Search

Terrestrial water storage (TWS) is an integrative hydrological state that is key for our understanding of the global water cycle. The TWS observation from the GRACE missions has, therefore, been instrumental in the calibration and validation of hydrological models and understanding the variations in the hydrological storage. The models, however, still show significant uncertainties in reproducing observed TWS variations, especially for the interannual variability (IAV) at the global scale. Here, we diagnose the regions dominating the variance in globally integrated TWS IAV and the sources of the errors in two data-driven hydrological models that were calibrated against global TWS, snow water equivalent, evapotranspiration, and runoff data. We used (1) a parsimonious process-based hydrological model, the Strategies to INtegrate Data and BiogeochemicAl moDels (SINDBAD) framework and (2) a machine learning, physically based hybrid hydrological model (H2M) that combines a dynamic neural network with a water balance concept. While both models agree with the Gravity Recovery and Climate Experiment (GRACE) that global TWS IAV is largely driven by the semi-arid regions of southern Africa, the Indian subcontinent and northern Australia, and the humid regions of northern South America and the Mekong River basin, the models still show errors such as the overestimation of the observed magnitude of TWS IAV at the global scale. Our analysis identifies modeling error hotspots of the global TWS IAV, mostly in the tropical regions including the Amazon, sub-Saharan regions, and Southeast Asia, indicating that the regions that dominate global TWS IAV are not necessarily the same as those that dominate the error in global TWS IAV. Excluding those error hotspot regions in the global integration yields large improvements in the simulated global TWS IAV, which implies that model improvements can focus on improving processes in these hotspot regions. Further analysis indicates that error hotspot regions are associated with lateral flow dynamics, including both sub-pixel moisture convergence and across-pixel lateral river flow, or with interactions between surface processes and groundwater. The association of model deficiencies with land processes that delay the TWS variation could, in part, explain why the models cannot represent the observed lagged response of TWS IAV to precipitation IAV in hotspot regions that manifest as errors in global TWS IAV. Our approach presents a general avenue to better diagnose model simulation errors for global data streams to guide efficient and focused model development for regions and processes that matter the most.

Long-term global monitoring of terrestrial gross primary production (GPP) is crucial for assessing ecosystem responses to global climate change. In recent decades, great advances have been made in estimating GPP and many global GPP datasets have been published. These datasets are based on observations from optical remote sensing, are upscaled from in situ measurements, or rely on process-based models. Although these approaches are well established within the scientific community, datasets nevertheless differ significantly. Here, we introduce the new VODCA2GPP dataset, which utilizes microwave remote sensing estimates of vegetation optical depth (VOD) to estimate GPP at the global scale for the period 1988–2020. VODCA2GPP applies a previously developed carbon-sink-driven approach (Teubner et al., 2019, 2021) to estimate GPP from the Vegetation Optical Depth Climate Archive (Moesinger et al., 2020; Zotta et al., 2022​​​​​​​), which merges VOD observations from multiple sensors into one long-running, coherent data record. VODCA2GPP was trained and evaluated against FLUXNET in situ observations of GPP and compared against largely independent state-of-the-art GPP datasets from the Moderate Resolution Imaging Spectroradiometer (MODIS), FLUXCOM, and the TRENDY-v7 process-based model ensemble. The site-level evaluation with FLUXNET GPP indicates an overall robust performance of VODCA2GPP with only a small bias and good temporal agreement. The comparisons with MODIS, FLUXCOM, and TRENDY-v7 show that VODCA2GPP exhibits very similar spatial patterns across all biomes but with a consistent positive bias. In terms of temporal dynamics, a high agreement was found for regions outside the humid tropics, with median correlations around 0.75. Concerning anomalies from the long-term climatology, VODCA2GPP correlates well with MODIS and TRENDY-v7 (Pearson's r 0.53 and 0.61) but less well with FLUXCOM (Pearson's r 0.29). A trend analysis for the period 1988–2019 did not exhibit a significant trend in VODCA2GPP at the global scale but rather suggests regionally different long-term changes in GPP. For the shorter overlapping observation period (2003–2015) of VODCA2GPP, MODIS, and the TRENDY-v7 ensemble, significant increases in global GPP were found. VODCA2GPP can complement existing GPP products and is a valuable dataset for the assessment of large-scale and long-term changes in GPP for global vegetation and carbon cycle studies. The VODCA2GPP dataset is available at the TU Data Repository of TU Wien (https://doi.org/10.48436/1k7aj-bdz35, Wild et al., 2021).

Masting is a well-known phenomenon in forest science in which trees recurrently produce exceptional high seed amounts throughout their lifespan. Occurring simultaneously across wide regions, masting may impact carbon sequestration and carbon cycling of forest ecosystems. To identify masting-induced changes in tree canopies, remote sensing methods may offer a suitable tool, though so far only few studies have used RS methods to investigate the relationship between masting and tree-ring width (TRW). Therefore, we employed a regional network of 19 European beech sites along a precipitation gradient in north-eastern Germany, providing time series of TRW (1961–2017) and categorical masting data (1992–2020). Combined with leaf area index (LAI) from the medium-resolution satellite sensors SPOT-VGT/PROBA-V and MODIS, we assessed the relationship between all three parameters by correlation analysis. Furthermore, we identified common climatic drivers and spatial masting patterns along gradients of water availability. We found negative relationships for both TRW and LAI with masting and a positive correlation for TRW with LAI for SPOT-VGT/PROBA-V data. Contrary, MODIS LAI provided no significant relationships. Water availability did not impact masting occurrence or intensity. The climate correlations showed that masting and TRW are commonly driven by summer climate in June and July, while we found no correlations between climate and LAI. We conclude that medium-resolved SPOT-VGT/PROBA-V LAI is suitable to observe masting but not to estimate TRW. Nonetheless, the results should be thoroughly interpreted in the context of uncertainties of RS measurements and complex interactions of different environmental drivers.

Remote sensing has played a fundamental role for land cover mapping and change detection at least since the launch of the Landsat satellite program in 1972. In 1995, the Central Intelligence Agency of the United States of America released previously classified spy imagery taken from 1960 onwards with near-global coverage from the Keyhole programme, which includes the CORONA satellite mission. CORONA imagery is a treasure because it contains information about land cover 10 years before the beginning of the civilian Earth observation and has a high spatial resolution < 2m. However, this imagery is only pan-chromatic and usually not georeferenced, which has so far prevented a large-scale application for land cover mapping or other geophysical and environmental applications.Here, we aim to harvest the valuable information about past land cover from CORONA imagery for a state-wide mapping of past land cover changes between 1965 and 1978 by training, testing and validating various deep learning models.To the best of our knowledge, this is the first work to analyse land cover from CORONA data on a large scale, dividing land cover into six classes based on the CORINE classification scheme. The particular focus of the work is to test the transferability of the deep learning approaches to unknown CORONA data.To investigate the transferability, we selected 27 spatially and temporally distributed study areas (each 23 km²) in the Free State of Saxony (Germany) and created semantic masks to train and test 10 different U-shaped neuronal network architectures to extract land cover from CORONA data. As input, we use either the original panchromatic pixel values and different texture measures. From these input data, ten different training datasets and test datasets were derived for cross-validation.The training results show that a semantic segmentation of land cover from CORONA data with the used architectures is possible. Strong differences in model performance (based on cross validation and the intersection over union metric, IOU) were detected among the classes. Classes with many sample data achieve significantly better IOU values than underrepresented classes. In general, a U-shaped architecture with a Transformer as Encoder (Transformer U-Net) achieved the best results. The best segmentation performance (IOU 83.29%), was obtained for forests, followed by agriculture (74.21%). For artificial surfaces, a mean IOU of 68.83% was achieved. Water surfaces achieved a mean IOU of 66.49%. For the shrub vegetation and open areas classes only IOU values mostly below 25% were achieved. The deep learning models were successfully transferable in space (between test areas) and time (between CORONA imagery from different years) especially for classes with many sample data. The transferability of deep learning models was difficult for the mapping of water bodies. Despite the general good model performance and successful transferability for most classes, the transferability was limited especially for imagery of very poor quality. Our approach enabled the state-wide mapping of land cover in Saxony between 1965 and 1978 with a spatial resolution of 2 m. We identify an increase in urban cover and a decrease in cropland cover

Changes in plant phenology as for example earlier leaf unfolding and delayed autumn senescence can result in variations in the carbon and water cycle. Studies investigating the impact of phenological shifts on biophysical processes such as water availability are still limited. Due to the sensitivity of radar satellite observations to both, structural and dielectric properties of the scattering materials, microwave remote sensing offers the potential to analyse structural (i.e. canopy biomass) and physiological (i.e. water status) dynamics in vegetation.Here, we aim to derive annual water dynamics of vegetation canopies from the Sentinel-1 C-band radar backscatter signal by removing the influence of vegetation structure on the backscatter seasonality. To decouple the phenology of vegetation structure from the moisture content dynamics, a semi-empirical backscattering model (Water Cloud Model, WCM) is combined with a canopy water balance model. The WCM aims to separate contributions of soil and vegetation to the total backscatter. When introducing physical parameters for vegetation structure like leaf area index (LAI) and and moisture like leaf fresh moisture content (LFMC) to describe the vegetation backscatter, the effect of the seasonal variability of both variables on the radar signal can be assessed. The canopy water balance model estimates interception and changes in the canopy saturation and storage capacity of the vegetation using precipitation and throughfall measurements. Both models are combined to iteratively estimate measures of vegetation moisture. To calibrate the two models, we use measurements of LFMC and of canopy interception for the Tharandt ecosystem site in Germany in 2022, which is part of the ICOS and FLUXNET network. The calibrated model is then used to analyse the individual effects of both vegetation descriptors, LAI and LFMC, by fixing either one and looking at the changes in the seasonality of the S1 signal. The combined use of both models will allow to remove the structural-related changes in the Sentinel-1 radar backscatter to finally retrieve vegetation water dynamics over larger areas.

Different disturbance events lead to varied response patterns in terrestrial biomass, while regulating the terrestrial ecosystems' short- and long-term carbon cycle dynamics. Quantifying the disturbance regimes is essential for understanding and reducing the uncertainty of climate factors affecting vegetation mortality and its responses on the carbon cycle. Based on model-based exercise, we built a machine learning model to predict three disturbance regime parameters, μ (probability scale), α (clustering degree), β (intensity slope) using the spatial pattern of emergent biomass. Here, relying on the model relationships, we utilize Earth observation data of high-resolution biomass, the GlobBiomass with a spatial resolution of 25m, to infer regional disturbance regime statistics.We first conduct a series of comparison exercises to test whether the current framework is robust for retrieving realistic disturbance regimes, including varying factors controlling: (i) the impacts of disturbance shape setting; (ii) the recovery pattern after perturbation; and (iii) the downsampling process for the biomass simulation. It was found that different model settings mainly lead to the inconsistency of texture features, and the disturbance regime prediction accuracy was maintained with different shape settings, or even higher after downsampling with a mean of 0.98 for Nash-Sutcliffe model efficiency coefficient (NSE).Given the robustness in the framework for retrieving disturbance regimes statistics from modelled biomass results we contrasted these spatial patterns with local GlobBiomass patterns across the world. The comparison between model and observations show data aggregation needs that provide information on aspects of scale and spatial resolution required for simulations. Finally, we provide a sparse, but globally distributed, characterization of disturbance regimes based on remote sensing observations and discuss potential climate links, and mechanisms behind, the spatially continuous distribution of disturbance regime.Given the novelty of the assessment of disturbance regimes with high-resolution biomass data, our study provides opportunities to evaluate and improve the representation of disturbance dynamics in dynamic vegetation and Earth System models.

Climate change leads to a change of precipitation frequency and quantity as well as to increased temperature inducing extreme weather events like floods but also more intensive and longer drought periods.The response of the vegetation to these trends is of high interest because vegetation regulates interception, transpiration and is a water storage which is important for plant productivity, agriculture, carbon cycling and the danger of wild fire occurrence. Reduced precipitation in combination with increased temperature lead to water stressed vegetation which might not only behave different in regards of evapotranspiration but are also prone to wildfires. However, currently we don’t know how the water status changes in the long-term. A long-term time series of the vegetation leaf moisture content can help to understand the consequences of changing environmental conditions on the vegetation layer as part of the water as well as the carbon cycle.Measurements of vegetation leaf moisture are usually only available for single test-sites (missing spatial coverage), often measured for a short time span and might hold missing data. Estimations of vegetation leaf moisture are able to provide consistent time series but are mostly done on regional scale which are also missing spatial transferability. However, long-term data with a consistent time series and large spatial coverage are necessary to address a reliable time series analyses in the context of climate change.Our trend analysis will focus on the live-fuel moisture content (LFMC) which is based on the vegetation optical depth (VOD) and Leaf Area Index (LAI). LFMC is defined as the water mass of living vegetation to the dry mass of the vegetation, usually expressed in percentage. LFMC is an important variable in the field of wild fire analyses as it is one of the key predictors for risk and development of a fire. LFMC can be estimated on ecosystem level due to its independence of plant type. Here we use VODCA VOD and GLOBMAP LAI data to create a longer time series of LFMC for the period 1988-2017 on global scale to analyse temporal changes in LFMC. Initial results indicate a heterogeneous pattern of LFMC trends which depend on land cover type, e.g., with a decreasing trend for shrublands but an increasing trend for needle-leaved forests. We compare the trends in LFMC with trends in heat and drought events as well as fire weather indices. Inter-annual changes in LFMC correspond to multi-year drought events.

Emissions from vegetation fires in tropical forests have the potential to turn the global land carbon sink into a source, affect atmospheric chemistry, and hence air quality. While natural forest fires are a rare phenomenon in tropical forests of South America and are usually of rather low intensity, deforestation fires and small land clearings in systems with high fuel loads can cause intense fires and high emissions. However, the high moisture content in tropical forests causes incomplete combustion and higher emissions of carbon monoxide (CO) than of carbon dioxide. The interacting effects of land use change, fuel load and moisture on fire intensity and emissions is, however, difficult to quantify at large scales because not all of those components are readily available from Earth observations in a consistent way. Here, we make use of several satellite products on vegetation, fire activity and atmospheric composition to quantify the effects of land use, fuel loads, fuel moisture on fuel consumption, emission factors and hence on emissions and atmospheric trace gas concentration. First, we use observations of active fires and fire radiative power from the VIIRS and Sentinel-3 SLSTR sensors to map different fire types (forest fires, deforestation fires, small land clearing and agricultural fires, savannah fires). Second, we integrate satellite products of canopy height, above-ground biomass, leaf area index, land cover and soil moisture in a novel data-model fusion framework to estimate fuel loads and moisture in vegetation, surface litter and woody debris. We then combine in a bottom-up approach the fire types with fuel loads and moisture to estimate fuel consumption and fire emissions using default emission factors. Third, we use observations from Sentinel-5p TROPOMI and the Integrated Forecast Systems (IFS) of the Copernicus Atmosphere Monitoring Service to compare the bottom-up estimates with distributions of CO and NOx in the atmosphere, which allows optimising emissions and associated emission factors.Our reconciled estimates of fire emissions outperform previous CO estimates e.g. from the Global Fire Assimilation System, which demonstrates an improved estimation of fire carbon emissions. The results show that the high fire intensity and emissions in tropical deforestation fires originate from the burning of high loads of woody biomass and coarse woody debris. The high fuel moisture content causes higher emission factors of CO in tropical forests than in savannah fires and hence higher absolute emissions of CO. Our new model approaches and satellite products allow to provide an integrated assessments on the effects of fuel and fire behaviour on fire emissions.

Wildfires are a pervasive feature of the terrestrial biosphere and contribute large carbon emissions within the Earth system. Humans are responsible for the majority of wildfire ignitions. Physical and empirical models are used to estimate the effects of fires on vegetation dynamics and the Earth system. However, there is no consensus on how human-caused fire ignitions should be represented in such models. We aim to identify which globally available predictors of human activity explain global fire ignitions as observed by satellites. We apply a random forest machine learning framework to state-of-the-art global climate, vegetation, and land cover datasets to predict global fire ignition density remote sensing data. We establish a baseline against which influences of socioeconomic data (cropland fraction, gross domestic product (GDP), road density, livestock density, and grazed lands) are evaluated to determine their effects on fire ignition occurrence predictions. Our results show that a baseline random forest without human predictors captures the spatial patterns of fire ignitions globally but overestimates ignition occurrence, with the highest predictions over Sub-Saharan Africa and South East Asia. Adding human predictors one by one to the baseline model reveals that human variables vary in their effects on fire ignitions, and GDP is the most vital driver of fire ignitions. We also find that high road density leads to decreased ignitions globally despite improved environmental access, mainly because these regions are primarily human settlements and will have fewer fires. A combined model with all human predictors shows that the human variables improve the ignition predictions in most areas of the world; still, some regions, e.g., East Africa, have worse predictions than the baseline model. We conclude that an ensemble of human predictors can add value to physical and empirical models. There are complex relationships between the variables, as evidenced by the improvement in bias in the combined model compared to the individual models. Furthermore, the variables tested have complex relationships that random forests may struggle to disentangle. Further work is needed to reach concrete conclusions at a global scale, but rather introduce the need for additional work to detangle the complex regional relationships between these variables, particularly across Central and Eastern Africa, where the full model performs poorly despite the high availability of fire ignition data.

The distribution of leaves, branches and trunks in the canopy is critical for water, carbon and energy cycles in forests. Light ranging and detection (lidar) and synthetic aperture radar (SAR) are two active remote sensing methods which show the potential to detect forest structure dynamics at different height. Lidar can penetrate the canopy gap and record a reflectivity profile. The vertical distribution of structure metrics, for example, canopy cover, can then be estimated based on this profile. For example, in Amazonian tropical forests, the leaf area index derived using ground- and space-borne lidar at canopy layer and understory layer shows converse behaviour over seasons. For SAR systems, wavelength determines the penetration depth of microwave signal and polarisation reflects the scattering mechanisms between signal and objects. SAR backscatter has been used for monitoring forest phenology and forest classification. However, the potential of using lidar and SAR to monitor canopy structure changes in temperate forests has not been analysed. The relationships between SAR backscatter with the lidar-derived structure metrics at height levels are also not clear.In this study, we attempt to investigate the influence of forest structure at height levels on SAR backscatter in two study sites in Germany: one deciduous forest in Hainich and one coniferous forest in Tharandt. Level 2B product of the Global Ecosystem Dynamics Investigation (GEDI) and Sentinel-1 ground range detected (GRD) backscatter are used in this study. There are dense multi-temporal GEDI observations around 51ºN, which makes it possible to continuously monitor the forest structure at same area. Auxiliary data including digital terrain model, slope, forest type and forest phenology data are used to extract GEDI footprints which cover similar forests. We only focus on GEDI footprints with forest height between 25 m and 30 m since most trees in study sites are distributed in this range and they have distinguishable canopies. Sentinel-1 data with fixed relative orbit is used for mitigating the influence of incidence angle. SAR backscatter at filtered GEDI locations are extracted for comparison.We analyse the correlation between the time series of GEDI-derived total structure metrics (e.g., cover), structure metric profiles and Sentinel-1 backscatter metrics (e.g., VH, VV, VV/VH ratio). For deciduous forest, the 15-20 m layer and 20-25 m layer have stronger correlation to Sentinel-1 VH backscatter and VV/VH ratio than other layers as well as the total structure metric. No significant correlation is found between structure metrics and Sentinel-1 backscatter in coniferous forest. We propose to further develop approaches to investigate the joint potential of space-borne lidar and SAR observations to monitor changes in forest canopy structure.

Land ecosystems dampen the increase of atmospheric CO2 by storing carbon in soils and vegetation. In order to estimate how long carbon stays in land ecosystems, a detailed knowledge about the distribution of carbon in different vegetation components is needed. Current Earth observation products provide estimates about total above-ground biomass but do not further separate between carbon stored in trees, understory vegetation, shrubs, grass, litter or woody debris. Here we present an approach in which we link several Earth observation products with a ground-based database to estimate biomass in various vegetation components. Therefore, we use information about the statistical distribution of biomass components provided by the North American Wildland Fuels Database (NAWFD), which are however not available as geocoded data. We use ESA CCI AGB version 3 data from 2010 as a proxy in order to link the NAWFD data to the spatial information from Earth observation products. The biomass and corresponding uncertainty from the ESA CCI AGB and a map of vegetation types are used to select the likely distribution of vegetation biomass components from the set of in-situ measurements of tree biomass. We then apply Isolation Forest outlier detection and bootstrapping for a robust comparison of both datasets and for uncertainty estimation. We use Random Forest and Gaussian Process regression to predict the biomass of trees, shrubs, snags, herbaceous vegetation, coarse and fine woody debris, duff and litter from ESA CCI AGB and land cover, GEDI canopy height, Sentinel-3 LAI and bioclimatic data. The regression models reach high predictive power and allow to also extrapolate to other regions. Our derived estimates of vegetation carbon stock components provide a more detailed view on the land carbon storage and contribute to an improved estimate of potential carbon emissions from respiration, disturbances and fires.

Knowledge about the state of the vegetation at fire occurrence is essential for estimating fire behavior and fire emissions. The spatial distribution and the temporal dynamics of the biomass in various vegetation components, surface litter and woody debris are important controls on fire spread and emissions. So far, no large-scale product exists that combines all these requirements. Maps of canopy height and above ground biomass (AGB) from satellite retrievals provide information on the regional variability of forest biomass, but they have a limited use for fire-related applications because they do not provide information on different fuel components such as biomass in the canopy, wood, grass or litter. Fire-targeted products like the global fuelbed database and the North American Wildland Fuel Database (NAWFD) combine land cover maps with representative values or statistical distributions of fuel properties such as biomass values for trees, shrubs, grass, woody debris and litter. However, those maps do not provide information on the spatial variability of fuel loads within one vegetation type (fuelbed). In addition, information on the temporal dynamics of the fuels is missing. Temporal dynamics of fuels can be approximated by satellite-derived time series of vegetation indices or biophysical parameters. Here, we aim to develop an approach that combines the spatial information from remotely-sensed AGB and canopy height maps, the annual temporal information from land cover maps with the high temporal information from leaf area index (LAI) time series to retrieve information on the spatial variability and temporal dynamics of fuel loads. Therefore, we developed a data-model fusion approach that uses the 10-daily LAI product from Sentinel-3 OLCI and Proba-V and land cover maps as input. We apply the approach to a spatial resolution of 333 x 333 m across different study regions in the Amazon, southern Africa, Siberia and the United States. In a first step, the temporal dynamics in tree height is computed from long-term changes in mean LAI and the fractional tree cover by taking observation of canopy height from GEDI as reference. Since canopy height is closely related to AGB through allometric relations between tree height and biomass, the estimated tree height is then used to estimate stem biomass and consecutively branches and leaf biomass, which is calibrated against maps of AGB. The Biomass and Allometry Database is used to calibrate model parameters that regulate the relationships between canopy height, leaf and woody biomass. Estimated temporal changes in tree height directly translate into changes in stem, branches and leaf biomass and hence result in a carbon turnover (e.g. leaf fall, transfer of woody biomass). Based on a simple decomposition model we then compute the dynamics of surface litter and woody debris. The estimates of litter, fine and coarse woody debris correspond well with databases of in situ observations. Our fuel data-model fusion approach allows estimating spatial patterns and temporal dynamics of vegetation and surface carbon pools for the analysis of carbon turnover and pools and as input into fire behaviour and emission models.

Space-for-time substitution has been used to infer long-term ecological processes such as vegetation dynamics, turnover of species, nutrient cycling, etc. The theory of space-for-time substitution was established to understand temporal processes from contemporary spatial patterns or gradients due to the lack of long-term temporal observations on the response of vegetation to climate change. However, the validity of this theory has been largely debated mostly due to the fact that the fundamental assumption, a climate or an environmental driver of the spatial gradient also drive its temporal change, has not been systematically tested. There is still lack of quantitative understanding of the interaction between climate and vegetation at different spatial and temporal scales. In this study, we used global observations of spatiotemporal changes in several proxies of vegetation (e.g., NDVI) to investigate the link between space and time in the responses of vegetation to climate. We show that the temperature sensitivities of vegetation derived from large scale spatial gradient (space) are highly correlated with the temporal temperature sensitivity (time). Our goal of this study is not only providing quantitively analysis on the spatiotemporal linkage in terrestrial vegetation but also to provide a broader perspective on the methodology that links space and time in understanding the variability of global vegetation.

Vegetation optical depth (VOD) from microwave satellite observations has received much attention in global vegetation studies in recent years due to its relationship to vegetation water content and biomass. We recently have shown that VOD is related to plant productivity, i.e., gross primary production (GPP). Based on this relationship between VOD and GPP, we developed a theory-based machine learning model to estimate global patterns of GPP from passive microwave VOD retrievals. The VOD-GPP model generally showed good agreement with site observations and other global data sets in temporal dynamic but tended to overestimate annual GPP across all latitudes. We hypothesized that the reason for the overestimation is the missing effect of temperature on autotrophic respiration in the theory-based machine learning model. Here we aim to further assess and enhance the robustness of the VOD-GPP model by including the effect of temperature on autotrophic respiration within the machine learning approach and by assessing the interannual variability of the model results with respect to water availability. We used X-band VOD from the VOD Climate Archive (VODCA) data set for estimating GPP and used global state-of-the-art GPP data sets from FLUXCOM and MODIS to assess residuals of the VOD-GPP model with respect to drought conditions as quantified by the Standardized Precipitation and Evaporation Index (SPEI). Our results reveal an improvement in model performance for correlation when including the temperature dependency of autotrophic respiration (average correlation increase of 0.18). This improvement in temporal dynamic is larger for temperate and cold regions than for the tropics. For unbiased root-mean-square error (ubRMSE) and bias, the results are regionally diverse and are compensated in the global average. Improvements are observed in temperate and cold regions, while decreases in performance are obtained mainly in the tropics. The overall improvement when adding temperature was less than expected and thus may only partly explain previously observed differences between the global GPP data sets. On interannual timescales, estimates of the VOD-GPP model agree well with GPP from FLUXCOM and MODIS. We further find that the residuals between VOD-based GPP estimates and the other data sets do not significantly correlate with SPEI, which demonstrates that the VOD-GPP model can capture responses of GPP to water availability even without including additional information on precipitation, soil moisture or evapotranspiration. Exceptions from this rule were found in some regions: significant negative correlations between VOD-GPP residuals and SPEI were observed in the US corn belt, Argentina, eastern Europe, Russia and China, while significant positive correlations were obtained in South America, Africa and Australia. In these regions, the significant correlations may indicate different plant strategies for dealing with variations in water availability. Overall, our findings support the robustness of global microwave-derived estimates of gross primary production for large-scale studies on climate–vegetation interactions.

Terrestrial water storage (TWS) is an integrative hydrological state that is key for our understanding of the global water cycle. The TWS observation from GRACE missions has, therefore, been instrumental in calibration and validation of hydrological models and understanding the variations of the hydrological storages. The models, however, still show significant uncertainties in reproducing observed TWS variations, especially for the interannual variability (IAV) at the global scale. Here, we diagnose the regions dominating the variance of globally integrated TWS IAV, and sources of the errors in two data-driven hydrological models that were calibrated against global TWS, snow water equivalent, evapotranspiration, and runoff data: 1) a parsimonious process-based hydrological model, the Strategies to INtegrate Data and BiogeochemicAl moDels (SINDBAD) framework, and 2) a machine learning-physically based hybrid hydrological model (H2M) that combines a dynamic neural network with a water balance concept. While both models agree with GRACE that global TWS IAV is largely driven by the semi-arid regions of southern Africa, Indian subcontinent and northern Australia, and the humid regions of northern South America and Mekong River Basin, the models still show errors such as overestimation of the observed magnitude of TWS IAV at the global scale. Our analysis identifies modeling error hotspots of the global TWS IAV mostly in the tropical regions including Amazon, sub–Saharan regions, and Southeast Asia, indicating that the regions that dominate global TWS IAV are not necessarily the same as those that dominate the error in global TWS IAV. Excluding those error hotspot regions in the global integration yields large improvements of simulated global TWS IAV, which implies that model improvements can focus on improving processes in these hotspot regions. Further analysis indicates that error hotspot regions are associated with lateral flow dynamics, including both sub-pixel moisture convergence and across pixel lateral river flow, or with interactions between surface processes and groundwater. The association of model deficiencies with land processes that delay the TWS variation could, in part, explain why the models cannot represent the observed lagged response of TWS IAV to precipitation IAV in hotspot regions that manifest to errors in global TWS IAV. Our approach presents a general avenue to better diagnose model simulation errors for global data streams to guide efficient and focused model development for regions and processes that matter the most.

Vegetation attenuates the microwave emission from the land surface. The strength of this attenuation is quantified in models in terms of the parameter vegetation optical depth (VOD) and is influenced by the vegetation mass, structure, water content, and observation wavelength. Earth observation satellite sensors operating in the microwave frequencies are used for global VOD retrievals, enabling the monitoring of vegetation at large scales. VOD has been used to determine above-ground biomass, monitor phenology, or estimate vegetation water status. VOD can be also used for constraining land surface models or modelling wildfires at large scales. Several VOD products exist, differing by frequency/wavelength, sensor, and retrieval algorithm. Numerous studies present correlations or empirical functions between different VOD datasets and vegetation variables such as the normalized difference vegetation index, leaf area index, gross primary production, biomass, vegetation height, or vegetation water content. However, an assessment of the joint impact of land cover, vegetation biomass, leaf area, and moisture status on the VOD signal is challenging and has not yet been done. This study aims to interpret the VOD signal as a multi-variate function of several descriptive vegetation variables. The results will help to select VOD at the most suitable wavelength for specific applications and can guide the development of appropriate observation operators to integrate VOD with large-scale land surface models. Here we use VOD from the Land Parameter Retrieval Model (LPRM) in the Ku, X, and C bands from the harmonized Vegetation Optical Depth Climate Archive (VODCA) dataset and L-band VOD derived from Soil Moisture and Ocean Salinity (SMOS) and Soil Moisture Active Passive (SMAP) sensors. The leaf area index, live-fuel moisture content, above-ground biomass, and land cover are able to explain up to 93 % and 95 % of the variance (Nash–Sutcliffe model efficiency coefficient) in 8-daily and monthly VOD within a multi-variable random forest regression. Thereby, the regression reproduces spatial patterns of L-band VOD and spatial and temporal patterns of Ku-, X-, and C-band VOD. Analyses of accumulated local effects demonstrate that Ku-, X-, and C-band VOD are mostly sensitive to the leaf area index, and L-band VOD is most sensitive to above-ground biomass. However, for all VODs the global relationships with vegetation properties are non-monotonic and complex and differ with land cover type. This indicates that the use of simple global regressions to estimate single vegetation properties (e.g. above-ground biomass) from VOD is over-simplistic.

The moisture content of vegetation canopies controls various ecosystem processes such as plant productivity, transpiration, mortality, and flammability. Leaf moisture content (here defined as the ratio of leaf water mass to leaf dry biomass, or live-fuel moisture content, LFMC) is a vegetation property that is frequently used to estimate flammability and the danger of fire occurrence and spread, and is widely measured at field sites around the globe. LFMC can be retrieved from satellite observations in the visible and infrared domain of the electromagnetic spectrum, which is however hampered by frequent cloud cover or low sun elevation angles. As an alternative, vegetation water content can be estimated from satellite observations in the microwave domain. For example, studies at local and regional scales have demonstrated the link between LFMC and vegetation optical depth (VOD) from passive microwave satellite observations. VOD describes the attenuation of microwaves in the vegetation layer. However, neither were the relations between VOD and LFMC investigated at large or global scales nor has VOD been used to estimate LFMC. Here we aim to estimate LFMC from VOD at large scales, i.e. at coarse spatial resolution, globally, and at daily time steps over past decadal timescales. Therefore, our objectives are: (1) to investigate the relation between VOD from different frequencies and LFMC derived from optical sensors and a global database of LFMC site measurements; (2) to test different model structures to estimate LFMC from VOD; and (3) to apply the best-performing model to estimate LFMC at global scales. Our results show that VOD is medium to highly correlated with LFMC in areas with medium to high coverage of short vegetation (grasslands, croplands, shrublands). Forested areas show on average weak correlations, but the variability in correlations is high. A logistic regression model that uses VOD and additionally leaf area index as predictor to account for canopy biomass reaches the highest performance in estimating LFMC. Applying this model to global VOD and LAI observations allows estimating LFMC globally over decadal time series at daily temporal sampling. The derived estimates of LFMC can be used to assess large-scale patterns and temporal changes in vegetation water status, drought conditions, and fire dynamics.

The frequency and the intensity of drought events have increased during the past decades in some regions, yet the implications of drought for the terrestrial vegetation functioning are not fully understood. In particular, drought in related studies has often been characterized by meteorological conditions rather than the actual soil moisture deficit. Further, previous research focused predominantly on the structural vegetation response, such that the large-scale physiological response remains poorly understood.Here, we analyze and compare the vegetation’s physiological and structural responses to drought across the globe using high-resolution daily TROPOMI sun-induced fluorescence (SIF) as a proxy for productivity, short-wavelength vegetation optical depth (VOD) as a proxy for canopy water conditions and biomass, and near-infrared reflectance of terrestrial vegetation (NIRv) during the period March 2018 - August 2021. Taking advantage of an extended soil moisture record (1979-2021, ERA5-Land reanalysis) we identify and focus on regions where severe soil moisture droughts occurred during our relatively short analysis period. Therein, we quantify and compare the amounts of SIF, VOD and NIRv changes during the early and late drought stages as well as for the recovery period. We also compute the vegetation response to short-term soil moisture changes, i.e. the vegetation sensitivity to short-term soil moisture and the respective changes during the course of droughts. The absolute changes of vegetation indices allow to disentangle physiological and structural responses, while the sensitivity analysis can quantify vegetation responses straightforward to water limitation by accounting for meteorological forcings. To infer vegetation sensitivity, we train random forest regression models at each grid cell, and apply the SHapley Additive exPlanations (SHAP) method to isolate the influence of soil moisture on vegetation from that of other meteorological variables such as temperature, solar radiation, precipitation and vapor pressure deficit. Analyzing the absolute value changes of NIRv, SIF and VOD during the 2018 European drought and the 2020 Russian drought events, we find asynchronous responses of vegetation productivity and vegetation water content. This indicates different hydraulic regulation strategies in response to drought. Moving beyond these case studies, we quantified and averaged the vegetation sensitivity to soil moisture across many severe droughts across the globe, which reveals systematic sensitivity increases during the course of the drought. From an ecological perspective, this indicates increases in ecosystem vulnerability to drought, and can induce feedback in climate and ecosystem services. Vegetation responses to drought differ depending on different vegetation types, climate and soil conditions. In addition, we conduct carbon fluxes from eddy covariance measurements to confirm the vegetation responses to drought derived from Earth observation satellites. In summary, this study provides insights into vegetation responses to droughts at large spatial scales which can help to accurately quantify respective anomalies in terms of land carbon uptake to consequently improve climate projections.

Microwave radiation emitted by the Earth surface is attenuated by vegetation. The magnitude of attenuation is commonly referred to as Vegetation Optical Depth (VOD) and depends on several factors, such as the water content of the vegetation and the wavelength of the sensor. It can be derived from space-borne microwave radiometers and has been explored for multiple applications including drought and biomass monitoring. Here, we present two novel applications based on long-term multi-satellite observations of VOD: 1) the Standardized VOD Index (SVODI) for monitoring vegetation condition and 2) modelling Gross Primary Production (VODCA2GPP). SVODI [1] is a standard normally distributed index indicating the condition of vegetation. Traditional index generation methods struggle to combine data from multiple data sources and often either exhibit a changing variance over time due to changing input sensors, or are biased. We solve these issues with a new probabilistic merging method, which in principle can be applied to other datasets as well. We show that SVODI, which is an indicator of vegetation structure, biomass and water content, supplements existing vegetation condition indexes based on optical data, which mostly represent the greenness of the vegetation. In the presentation, we also assess spatio-temporal patterns and trends of extreme SVODI values. VODCA2GPP [2] is a global GPP dataset derived from VOD using a sink-driven approach. It uses a machine learning framework using the VOD Climate Archive (VODCA), a long-term coherent dataset based on multiple sensors [3], in combination with FLUXNET in situ GPP observations to predict GPP at a global scale for a period of 30 years. VODCA2GPP exhibits similar spatial and temporal patterns as existing observation-based global GPP datasets from FLUXCOM and MODIS, but with a consistent positive bias. While VODCA2GPP reveals no significant global trend, it does show several regional trends, some of which corroborate trends found in the TRENDY model ensemble. Both datasets are available for free from open data repositories. [1] Moesinger, L., Zotta, R.M., van der Schalie, R., Scanlon, T., de Jeu, R.A.M., Dorigo, W. Monitoring Vegetation Condition using Microwave Remote Sensing: The Standardized Vegetation Optical Depth Index SVODI, in rev. [2] Wild, B., Teubner, I., Moesinger, L., Zotta, R.-M., Forkel, M., van der Schalie, R., Sitch, S., Dorigo, W. A. (2021). VODCA2GPP -- A new global, long-term (1988--2020) GPP dataset from microwave remote sensing. Earth System Science Data Discussions, 2021, 1–37. https://doi.org/10.5194/essd-2021-209 [3] Moesinger, L., Dorigo, W., de Jeu, R., van der Schalie, R., Scanlon, T., Teubner, I., and Forkel, M.: The global long-term microwave Vegetation Optical Depth Climate Archive (VODCA), Earth Syst. Sci. Data, 12, 177–196, https://doi.org/10.5194/essd-12-177-2020, 2020.

Terrestrial water storage (TWS) is a socially (e.g., floods and drought) and scientifically (e.g., water and carbon cycles) important variable. Hydrological models have been extensively used to study variations in TWS but the models are showing significant uncertainties, especially for the interannual variability (IAV). It is therefore essential for TWS IAV studies to further improve model accuracy, calling for a better understanding of the TWS IAV simulation error. We conducted a covariance matrix analysis to spatially attribute the contributions to global TWS IAV and its simulation error by two hydrological models: 1) a parsimonious process-based one, implemented in the Strategies to INtegrate Data and BiogeochemicAl moDels (SINDBAD) framework, and 2) a hybrid one, the hybrid hydrological model (H2M), which combines a dynamic neural network and a water balance concept. Both models were calibrated against observation-based data streams for evapotranspiration, snow water equivalent, and runoff, as well as against the Gravity Recovery and Climate Experiment (GRACE) satellite observations of TWS. Both models indicate that the global TWS IAV is largely driven by some regions such as Amazon, Zambezi, Mekong basins, and India. The analysis also identified hotspots of the global TWS IAV error from river basins (e.g., Amazon, Paraná, Congo, and Mekong basins) and inland water bodies (e.g., the Laurentian Great Lakes). Excluding those hotspots in the global integration, the 12-month running mean of the global TWS IAV showed a large improvement: R2 of the global TWS IAV time series was improved from 0.62 to 0.82 for SINDBAD and from 0.62 to 0.88 for H2M. Therefore, the model simulation of the global TWS can efficiently be improved by focusing on correcting the hotspot regions. Comparing GRACE and modelled TWS IAV time series revealed various potential sources of errors, including anthropogenic factors (e.g., reservoir management) and model structure (e.g., insufficient storage capacity and missing storage processes), while the latter was prevalent over many regions. A further comparison to surface water data could characterize the hotspots as areas of 1) more dynamic surface water, and 2) wetlands (i.e., near-inland water bodies). These characteristics of hotspots imply that surface water processes (e.g., seasonal inundation) are relevant for understanding global TWS IAV and are underestimated in the two tested models, calling for further improvement in that respect. Our approach presents a general avenue to identify model simulation errors for global data streams and can guide efficient model development.

Fire risk is projected to increase under future climate change. Most projections focus on fire-prone regions, such as the Mediterranean-type ecosystems, whereas little attention has been paid to regions of low fire risk such as Central Europe. Here, future projections of fire risk which are tailor-made for its specific conditions are scarce. With our study we aim to fill this gap. We use meteorological station data and interpolated climate datasets to compute future fire risk for Central Europe (covering Germany, Poland, the Czech Republic) using the Fire Weather Index. In a next step, we analyse the spatial distribution of reported fire ignitions to identify additional drivers that can explain the spatial pattern of fire ignition and risk, or accelerate fire risk under climate extremes (drought or extreme heat). We analyse how transport infrastructure and proximity to settlements have influenced fire ignition in Central Europe and compare it against relationships known from fire-prone regions. We aim to build on recent adjustments of the FWI to account for respective increased fire risk and apply it to our study area. We conduct high-resolution analysis of fire-risk analysis by computing the FWI along the wildland-urban and wildland-rural interface in individual sites in the study region. In a next step, downscaled future climate scenarios (CMIP6) are applied to compute changes in future fire risk for the entire study area as well as the selected sites in Central Europe. Uncertainty ranges of future fire risk projections will be covered by using several climate scenarios for the entire study region. Detailed analysis will be conducted at the local scale by further refining changes in ignition potentials along the wildland-urban and the wildland-rural interface in selected sites in Central Europe under climate change conditions.

The leaf economics spectrum1,2 and the global spectrum of plant forms and functions3 revealed fundamental axes of variation in plant traits, which represent different ecological strategies that are shaped by the evolutionary development of plant species2. Ecosystem functions depend on environmental conditions and the traits of species that comprise the ecological communities4. However, the axes of variation of ecosystem functions are largely unknown, which limits our understanding of how ecosystems respond as a whole to anthropogenic drivers, climate and environmental variability4,5. Here we derive a set of ecosystem functions6 from a dataset of surface gas exchange measurements across major terrestrial biomes. We find that most of the variability within ecosystem functions (71.8%) is captured by three key axes. The first axis reflects maximum ecosystem productivity and is mostly explained by vegetation structure. The second axis reflects ecosystem water-use strategies and is jointly explained by variation in vegetation height and climate. The third axis, which represents ecosystem carbon-use efficiency, features a gradient related to aridity, and is explained primarily by variation in vegetation structure. We show that two state-of-the-art land surface models reproduce the first and most important axis of ecosystem functions. However, the models tend to simulate more strongly correlated functions than those observed, which limits their ability to accurately predict the full range of responses to environmental changes in carbon, water and energy cycling in terrestrial ecosystems7,8., Three key axes of variation of ecosystem functional changes and their underlying causes are identified from a dataset of surface gas exchange measurements across major terrestrial biomes and climate zones.

Since the late 1970s, space-borne microwave radiometers have been providing measurements of radiation emitted by the Earth’s surface. From these measurements it is possible to derive vegetation optical depth (VOD), a model-based indicator related to the density, biomass, and water content of vegetation. Because of its high temporal resolution and long availability, VOD can be used to monitor short- to long-term changes in vegetation. However, studying long-term VOD dynamics is generally hampered by the relatively short time span covered by the individual microwave sensors. This can potentially be overcome by merging multiple VOD products into a single climate data record. However, combining multiple sensors into a single product is challenging as systematic differences between input products like biases, different temporal and spatial resolutions, and coverage need to be overcome. Here, we present a new series of long-term VOD products, the VOD Climate Archive (VODCA). VODCA combines VOD retrievals that have been derived from multiple sensors (SSM/I, TMI, AMSR-E, WindSat, and AMSR2) using the Land Parameter Retrieval Model. We produce separate VOD products for microwave observations in different spectral bands, namely the Ku-band (period 1987–2017), X-band (1997–2018), and C-band (2002–2018). In this way, our multi-band VOD products preserve the unique characteristics of each frequency with respect to the structural elements of the canopy. Our merging approach builds on an existing approach that is used to merge satellite products of surface soil moisture: first, the data sets are co-calibrated via cumulative distribution function matching using AMSR-E as the scaling reference. To do so, we apply a new matching technique that scales outliers more robustly than ordinary piecewise linear interpolation. Second, we aggregate the data sets by taking the arithmetic mean between temporally overlapping observations of the scaled data. The characteristics of VODCA are assessed for self-consistency and against other products. Using an autocorrelation analysis, we show that the merging of the multiple data sets successfully reduces the random error compared to the input data sets. Spatio-temporal patterns and anomalies of the merged products show consistency between frequencies and with leaf area index observations from the MODIS instrument as well as with Vegetation Continuous Fields from the AVHRR instruments. Long-term trends in Ku-band VODCA show that since 1987 there has been a decline in VOD in the tropics and in large parts of east-central and north Asia, while a substantial increase is observed in India, large parts of Australia, southern Africa, southeastern China, and central North America. In summary, VODCA shows vast potential for monitoring spatial–temporal ecosystem changes as it is sensitive to vegetation water content and unaffected by cloud cover or high sun zenith angles. As such, it complements existing long-term optical indices of greenness and leaf area. The VODCA products (Moesinger et al., 2019) are open access and available under Attribution 4.0 International at https://doi.org/10.5281/zenodo.2575599.

Secondary metabolites play important roles in plant responses to environmental stress but may also represent a large carbon (C) cost, resulting in trade-offs with other C sinks like growth and storage. However, it remains uncertain how such trade-offs may vary with changes in resource availability including water and CO2 availability. We conducted a glasshouse experiment with peppermint (Mentha x piperita L.) exposed to four treatments: control (sufficient irrigation and near-ambient CO2), water deficit (50% irrigation and near-ambient CO2), CO2 limitation (sufficient irrigation and below-ambient CO2) and the combination of water and CO2 deficits. Continuous 13CO2 labelling was used to trace allocation of newly-assimilated C. Concentrations of soluble sugars significantly increased under water deficit but decreased along with aboveground biomass under low CO2, while monoterpene concentrations remained relatively constant, independent of treatments. Under water deficit, there were no differences in allocation of new vs old C to monoterpenes production, structural growth and storage; plants grown under low CO2 even invested proportionally more newly-assimilated C for monoterpenes production, suggesting a preferential allocation to defense at the expense of growth. We concluded that C allocation to monoterpenes is actively regulated in coordination with growth and storage under water and C stresses, consistent with an optimal defense strategy to protect young tissues.

Long-term global monitoring of terrestrial Gross Primary Production (GPP) is crucial for assessing ecosystem response to global climate change. In recent years and decades, great advances in estimating GPP on a global level have been made and many global GPP datasets have been published. These global data records are either based on observations from optical remote sensing, are upscaled from in situ measurements, or rely on process-based models. The different estimation approaches are well established within the scientific community but also exhibit significant discrepancies among each other. Here, we introduce the new VODCA2GPP dataset, which utilizes microwave remote sensing estimates of Vegetation Optical Depth (VOD) to estimate GPP on a global scale. VODCA2GPP is able to complement existing products with long-term GPP estimates covering the period 1988–2020. VODCA2GPP applies a previously developed carbon sink-driven approach (Teubner et al., 2019, 2021) to estimate GPP from the Vegetation Optical Depth Climate Archive (Zotta et al., in prep.; Moesinger et al., 2020), which merges VOD observations from multiple sensors into one long-running, coherent data record. VODCA2GPP was trained and evaluated against FLUXNET in situ observations of GPP and assessed against largely independent state-of-the art GPP datasets (MODIS GPP, FLUXCOM GPP, and GPP estimates from the TRENDY-v7 model ensemble). These assessments show that VODCA2GPP exhibits very similar spatial patterns compared to existing GPP datasets across all biomes but with a consistent positive bias. In terms of temporal dynamics, a high agreement was found for regions outside the humid tropics, with median correlations around 0.75. Concerning anomalies from the long-term climatology, VODCA2GPP correlates well with MODIS and TRENDY-v7 GPP (Pearson’s r: 0.53 and 0.61) but less with FLUXCOM GPP (Pearson’s r: 0.29). A trend analysis for the period 1988–2019 did not exhibit a significant trend in VODCA2GPP on a global scale but rather suggests regionally differing long-term changes in GPP. Significant similar increases of global GPP that were found for VODCA2GPP, MODIS GPP, and the TRENDY-v7 ensemble for the shorter overlapping observation period (2003–2015) supports the theory of elevated CO2 uptake potentially induced by increased atmospheric CO2 concentrations and the associated rising temperatures. The VODCA2GPP dataset is available at TU Data (https://doi.org/10.48436/1k7aj-bdz35; Wild et al., 2021).

The seasonal and longer-term dynamics of fuel accumulation affect fire seasonality and the occurrence of extreme wildfires. Failure to account for their influence may help to explain why state-of-the-art fire models do not simulate the length and timing of the fire season or interannual variability in burnt area well. We investigated the impact of accounting for different timescales of fuel production and accumulation on burnt area using a suite of random forest regression models that included the immediate impact of climate, vegetation, and human influences in a given month and tested the impact of various combinations of antecedent conditions in four productivity-related vegetation indices and in antecedent moisture conditions. Analyses were conducted for the period from 2010 to 2015 inclusive. Inclusion of antecedent vegetation conditions representing fuel build-up led to an improvement of the global, climatological out-of-sample R 2 from 0.579 to 0.701, but the inclusion of antecedent vegetation conditions on timescales ≥ 1 year had no impact on simulated burnt area. Current moisture levels were the dominant influence on fuel drying. Additionally, antecedent moisture levels were important for fuel build-up. The models also enabled the visualisation of interactions between variables, such as the importance of antecedent productivity coupled with instantaneous drying. The length of the period which needs to be considered varies across biomes; fuel-limited regions are sensitive to antecedent conditions that determine fuel build-up over longer time periods (∼ 4 months), while moisture-limited regions are more sensitive to current conditions that regulate fuel drying.

Terrestrial vegetation couples the global water, energy and carbon exchange between the atmosphere and the land surface. Thereby, vegetation productivity is determined by a multitude of energy- and water-related variables. While the emergent sensitivity of productivity to these variables has been inferred from Earth observations, its temporal evolution during the last decades is unclear, as well as potential changes in response to trends in hydro-climatic conditions. In this study, we analyze the changing sensitivity of global vegetation productivity to hydro-climate conditions by using satellite-observed vegetation indices (i.e. NDVI) at the monthly timescale from 1982–2015. Further, we repeat the analysis with simulated leaf area index and gross primary productivity from the TRENDY vegetation models, and contrast the findings with the observation-based results. We train a random forest model to predict anomalies of productivity from a comprehensive set of hydro-meteorological variables (temperature, solar radiation, vapor pressure deficit, surface and root-zone soil moisture and precipitation), and to infer the sensitivity to each of these variables. By training models from temporal independent subsets of the data we detect the evolution of sensitivity across time. Results based on observations show that vegetation sensitivity to energy- and water-related variables has significantly changed in many regions across the globe. In particular we find decreased (increased) sensitivity to temperature in very warm (cold) regions. Thereby, the magnitude of the sensitivity tends to differ between the early and late growing seasons. Likewise, we find changing sensitivity to root-zone soil moisture with increases predominantly in the early growing season and decreases in the late growing season. For better understanding the mechanisms behind the sensitivity changes, we analyse land-cover changes, hydro-climatic trends, and abrupt disturbances (e.g. drought, heatwave events or fires could result in breaking points of sensitivity evolution in the local interpretation). In summary, this study sheds light on how and where vegetation productivity changes its response to the drivers under climate change, which can help to understand possibly resulting changes in spatial and temporal patterns of land carbon uptake.