A convolutional neural network model for accurate short-term leaf area index prediction

The leaf area index (LAI) is a crucial biophysical parameter that significantly influences carbon, water, and energy cycles within terrestrial ecosystems. While short-term LAI prediction has been extensively studied, most research has primarily focused on specific ecosystem types. This comprehensive study evaluates the performance of the convolutional neural network (CNN) model across eleven diverse land cover types within global terrestrial ecosystems. Our results reveal the promising predictive capabilities of the CNN model, achieving an overall RAcirc;(2) of 0.845 and RMSE of 0.301, outperforming all the other baseline models. Notably, seasonal analysis demonstrates higher prediction accuracy (lower SMAPE) during summer than winter for most studied land cover types. We further identify radiation as a key environmental factor influencing LAI prediction accuracy across various land cover types. Overall, this research contributes to advancements in short-term LAI prediction, highlighting the efficacy of the tested deep learning models in time-series ecological modeling. These findings have broad implications for climate change modeling, resource management, and agricultural planning.