Neural Network Analysis to Evaluate Ozone Damage to Vegetation Under Different Climatic Conditions

Tropospheric ozone (O3) is probably the air pollutant most damaging to vegetation. Understanding how plants respond to O3 pollution under different climate conditions is of central importance for predicting the interactions between climate change, ozone impact and vegetation. This work analyses the effect of O3 fluxes on net ecosystem productivity (NEP), measured directly at the ecosystem level with the eddy covariance (EC) technique. The relationship was explored with artificial neural networks (ANNs), which were used to model NEP using environmental and phenological variables as inputs in addition to stomatal O3 uptake in Spring and Summer, when O3 pollution is expected to be highest. A sensitivity analysis allowed us to isolate the effect of O3, visualize the shape of the O3-NEP functional relationship and explore how climatic variables affect NEP response to O3. This approach has been applied to eleven ecosystems covering a range of climatic areas. The analysis highlighted that O3 effects over NEP are highly non-linear and site-specific. A significant but small NEP reduction was found during Spring in a Scottish shrubland (−0.67%), in two Italian forests (up to −1.37%) and during Summer in a Californian orange orchard (−1.25%). Although the overall seasonal effect of O3 on NEP was not found to be negative for the other sites, with episodic O3 detrimental effect still identified. These episodes were correlated with meteorological variables showing that O3 damage depends on weather conditions. By identifying O3 damage under field conditions and the environmental factors influencing to that damage, this work provides an insight into O3 pollution, climate and weather conditions.