Global environmental change is strongly dependent on land surface processes. Namely, the vegetation dynamics is closely related to the atmosphere through water, CO
and energy exchanges [1
]. The leaf area index (LAI), defined as the area of green leaves per unit ground horizontal surface area [2
], is a good indicator of the vegetation growth. As such it is an essential parameter for the description of the vegetation dynamics [4
], crop monitoring [5
] or climate change studies [6
]. Nunes, C. and Auge, J.I. [8
] reported that LAI can be used for detection of change and for providing information on shifting trends or trajectories in land use and cover change. With recent advances in remote sensing techniques, it has become possible to study the LAI variations at the global scale and in a consistent way over the last decades using a variety of techniques (see, e.g., the review of Fang, H. and Jiang, C. et al. [9
How vegetation will respond to climate change has been the topic of several recent studies based on remotely derived LAI. Most of them reported positive trends in vegetation (greening) during recent decades [10
]. In return, changes in vegetation dynamics may affect local climate via several biophysical processes, leading to positive or negative feedback in the land-climate system [16
]. For instance, Forzieri, G. et al. [18
] showed that positive trend in LAI contributed to the warming of boreal zones and to a cooling in arid regions. The impacts of climate change on the vegetation dynamics and their potential interactions have been increasingly studied, given their possible benefits in carbon sequestration [19
Zhu, Z. et al. [14
] related changes in vegetation greenness to multiple biogeochemical drivers (such as CO
concentration in the atmosphere or changes in temperature and precipitation) and land-use effects (fertilization, irrigation, etc.). They also identified local impact at regional scale of such drivers on the vegetation dynamics. For instance, rising CO
is the dominant factor in the tropical zone while changes in the vegetation dynamics are mainly driven by climate warming in high latitudes of the Northern Hemisphere.
Vegetation growth highly depends on climate conditions but also on the vegetation coverage density (sparsely, moderately or highly vegetated regions), as shown by Feng, H. et al. [22
]. In addition, the type of vegetation is expected to play a role in the relationship between vegetation and climate as the growing mechanisms of the plant differ from one type to another (e.g., drought avoiding or drought tolerant behaviours for crops, [23
]). There is a quite extensive literature on the recent greening of different vegetation types, including forests (e.g., [16
]) and grasslands (e.g., [18
These studies were based on land cover maps to select pixels or areas corresponding to a specific vegetation type. Land cover maps describe the partition of land surfaces into several biomes depending on their seasonal vegetation cycle. These maps generally provide, at high spatial resolution, either one type of vegetation (or plant functional type, PFT) per pixel (e.g., ESA-CCI Land Cover, [29
]), or cover fractions of several PFTs per pixel (e.g., ECOCLIMAP-II, [30
]). In both cases, assigning only one vegetation type per pixel, generally the dominant one, has two main drawbacks. First the LAI supposed to represent the dominant vegetation type is contaminated by spurious signal due to other types. Second, at the global scale, significant areas of each vegetation type are neglected. Figure 1
presents the fraction of the vegetated area (bare soil excluded) corresponding to the dominant vegetation type based on the ECOCLIMAP-II land cover map at 5 km resolution. Some regions in the world show values very close to 1, meaning that these areas are very homogeneous (only one vegetation type within the pixel), like the Amazon forest or the Tibetan plateau for instance. However, in most parts of the globe, pixels at this resolution are mixed and the signal is not necessarily representative of the dominant vegetation type. At this resolution, 12% of the total grassland area, 23% for evergreen forests, 36% for coniferous forests and 72% of broadleaf forests are neglected. In the following, the method that consists in selecting only the dominant vegetation type for each pixel will be referred as the single-dominant selection method.
Retrieving the LAI of each vegetation type within a pixel from the total observed LAI is not straightforward. Because these products present a coarse spatial resolution (e.g., 1 km for GEOV1, [31
]), the spatial scale does not permit discriminating between vegetation types within mixed pixels. The observed LAI over a pixel represents the weighted average of LAI of each type of vegetation within the pixel. Thus, despite the importance of improving our understanding of vegetation dynamics and its relationship with climate change, the mechanisms driving the vegetation growth are still unclear [32
In a previous study, Carrer, D. et al. [33
] presented an innovative method to disaggregate satellite-based albedo into bare soil and vegetation albedos. They developed a Kalman Filter (KF) approach to retrieve both quantities from a decade of MODIS data on a global scale. The derived product showed good agreement with previous studies and was indirectly assessed through the relationship between bare soil albedo and soil moisture. The method was further improved by Planque, C. et al. [34
] to include a dynamic fraction of vegetated area into the retrieval.
The main objective of this study is to detect and quantify changes in the vegetation dynamics for individual vegetation type at the global scale over the last 17 years. Following the method of Carrer, D. et al. [33
], we first develop a Kalman Filter approach to disaggregate the satellite-derived LAI from GEOV1 over nine main vegetation types, including broadleaf, coniferous and evergreen forests as well as crops and grasslands. We use data from the ECOCLIMAP-II land cover database as prior information. This KF approach permits the separation the individual LAI of different vegetation types that co-exist in a grid pixel. In a second step a trend analysis is conducted using a statistical test for each vegetation type over the 1999–2015 period. The analysis is performed at both global and regional scales (continental scale and main climate zones).
presents the GEOV1-LAI satellite product and the ECOCLIMAP-II database, as well as in situ LAI observations used for validation purposes. The disaggregation method based on the KF is detailed in Section 3
. Validation against in situ observation and trend analysis are presented in Section 4
and discussed in Section 5
This study presents an innovative method to disaggregate the total LAI observed from satellites into LAIs of several vegetation types, including broadleaf, evergreen and coniferous forests, grasslands and summer/winter crops. The algorithm is an extension of the method from Carrer, D. et al. [33
] to reconstruct bare soil and vegetation albedos from total albedo. It is based on a Kalman Filter approach and makes use of cover fraction and LAI seasonal cycle per vegetation types from the ECOCLIMAP-II database as prior information. The derived product, called LAI-MC (Multi-Cover), consists of world-wide LAI maps provided every 10 days for each vegetation type over the period 1999–2015. The LAI-MC product is validated against in situ measurements with overall good performances, at least as good as for the total LAI, even for pixels where the vegetation type corresponding to the local site is not dominant.
A trend analysis is performed at the global scale and over different regions (continents and climate zones). On average, all vegetation types have experienced greening over the last two decades at rates ranging from 0.026 mmyr for winter crops to 0.042 mmyr for coniferous forests. Coniferous forests are mainly greening in temperate regions and show the largest area affected by high positive trends. By contrast, grasslands are greening at a moderate average rate, but since they cover almost half of the total vegetated area, the grassland area affected by high trend values is greater than for any other vegetation type but coniferous forests. Evergreen forests, which are mainly located in the tropical regions, show high positive trends except for deforested areas of the Amazon Basin. Coniferous and broadleaf forests have been browning in Central Asia and greening in most other regions of the Northern Hemisphere. Globally, grasslands and summer crops experienced greening while winter crops show small or even negative trends.
These results are in agreement with previous studies on the regions of the world that show greening or browning. Nevertheless, some differences can be noted among vegetation types that other studies could not identify. For instance, our study shows that coniferous forests of Central Asia are greening while broadleaf forests in the same region are mostly browning. In the North-Eastern part of Asia, coniferous and broadleaf forests are greening while grasslands are browning. Coniferous forests, winter and summer crops and grassland in China are all greening but at different rates. The same remark applies for South America.
LAI trends of each vegetation type are compared to the single-dominant selection method (obtained by considering only the dominant vegetation type for each pixel). Small differences are observed for evergreen and coniferous forests. On the contrary, the spatial average of the relative difference reaches ±27% to 53% for other vegetation types. This result highlights the limited representativeness of the total LAI with respect to the dominant vegetation type, as well as the added value of the LAI-MC disaggregated product.
Finally, the new LAI-MC dataset is the first product providing global maps of vegetation growth every 10 days for the main vegetation types over the last 17 years. Although the problem of detection/attribution is behind the scope of this study, the LAI-MC product can help such analyses by offering a more consistent comparison with observations and derived products. Apart from climate related studies, it may be valuable for Land Surface Models (LSMs) to update phenology parameters. Also the assimilation of satellite derived LAI observations into the ISBA LSM as done recently by Albergel, C. and Munier, S. et al. [83
] is based on the comparison between the total observed LAI and the aggregated LAI simulated over the different vegetation types. An extension is under development to assimilate the disaggregated LAI-MC product for each vegetation type independently, with very promising preliminary results.