Abstract :
[en] Comprehensive understanding of the interactions between climate and vegetation is a key issue in environmental sciences, and especially for researchers studying climate change impacts on terrestrial ecosystems. Indeed in order to better predict changes in ecosystems productivity, scientists are investing time and e ort in assessing how environmental changes are influencing - and are going to influence in the near future - the vegetation distribution and dynamics.
Temperature, precipitation and atmospheric CO2 are the key determinants of the distribution of vegetation on Earth. Over the last 150 years, it has been reported that the global surface temperature has increased on average by around 0:8 C. Several studies mentioned that this rapid warming has resulted in reduction of climatic constraints to biological activity and shift in growing season. However changes in vegetation dynamics are not uniform spatially. From a methodological point of view, annual and seasonal metrics were commonly used to assess the impact of climate variability on vegetation at global and continental scales. The studies therefore neglected that intra-annual variability in the response of terrestrial ecosystems to such changes may exist. This intra-annual variability can be seen as the difference in vegetation response to a given environmental change according to its phenological development.
In this research we investigated the intra-annual variation of the climatic constraints over croplands and grasslands in 25 regions located in Europe and Africa. The central question was: how best can we identify the climate footprint on vegetation development during the growing season, using global datasets of Normalized Difference Vegetation Index (NDVI) and the JRC-MARS meteorological indicators? The structure of this study is as follows. First we provide an overview of major studies linking climate variability and vegetation dynamic at global, continental and regional scales. Then we describe the NDVI and meteorological datasets used in this research, as well as the methodology developed to select optimal regions of interest for the study of 'climate-vegetation' interactions at regional scale. Indeed external factors - such as land cover changes, landscape fragmentation, etc. - need to be minimized to ensure that the variations in the NDVI signal can be attributed to climate variability.
Preliminary time series analyses are then performed to characterize the long-term climate and vegetation conditions in each region of interest. We further present the approach developed in this research to decompose and to analyse jointly time series of remote sensing derived observation and climate dataset. We focus specifically on the adjustment of the 'climate-vegetation' relationships for specific periods within the growing season. Indeed we demonstrate that the relationship between NDVI and the meteorological parameters is highly complex and vary significantly trough the phenological cycle of the plants. Hence, interactions between vegetation dynamics and climate variability need to be studied at a smaller time scale than the year or the growing season, in order to identify properly the limiting factors to vegetation growth. Our analysis revealed that, in most of the cases, the best correlations are obtained when we considered the vegetative phase (from green-up to maximum of NDVI) and the reproductive phase (from maximum of NDVI to maturity) separately. We also show that climatic constraints identified using yearly proxies of climate and vegetation do not depict correctly, or completely, the climate control on vegetation development.
Finally we evaluate the performance of climate-driven growth models in two sites of croplands and two sites of grasslands. The models were adjusted per phenological phases and set to provide 1-month forecast of NDVI. Pure climatic models (CLIM) were compared to auto-regressive climatic model (CLIM-AR). Apart in the Irish grasslands, the CLIM-AR models were performing better than CLIM models during the vegetative phase. On the other hand, during the reproductive phase, the introduction of the auto-regressive term did not improve significantly the performance of the CLIM model. Moreover the autoregressive term did never appear as first predictor, demonstrating that, in the selected sites, short to medium atmospheric conditions were explaining most of the variance in the 1-month forecast NDVI.
