sequestration on the Earth. Undoubtably, any substantial variation in the vegetation of this basin have tremendous impact upon the carbon absorption. Nonetheless, the impact of changing climate on the Amazon rainforest further complicates the matter. This study, for the first time, utilises the system identification method, under a wider realm of machine learning, for modelling the nonlinear dynamical relationship among the Leaf Area Index (LAI) and surface temperature for an Amazon rainforest site. The chosen model structure is Nonlinear Autoregressive with Exogenous Inputs (NARX). The training and testing datasets involved in this study correspond to the NASA Earth Observations. On contrary to the existing modelling methods performed for the Amazon, this data driven method results into a parsimonious model structure consisting of autoregressive terms as well as time lagged surface temperature. It therefore gives a deeper insights about the effects of temperature variation on the Amazon vegetation, emboldening the potential management of this crucial rainforest. A temperature dependent model also facilitates the forecasting under the various scenarios of the Intergovernmental Panel on Climate Change (IPCC).