Global climate/carbon cycle models show a large spread in simulations of the land carbon sink under future climate change. This spread originates from how Dynamic Global Vegetation Models (DGVMs) simulate the response of plant carbon uptake and carbon release to climate variability and depends on differences in model structures and parameterizations. Carbon uptake and release by land ecosystems are controlled by processes such as vegetation productivity, biomass allocation and turnover, disturbances such as fires and plant mortality. DGVMs account for such processes but have not yet been systematically assessed over large spatial scales. However, the availability of global satellite observations on vegetation distribution, biomass, greenness and productivity allows to benchmark and potentially improve DGVMs. We give examples how satellite observations are used to identify deficiencies in DGVMs. Beyond the application of satellite observations for model benchmarking, model-data integration approaches allow to estimate model parameters and hence to directly improve models. Here, we use multiple satellite observations to estimate parameters of the LPJmL (Lund-Potsdam-Jena managed Lands) DGVM. Satellite observations include maps of above-ground biomass from the ESA BIOMASAR algorithm, multi-temporal land cover and burned area maps from ESA’s Climate Change Initiative (CCI), time series of the fraction of absorbed photosynthetic active radiation (FAPAR) from MODIS and of sun-induced florescence (SIF) from GOME-2. We use these datasets to compute a joint multi-dimensional model-data error and use a genetic optimization algorithm to estimate model parameters and plant traits that control phenology, productivity, vegetation dynamics, and mortality in LPJmL. Our results show that both the prior and the optimized model accurately reproduce observed temporal dynamics of FAPAR and SIF and independent datasets of the global carbon cycle. However, the optimized model better reproduces observed spatial patterns of biomass and vegetation distribution especially in northern boreal ecosystems and in tropical forests (Figure). Despite these clear model improvements, the results demonstrate that the model cannot well reproduce the observed mixture of woody and herbaceous vegetation in Savannah regions. Our results demonstrate that model-data integration techniques are the key to harvest the value of multiple and long-term satellite observations for the continuous development of global vegetation models.

Transpiration (Tr) is an important and complex plant physiological process in terrestrial ecosystems and significantly impacts the global water and energy balance. Assessing the dynamics of Tr has versatile implications to understand causes of water scarcity, to advance understating of interactions and feedbacks between global climate change and ecosystem functioning, to assess plant water use efficiency for optimized agricultural management, and to exploit the coupled carbon and water cycle. Nevertheless, spatio-temporal assessments of Tr using models or Earth observation (EO) informed approaches are still uncertain due to the complexity of Tr, limitations in existing methodology, and lack of observational approaches.
In this study, we apply a novel mechanistic approach to estimate Tr and understand its dynamics under an extreme drought event in Europe in summer 2018. The approach combines mechanistic modeling (i.e. a combination of the Penman-Monteith (PM) model and the Ball-Berry-Leuning (BBL) model) and multi-source EO data (i.e. Sentinel-3 satellite, European Centre for Medium-Range Weather Forecasts (ECMWF)). We calculate potential and actual Tr by parameterizing the canopy surface resistance term with a greenness based estimate of photosynthesis (potential Tr) and a sun-induced chlorophyll fluorescence (SIF) surrogate (actual Tr). An assessment of monthly time series of a study site in southwestern Europe over two years, the drought year 2018 and the baseline year 2017, allowed investigating the dynamics of Tr under different climatic conditions. The results of potential and actual Tr were validated with flux data from eddy covariance sites in Europe.
The study revealed complex pattern of Tr dynamics (potential and actual) in response to the 2018 drought event in Europe. We discuss these results considering possible superimposing environmental factors and for different ecosystem types. We evaluate the suitability of the novel mechanistic approach developed in this study to reveal understanding of Tr dynamics in space and time. We outline limitations in underlying methodology and identify gaps in current observational data. Our results also facilitate the definition of requirements for future missions to complement our multi-source approach.

Previous research has shown that seasonal gross primary productivity (GPP) can be modelled with fair accuracy across boreal and sub-arctic ecosystems based on coarse-resolution data from e.g. MODIS. However, in heterogeneous landscapes these estimates may be unreliable due to the coarseness of the data. We here report on work to utilize higher resolution remotely sensed data for better describing the landscape and to improve the matching with flux tower measurements. For regional applications, Sentinel-2 data is tested with data from eight flux tower sites in Scandinavia, spanning a climate gradient across temperate, boreal and sub-arctic areas. We apply a newly developed time-series processing method to Sentinel-2 time-series data, implemented in the TIMESAT package, to gap-fill and smooth radiometrically corrected Sentinel-2 data. We then develop models based on the vegetation index EVI2, incident photosynthetically active radiation (PAR), and air temperature, to estimate GPP for average flux footprint areas. We compare the results with models driven by MODIS vegetation index data to assess the improvement when going to the higher resolution of Sentinel-2.

To analyze the effect of landscape heterogeneity at finer scale on the estimation of GPP we focus on a palsa peatland in N. Sweden (Abisko-Stordalen), characterized by melting permafrost. The GPP data are derived from a flux tower with a footprint area influenced by different wetness zones in a pattern smaller than the resolution of Sentinel-2. Here we map the footprint area using multispectral camera images at 10 cm resolution from a UAV campaign. We apply a classifier to map the peatland into wetness classes based on spectral data and topography. For wet and dry classes we then apply spectral time-series data from two spectral masts. These masts measure reflected radiation in visible, near-infrared and short-wave infrared bands year round, in wet and in dry locations of the peatland respectively. We further use a footprint model to allocate CO2 fluxes to source areas depending on the atmospheric conditions. The analysis shows the degree to which carbon uptake of the mire is heterogeneous across the footprint area, and the capacity for spectral data to aid in flux data upscaling.

Comparing results from the two approaches provides an insight into the scalability of fluxes when going from flux towers to local areas and regions. It also provides insights into the degree to which models are constrained by access to spatial data or by model formulations.

We develop and apply new technologies, based on airborne and spaceborne imaging and ground surveys, climate measurements and modelling, to understand changes in the forests of northern Russia since 2000. Northern forests are extremely sensitive to climate change, and are important not just as indicators but also because of their contribution to global biosystems and biodiversity, supporting traditional land management, and as economic resources. New possibilities have been created by research on climate data from northern Russia combined with long datasets of coarse-resolution satellite data, high-resolution imaging systems, airborne imaging (‘drone’) technologies and in situ measurements of the tree stands. We aim to describe and explain changes in forest distribution since 2000 and predict future impacts. In this study, we present results of forest stand modelling for 28 test sites in the central Kola Peninsula, NW Russia, surveyed in the field and from UAVs in summer 2018, and of upscaling these data to MODIS scale. Test sites cover pine-, spruce- and birch-dominated forests.
We derive stand parameters: tree height, canopy cover, crown sizes, leaf area index (LAI) and growing stock volume (GSV), utilising DTMs and dense point clouds generated from UAV surveys taken from 50 m and 100 m altitude, and validated with ground measurements. These parameters are then statistically upscaled to MODIS-derived LAI and GSV products. Their trends are followed since 2000 and explained on the basis of stand structure in the test sites, as well as information on climate change and human-induced impacts in the region.
It is envisaged that further work will extend to test sites in larch forests in Eastern Siberia, and will include ground photogrammetry to improve stand characterization.
This work is funded by the British Council (Institutional Links grant no. 352397111) and the Ministry of Science and Higher Education of the Russian Federation (grant agreement no. 14.616.21.0099 dated 27 February 2018, project RFMEFI61618X0099).

Spatial differences and signal mixing between the canopy and ecosystem scales often cause for discrepancies in linking in-situ ground data with airborne and/or satellite earth observation data. Evapotranspiration (ET), the combination of land evaporation and plant transpiration, is a key process that interlinks the water and energy budget, along with carbon cycling through the processes of transpiration and photosynthesis. Thus, ET, and the ability to estimate it within a spatial and temporal context, has important implications for atmosphere-land interactions, ecosystem functional properties, and their related plant traits. The Two Source Energy Balance (TSEB) modeling scheme [1], which estimates ET by partitioning radiometric land surface temperature from thermal infrared (TIR) observations between the surface's soil and vegetation components, is applied in a Mediterranean tree-grass ecosystem in central Spain to investigate the retrieval of ET at different spatial scales. A major limitation in TIR remote sensing is the need to balance the sensor's radiometric capabilities to its spatial resolution, resulting in TIR imagery tending to be captured at a relatively coarse resolution. This is particularly an issue for heterogenous landscapes, such as tree-grass ecosystems, where different vegetation and/or land cover types, with different radiometric and physical properties, may not be separated at the coarse pixel level. As such, a thermal sharpening technique is tested where TIR images are disaggregated by relating it with homologous finer resolution optical data. A machine learning approach is used [2] to develop a statistical relationship between the high-resolution optical image and the lower resolution TIR data. TSEB is then applied to simulate ET on two distinct scales: fine-resolution airborne imagery by combining Compact Airborne Spectrographic Imager (CASI) and Airborne Hyperspectral Scanner (AHS) images at 1m and 4.5m resolution respectively; and medium-resolution satellite imagery by fusing Sentinel-2 and 3 images at 20m and 1000m resolution, where different levels of within-pixel heterogeneity (% of tree, grass and soil) are apparent. The fluxes derived from different data sources and resolutions, including the use of sharpened and non-sharpened TIR imagery, are compared and benchmarked against flux measurements from eddy-covariance (EC) towers located within the experimental site. At the tower scale, TSEB, when accounting for seasonal changes within the study site, adequately simulates sub-hourly ET compared to EC tower flux measurements with a root-mean-square deviation (RMSD) of 38 W/m2 and a Pearson's correlation coefficient (r) of 0.91.

[1] Norman, J. M., Kustas, W. P., & Humes, K. S. (1995). Source approach for estimating soil and vegetation energy fluxes in observations of directional radiometric surface temperature. Agricultural and Forest Meteorology, 77(3-4), 263-293
[2] Gao, F., Kustas, W.P., & Anderson, M.C., (2012). A Data Mining Approach for Sharpening Thermal Satellite Imagery over Land. Remote Sensing 4(11), 3287

The combination of models describing radiation radiative transfer, plant photosynthesis and surface energy balance bring new opportunities for the study of functional biodiversity of plants from spaceThe Soil-Canopy Observation Photosynthesis and Energy fluxes (SCOPE) model is currently the state-of-art model combining optical, thermal and sun-induced chlorophyll fluorescence (F) radiative transfer models with soil-vegetation-atmosphere transfer models. SCOPE is intended to allow monitoring of plant photosynthesis from remote observations of optical, thermal and F when ESA Fluorescence EXplorer mission became operational. Different sources of information can be combined to inform about key essential biodiversity variables describing plant function, which are relevant for land surface modeling and to understand vegetation responses to changing environmental conditions. In this work we propose data assimilation of optical remote sensing images with Eddy Covariance (EC) flux measurements to retrieve biophysical properties of vegetation together with maximum carboxylation rate (Vcmax) and the stomatal sensitivity parameter (m) in the Ball-Berry model.
The study site is a Mediterranean tree-grass ecosystem consisting of Holm Oaks (Quercus ilex, ~20% canopy cover) with an understory of annual grasses in Majadas de Tiétar, Cáceres, Spain. An EC tower operates since 2004. and 2014 two more towers were included in the ecosystem as a part of a large scale manipulation experiment where the footprints of the new towers were fertilized with nitrogen and nitrogen plus phosphorous respectively. Since 2010, time series of airborne hyperspectral imagery have been acquired almost yearly with the Compact Airborne Spectrograph Imager (CASI, ITRES Research Limited, Canada). Airborne campaigns took place in spring, close to the seasonal green peak of the ecosystem as well as in the dry season, when grasses are senescent. Simultaneously, destructive vegetation sampling allowed estimating biophysical parameters such as Leaf Area Index (LAI) and nitrogen content (Nmass). In this work we are using data from xxx campaigns from xx to xxx. For every image, we selected pixels around each tower within the cumulative footprint of a 7 day period centered around the flight dates. Averaged reflectance factors (RF) from these cut-outs were combined with EC Gross Primary Production (GPP) and up-welling thermal radiance (L↑T) observed by CNR4 net radiometers in the EC towers (Kipp & Zonen, Delft, Netherlands) to invert SCOPE model. A two-step methodology was followed: In the first step we combined RF and the GPP smoothed at the time of each overpass to retrieve leaf parameters (pigments, water, dry matter and the number of layers) plus and LAI, leaf angle distribution and Vcmax.. In a second step, time series of daytime GPP and L↑T ± three days around each overpass were combined to constrain Vcmax and m; biophysical parameters were kept fixed from the previous step. SCOPE was inverted using a nonlinear least-squares numerical approach. However, the starting solution point was set from an ad-hoc look-up table generated for each image optimized against the RF.
Estimation of biophysical parameters was successful. LAI was predicted with a coefficient of determination R2 = 0.53 and root mean squared error RMSE = 0.65 m2/m2. Chlorophyll content was evaluated against Nmass with R2 = 0.48 and RMSE = 0.65 g/cm2. Plant traits were also satisfactorily estimated. Vcmax was also evaluated against Nmass with R2 = 0.17 and RMSE = 56.1 μmol/m2/s. We also evaluated the capability of the constraint parameters to predict EC fluxes. GPP and L↑T used in the inversion were estimated with R2 = 0.57 and RMSE = 5.43 μmol/m2/s and R2 = 0.61 and RMSE = 32.96 W/m2 respectively. Predicted latent (λE) and sensible heat (H) fluxes did not take part in the inversion. λE was predicted with R2 = 0.08 and RMSE = 88.15 W/m2 and H with R2 = 0.15 and RMSE = 100.96 W/m2. We identified that most of the uncertainty came from the predictions during the dry season. In the spring campaigns, λE predictions also improved with R2 = 0.45 and RMSE = 81.22 W/m2 and H with R2 = 0.68 and RMSE = 55.00 W/m2.
Multiple constrained inversions of SCOPE using proximal sensing was successfully achieved in the same ecosystem mainly during spring campaigns (Pacheco-Labrador et al., under review). In this work, we evaluate the same method at ecosystem scale and with a larger contribution of data acquired during the dry season. Results are satisfactory considering the increased uncertainties included in the model as the 3-D structure of the ecosystem not being properly represented by the 1-D radiative transfer of SCOPE. Time series of fluxes used in the inversion are not accompanied of hyperspectral observations of down-welling irradiance. Despite of applying a footprint model, there might be still a certain spatial mismatch between the spectral and EC footprints. Also, higher uncertainties would be present in the spectral data propagated from the atmospheric correction of the imagery. Nonetheless, biophysical and functional traits seem to be acceptably retrieved, and fluxes adequately estimated. The methodology proposed could be applied with similar or better results in ecosystems with a less complex structure. The performance of the approach lost predictive power in the dry season where likely both, the radiative transfer and the energy balance, represent less accurately the processes and status of the ecosystem. Grass litter seems not to be well represented by the radiative transfer models (Melendo-Vega, et al., submitted). Furthermore, SCOPE does account for the presence of a dry fraction in LAI in the energy balance, which leads to an overestimation of λE. The approach successfully proposed in (Pacheco-Labrador et al., under review) seems to perform poorer in the scenario of increased uncertainties faced at ecosystem scale. More work should be done in this direction in the future.

References:
Pacheco-Labrador, J., Perez-Priego, O., El-Madany, T.S., Julitta, T., Rossini, M., Guan, J., Moreno, G., Carvalhais, N., Martín, M.P., González-Cascón, R., Kolle, O., Reichstein, M., van der Tol, C., Carrara, A., Martini, D., Hammer, T., Moossen, H., & Migliavacca, M. (under review). Multiple-constraint inversion and benchmarking of the scope model for functional traits retrieval. Remote Sensing of Environment.
Melendo-Vega, J.R., Martín, M.P., Pacheco-Labrador, J., González-Cascón, R., Moreno, G., Pérez-Cabello, F., Migliavacco, M., García, M., North, P., & Riaño, D. (under review). Modelling optical properties of a tree-grass ecosystem by coupling PROSAIL and FLIGHT Radiative Transfer Models. Remote Sensing