Final Report Summary - EUROSAT4PHENOCHANGES (Monitoring vegetation phenology at multiple scales in Europe from the GMES satellite sensor time-series: a special consideration to natura2000 areas)
Work progress and achievements during the period:
1. Progress towards objectives and details for each task
The overall objective of this project is to develop an accurate methodology based on satellite sensor imagery for studying and mapping the phenological response of European terrestrial vegetation to climate drivers. To this end, the following tasks were carried out:
Task 1: Remote sensing data acquisition and pre-processing
MERIS MTCI time-series data at medium (1 km) and fine spatial resolutions (300 m) were acquired and used to construct multi-temporal records of vegetation index values for the whole European continent, and the Iberian Peninsula for the last decade. To this aim, the research team worked in close collaboration with the company Astrium Services.
The time-series was successfully processed applying three major procedures: data cleaning and flagging, data smoothing and temporal base information extraction. A data cleaning and flagging procedure was used to remove missing data values from the original data and create a flag depending upon the quality of the temporal information available in each image pixel. The errors in the raw time-series were minimized by applying a Fourier smoothing algorithm.
Although in the proposal of the project we proposed to use and compare different vegetation indices (NDVI, LAI, FAPAR and MTCI), given that they were going to be available within the framework of biopar2 according to the information the European Union. We were informed on 25/06/2013 by the company supplier of the data (Astrium Services) that the processing of these time-series was cancelled. Therefore, this comparison is no longer possible, because the processing of 4 vegetation index products by our research team would be very time consuming and it will exceed the duration of the project. However, the processing of MTCI time-series has been prioritized as it is the most promising product. Time-series of MTCI were retrieved and composited from the beginning in close collaboration with Astrium Services. This is a deviation from the original plan because we built our proposal on the basis of the availability of different time-series of biophysical products which were going to be available since the beginning of the project.
Task 2. Phenological variables extraction and mapping
After the pre-processing described above, three phenological parameters (onset of greenness, OG; end of senescence, ES; and length of the season, LS) were estimated using an iterative search process over the annual smoothed data. These variables were mapped for the whole European continent at a 1 km spatial resolution, and the Iberian Peninsula at a finer resolution equal to 300 m.
Since vegetation phenology varies with different land covers and bio-geographical regions, the phenology of the main natural land covers (deciduous forest, evergreen forest, grasslands and shrublands) for the different European bio-geographical regions were characterised to establish a baseline to assess the impact of past and future environmental changes. This aspect was critical because many previous studies have not stratified phenological characteristics by land cover class and biogeographical region in Europe (see Rodriguez-Galiano et al. 2015 – Remote Sensing).
Our approach has established a potential reference for evaluating the susceptibility of vegetation to global climatic changes, and allows comparison of how different regions could respond under climatic conditions. The phenology maps produced here represent one of the most comprehensive and recent assessments of the land surface phenology of continental Europe at a 1 km spatial resolution. The spatial variability of phenology can be used for tuning models of carbon emissions and can play an important role in monitoring programmes for the management of natural vegetation.
Task 3. Validation of the phenology characterisation
The spatio-temporal reliability of the LSP estimates described above were assessed using ground phenology observations for a large number of points of different deciduous tree plant species collected across Europe within the framework of the PEP725 ground phenology (GP) network (http://www.pep725.eu/(opens in new window)). Correlations observed between the interannual time series of the satellite sensor estimates of phenology and PEP725 records revealed a close agreement (especially for Betula Pendula and Fagus Sylvatica species). In particular, 90% of the statistically significant correlations between LSP and GP were positive (mean R2 = 0.77). A large spatiotemporal correlation was observed between the dates of the start of season (end of season) from space and leaf unfolding (autumn coloring) at the ground (pseudo R2 of 0.70 (0.71)) providing, for the first time, the ability to predict accurately the date of leaf unfolding (autumn coloring) across Europe (root-mean-square error of 5.97 days (6.75 days) over 365 days).
A new methodology to relate LSP and GP has been proposed in this project through the application of Random Forest, a nonparametric method which allows for nonlinear relationships between phenology variables and for the inclusion in the modeling process of plant species. The proposed method can recognize complex patterns between LSP and the phenology of multiple specific plant species, integrating them into a unique overall model, rather than generating multiple models for every species. Additionally, it is data driven, which means that there is no need to incorporate previous knowledge about the species composition in the landscape, but it considers the different plant species of GP (see Rodriguez-Galiano et al 2015 – Geophysical Research Letters).
Task 4. Modelling of anomalies in phenology using climatic data.
Understanding the effect of inter-annual climatic variation on LSP is an essential step to establish a plausible link between recent climate variability and vegetation phenological responses at global or regional scales, and importantly to make reliable forecasts about future vegetation responses to different future climatic scenarios. Our aim was, therefore, to provide an explanation of the observed anomalies in LSP of the entire European forest during the last decade, identifying the main climatic drivers of spring and autumnal LSP at the continental scale applying new methodologies.
In Rodriguez-Galiano et al 2015 – Biogeosciences we modelled the climate-driven anomalies in phenology, rather than trends, and using innovative multivariate non-linear machine learning techniques to evaluate multiple climatic predictors at biological scales, and non-climatic predictors such as the legacy effect of the date of spring onset in leaf senescence. Climate predictors used range from monthly average values of temperature (max, min and avg), precipitation, short wave radiation and day length; trimestral cumulated values such as growing degree days or chilling requirements, among others; to the date of specific events such as the first freeze or the last freeze. Moreover, we considered flexible biological time scales in the analysis between climatic and phenological events rather than fixed calendar dates.
This part of the project revealed new insights into the climatic drivers of anomalies in LSP across the entire European forest, while at the same time established a new conceptual framework for predictive modelling of LSP. Specifically, the Random Forest method, a multivariate, spatially non-stationary and non-linear machine learning approach, was introduced for phenological modelling across very large areas and across multiple years simultaneously: the typical case for satellite-observed LSP. The RF model was fitted to the relation between LSP anomalies and numerous climate predictor variables. The models explained 81% and 62% of the variance in the spring and autumn LSP anomalies, with relative errors of 10% and 20%, respectively: a level of precision that has until now been unobtainable at the continental scale. A further comparison with a linear regression analysis suggested that there might be a non-linear relationship between the anomalies in LSP and the climatic drivers. It also allowed identification of the main drivers of the anomalies in LSP through its estimation of variable importance. Our results revealed that the accuracy of the broadly applied growing degree day-based models is overestimated using linear regression models and that non-linear multivariate relationships between temperature (especially minimum temperature) and radiation are needed to describe the relations between phenology and climatic drivers. We would like to highlight importance of minimum temperatures as a better indicator of climatic changes than either the average, maximum temperature, or growing degree days.
Our findings, thus, showed clearly the inadequacy of the hitherto applied linear regression approaches for modelling LSP and paves the way for a new set of scientific investigations based on machine learning methods.
1. Progress towards objectives and details for each task
The overall objective of this project is to develop an accurate methodology based on satellite sensor imagery for studying and mapping the phenological response of European terrestrial vegetation to climate drivers. To this end, the following tasks were carried out:
Task 1: Remote sensing data acquisition and pre-processing
MERIS MTCI time-series data at medium (1 km) and fine spatial resolutions (300 m) were acquired and used to construct multi-temporal records of vegetation index values for the whole European continent, and the Iberian Peninsula for the last decade. To this aim, the research team worked in close collaboration with the company Astrium Services.
The time-series was successfully processed applying three major procedures: data cleaning and flagging, data smoothing and temporal base information extraction. A data cleaning and flagging procedure was used to remove missing data values from the original data and create a flag depending upon the quality of the temporal information available in each image pixel. The errors in the raw time-series were minimized by applying a Fourier smoothing algorithm.
Although in the proposal of the project we proposed to use and compare different vegetation indices (NDVI, LAI, FAPAR and MTCI), given that they were going to be available within the framework of biopar2 according to the information the European Union. We were informed on 25/06/2013 by the company supplier of the data (Astrium Services) that the processing of these time-series was cancelled. Therefore, this comparison is no longer possible, because the processing of 4 vegetation index products by our research team would be very time consuming and it will exceed the duration of the project. However, the processing of MTCI time-series has been prioritized as it is the most promising product. Time-series of MTCI were retrieved and composited from the beginning in close collaboration with Astrium Services. This is a deviation from the original plan because we built our proposal on the basis of the availability of different time-series of biophysical products which were going to be available since the beginning of the project.
Task 2. Phenological variables extraction and mapping
After the pre-processing described above, three phenological parameters (onset of greenness, OG; end of senescence, ES; and length of the season, LS) were estimated using an iterative search process over the annual smoothed data. These variables were mapped for the whole European continent at a 1 km spatial resolution, and the Iberian Peninsula at a finer resolution equal to 300 m.
Since vegetation phenology varies with different land covers and bio-geographical regions, the phenology of the main natural land covers (deciduous forest, evergreen forest, grasslands and shrublands) for the different European bio-geographical regions were characterised to establish a baseline to assess the impact of past and future environmental changes. This aspect was critical because many previous studies have not stratified phenological characteristics by land cover class and biogeographical region in Europe (see Rodriguez-Galiano et al. 2015 – Remote Sensing).
Our approach has established a potential reference for evaluating the susceptibility of vegetation to global climatic changes, and allows comparison of how different regions could respond under climatic conditions. The phenology maps produced here represent one of the most comprehensive and recent assessments of the land surface phenology of continental Europe at a 1 km spatial resolution. The spatial variability of phenology can be used for tuning models of carbon emissions and can play an important role in monitoring programmes for the management of natural vegetation.
Task 3. Validation of the phenology characterisation
The spatio-temporal reliability of the LSP estimates described above were assessed using ground phenology observations for a large number of points of different deciduous tree plant species collected across Europe within the framework of the PEP725 ground phenology (GP) network (http://www.pep725.eu/(opens in new window)). Correlations observed between the interannual time series of the satellite sensor estimates of phenology and PEP725 records revealed a close agreement (especially for Betula Pendula and Fagus Sylvatica species). In particular, 90% of the statistically significant correlations between LSP and GP were positive (mean R2 = 0.77). A large spatiotemporal correlation was observed between the dates of the start of season (end of season) from space and leaf unfolding (autumn coloring) at the ground (pseudo R2 of 0.70 (0.71)) providing, for the first time, the ability to predict accurately the date of leaf unfolding (autumn coloring) across Europe (root-mean-square error of 5.97 days (6.75 days) over 365 days).
A new methodology to relate LSP and GP has been proposed in this project through the application of Random Forest, a nonparametric method which allows for nonlinear relationships between phenology variables and for the inclusion in the modeling process of plant species. The proposed method can recognize complex patterns between LSP and the phenology of multiple specific plant species, integrating them into a unique overall model, rather than generating multiple models for every species. Additionally, it is data driven, which means that there is no need to incorporate previous knowledge about the species composition in the landscape, but it considers the different plant species of GP (see Rodriguez-Galiano et al 2015 – Geophysical Research Letters).
Task 4. Modelling of anomalies in phenology using climatic data.
Understanding the effect of inter-annual climatic variation on LSP is an essential step to establish a plausible link between recent climate variability and vegetation phenological responses at global or regional scales, and importantly to make reliable forecasts about future vegetation responses to different future climatic scenarios. Our aim was, therefore, to provide an explanation of the observed anomalies in LSP of the entire European forest during the last decade, identifying the main climatic drivers of spring and autumnal LSP at the continental scale applying new methodologies.
In Rodriguez-Galiano et al 2015 – Biogeosciences we modelled the climate-driven anomalies in phenology, rather than trends, and using innovative multivariate non-linear machine learning techniques to evaluate multiple climatic predictors at biological scales, and non-climatic predictors such as the legacy effect of the date of spring onset in leaf senescence. Climate predictors used range from monthly average values of temperature (max, min and avg), precipitation, short wave radiation and day length; trimestral cumulated values such as growing degree days or chilling requirements, among others; to the date of specific events such as the first freeze or the last freeze. Moreover, we considered flexible biological time scales in the analysis between climatic and phenological events rather than fixed calendar dates.
This part of the project revealed new insights into the climatic drivers of anomalies in LSP across the entire European forest, while at the same time established a new conceptual framework for predictive modelling of LSP. Specifically, the Random Forest method, a multivariate, spatially non-stationary and non-linear machine learning approach, was introduced for phenological modelling across very large areas and across multiple years simultaneously: the typical case for satellite-observed LSP. The RF model was fitted to the relation between LSP anomalies and numerous climate predictor variables. The models explained 81% and 62% of the variance in the spring and autumn LSP anomalies, with relative errors of 10% and 20%, respectively: a level of precision that has until now been unobtainable at the continental scale. A further comparison with a linear regression analysis suggested that there might be a non-linear relationship between the anomalies in LSP and the climatic drivers. It also allowed identification of the main drivers of the anomalies in LSP through its estimation of variable importance. Our results revealed that the accuracy of the broadly applied growing degree day-based models is overestimated using linear regression models and that non-linear multivariate relationships between temperature (especially minimum temperature) and radiation are needed to describe the relations between phenology and climatic drivers. We would like to highlight importance of minimum temperatures as a better indicator of climatic changes than either the average, maximum temperature, or growing degree days.
Our findings, thus, showed clearly the inadequacy of the hitherto applied linear regression approaches for modelling LSP and paves the way for a new set of scientific investigations based on machine learning methods.