Machine learning-based prediction and assessment of recent dynamics of forest net primary productivity in Romania

Abstract

While the analysis of spatio-temporal changes in the net primary productivity (NPP) of forests can provide critical information on carbon cycle and climate change, these ecological trends have remained unclear in many countries worldwide, including Romania. By using complex (satellite, forest and climate) data, many sophisticated (machine learning) algorithms and some widely applied (the Mann-Kendall test and Sen's slope estimator) statistical procedures, this study investigates, for the first time, recent forest NPP trends (1987-2018) that occurred in Romania, in relation to climate change that affected the country over the past decades. Following the modelling, mapping and assessment of NPP dynamics, results showed almost exclusively positive trends for this ecological parameter, which accounts for ∼99% of all forest NPP changes that occurred throughout the country, after 1987. Interestingly, almost three quarters (∼73%) of all NPP increasing trends are statistically significant, which indicates that Romania's forests have recently experienced a large-scale improvement in carbon fluxes and stocks. Investigations of eco-climatic relationships suggest that climate change has partially contributed to these surprising NPP dynamics observed in recent decades. All these findings can provide valuable information for forest management and for many stakeholders and policymakers who operate in the forestry and climate fields in Romania.

Keywords: Carbon fluxes; Climate change; Geostatistical modelling; Machine learning; NPP; Romania; Spatio-temporal trends.