Importance of the memory effect for assessing interannual variation in net ecosystem exchange

Abstract

Terrestrial carbon flux shows greater interannual variation (IAV) than the ocean and drives most temporal changes in the atmospheric CO2 growth rate. Accurate estimation of the interannual variation in the terrestrial carbon budget is crucial for predicting changes in the carbon balance of ecosystems and atmospheric CO2 concentrations in the context of climate change. However, the interannual variation simulation of the terrestrial carbon budget has considerable uncertainty. One of the important reasons is the neglect of the memory effect of the ecosystem on climate change, namely the potential lagged impact of climate change. In the present work, we integrated meteorological and remote sensing vegetation data and introduced the memory effect into a deep learning model to simulate the global terrestrial net ecosystem exchange (NEE). The conclusions were as follows: (a) Considering the memory effect not only improved the simulation accuracy of the global terrestrial NEE, R2 increased by 0.37–0.55, RMSE decreased by 38.97%–52.42%, and MAE decreased by 39.77%–52.38%, but also improved the simulation accuracy of the interannual variation of NEE, R2 increased by 0.50–0.84, RMSE decreased by 37.13%–72.07%, and MAE decreased by 40.72%–66.49%. (b) The memory effect of NEE varied by ecosystem type, environmental factors, and plant development phase. The response of NEE to precipitation in most ecosystems was slower than that to temperature and radiation. The length of the comprehensive memory effect of NEE on antecedent environmental variables in the non-growth season was longer than that in the growth season. (c) NEE in the tropics has shown greater interannual variation than that in the high latitudes and dominated interannual variation in global terrestrial NEE. The tropical savanna exceeded tropical forest and contributed the largest fraction of the interannual variation of NEE in the tropics.