The wheat canopy genome harbors abundant yet untapped genetic variation that could be harnessed to enhance yield potential. The green area index (GAI) is a structural metric that reflects the photosynthetically active canopy surface and is closely linked to final grain yield. Current image-based GAI retrieval methods often suffer from signal saturation and coarse structural depiction, constraining downstream genetic analyses. To address this limitation, we constructed a comprehensive image dataset spanning eight field experiments across China and France, encompassing approximately 600 genotypes under six distinct management regimes. Leveraging this diverse data, we developed a multimodal deep-learning framework augmented by simulated-to-realistic (sim2-real) synthetic data transfer. This framework fuses nadir and oblique RGB images with accumulated thermal time to produce high-precision, time-series GAI estimates. Validated on independent testing datasets from both China and France, the multimodal approach demonstrated robust performance with an accuracy of R 2 = 0.88 and an RMSE of 0.49 m 2 m-2 , representing an improvement of about 22% over the traditional gap fraction method. In three site-year field experiments involving 565 genotypes, the GAI dynamics derived from the multimodal approach showed higher broad-sense heritability (0.20-0.48) than those from the gap fraction approach (0.02-0.13) and stronger genotypic correlations with yield (0.19-0.40 versus 0.09-0.31). Furthermore, genetic analysis confirmed the biological fidelity of the estimated traits, identifying loci that co-localize with known architectural regulators such as Rht-D1, TaTB1-4D, and TaBGC1-4D. Consistently, the multimodal-derived phenotypes were specifically enriched in cell-wall remodeling and hormonal signaling pathways (e.g., brassinosteroid) that directly regulate canopy expansion. Overall, the proposed method offers a powerful tool for unlocking genetic gain in canopy architecture and accelerating canopy-targeted wheat improvement.

Green Area Index (GAI) is a key biophysical trait that underpins crop management decisions and plant breeding programs. Unmanned Aerial Vehicle (UAV)-based remote sensing has emerged as a powerful tool for high-throughput crop phenotyping and precision agriculture, with numerous GAI retrieval methods developed from multispectral imagery. These approaches typically rely on radiance or reflectance calibrated using either a reference panel (PanelCal) or a Downwelling Light Sensor (DLS). However, variable illumination conditions during UAV acquisitions introduce systematic radiometric distortions that degrade retrieval accuracy. To address this limitation, we propose the Spectral Normalization for Illumination-Invariant Calibration (SNIC) method to mitigate illumination-induced artifacts in UAV multispectral imagery. A physically based simulation framework was applied for evaluating radiometric calibration strategies under variable lighting conditions, by coupling the Digital Plant Phenotyping Platform (D3P) with a 3D radiative transfer model, LESS, enabling the generation of wheat canopy reflectance spectra and corresponding GAI values under a wide range of illumination conditions and canopy configurations. Based on the simulated dataset, we also demonstrate that inter-band spectral ratios remain stable under varying illumination conditions, thereby validating the fundamental assumption underlying the SNIC method.GAI retrieval was performed using an XGBoost regression model, and four input strategies (Radiance, PanelCal, DLS, and SNIC) were systematically evaluated on a 6,400-sample in situ dataset. The dataset was split into 3,200 samples under stable illumination conditions and 3,200 samples under variable illumination conditions. Under stable illumination conditions, all methods showed comparable performance, with R2 values of 0.83–0.85 and RMSE values of 0.81–0.83, while DLS performed worse (R2 = 0.71, RMSE = 1.01). Under variable illumination conditions, SNIC consistently outperformed all others, achieving R2= 0.90 and RMSE = 0.63. These results demonstrate that SNIC ensures the robustness of UAV-based GAI retrieval under varying illumination conditions. The proposed method provides a simple yet physically interpretable calibration strategy for enhancing the reliability of UAV phenotyping and precision agriculture applications in real-world operational environments.

Grain filling is the decisive period for rice grain weight formation. However, traditional static traits fail to capture its complex, nonlinear dynamics, while direct panicle weighing is hindered by canopy occlusion. Given the intrinsic synchronization between grain filling and dehydration from anthesis to physiological maturity, monitoring grain moisture content (GMC) dynamics serves as a robust proxy for characterizing the filling process. Here, we propose a high-throughput, physiology-informed phenotyping framework to monitor dehydration. Leveraging a 4-year dataset across 135 cultivar-environment combinations, we demonstrate that the GMC threshold for physiological maturity is relatively stable (≈25%). Concurrently, we developed 2 image-based models for GMC estimation, achieving high accuracies (R2 = 0.82 and 0.86). Integrating this physiological threshold with GMC estimation models enabled the successful reconstruction of the dehydration process. Validation on 26 independent cultivars across 2 sowing dates predicted physiological maturity with a root mean square error of 2.4 to 3.3 d. Traits extracted from these dehydration profiles accounted for 42% of the variance in grain weight, doubling the explanatory power of traditional traits. These gains are largely attributed to a new integrated trait, the moisture maintenance index, which showed a higher and more stable correlation with thousand-grain weight (r = 0.6). This framework offers a scalable approach for monitoring large-scale dehydration dynamics to deepen our understanding of grain weight formation, facilitating the genetic improvement of the filling process to enhance crop yield.

High-throughput phenotyping platforms (HTPPs) are widely used for efficient phenotyping. Near-surface image stitching is one key technology required by HTTPs. However, it is challenged by large-distortion, large-parallax, and homogenous features of images, leading to poor alignment, edge distortions, and ghosting. This study developed an artifact-free, robust, and scalable near-surface image stitching algorithm (NearStitch) with a selfconstructed dataset (PlantStitch). It is characterized by its hierarchical structure, primarily comprising three modules: layered feature extraction, layered image alignment, and warping and post-processing. The PlantStitch comprises 34,299 images of various crop types, growing stages, and altitudes, collected using field gantry phenotyping and unmanned aerial vehicle (UAV) platforms. The adjacent and multi-line stitched results showed that NearStitch had fewer artifacts and less distortion than other commercial software and public algorithms. Besides, NearStitch produced Integrated Local Natural Image Quality Evaluator (ILNIQE) scores of 22.97 for the GS3 dataset (gantry-collected soybean images at 3 m) and 23.12 for the GW3 dataset (gantry-collected wheat images at 3 m). Meanwhile, the average angle difference of checkerboard (AADC) values for the GS3 and GW3 datasets were 1.68 and 0.40, respectively. Compared to the six state-of-the-art (SOTA) methods, the ILNIQE scores improved by approximately 7 % on average, while the AADC scores showed an average improvement of around 10 %. Additionally, the algorithms are transferable to other image data types, such as multispectral and thermal infrared stitching. The NearStitch algorithm and software are publicly available on GitHub. We believe that the PlantStitch and NearStitch, as common image processing datasets and methods, will support image-based crop stress assessment and variety screening in high-throughput plant phenotyping by enabling reliable nearsurface image stitching.

Long-term time series of global leaf area index (LAI) and fraction of absorbed photosynthetic active radiation (FAPAR) are required for characterizing vegetation dynamics in global change studies. The recently developed Copernicus Land Monitoring Service GEOV2-CLMS products were demonstrated to outperform other existing LAI and FAPAR products in terms of completeness, temporal smoothness, consistency across variables and accuracy. However, these GEOV2-CLMS products are derived from the SPOT/VGT and PROBA-V constellation with temporal coverage from 1999 to 2020 which limits its applicability for global change studies. We present here an adaptation of the GEOV2-CLMS algorithm to AVHRR to extend these time series and generate long-term global vegetation products from July 1981 to December 2022. The GEOV2-AVHRR algorithm was specifically designed to maximize the temporal consistency over the successive AVHRR sensors on board NOAA and MetOp-B satellites while keeping high agreement with GEOV2-CLMS products. Neural networks first transform AVHRR surface reflectance into LAI and FAPAR values at the daily time step. The daily estimates are then filtered, smoothed, gap filled and composited every 10-day. GEOV2-AVHRR showed accuracy error between -0.2 and 0.3 LAI and similar to -0.03 FAPAR and uncertainty <1 LAI and similar to 0.10-0.15 for woody and non-woody sites of both DIRECT2.1 and GBOV V3 datasets. GEOV2-AVHRR agreed well with GEOV2-CLMS (MODIS): 92 % (76 %) of land pixels are within +/- max(20 %, 0.5) LAI and 71 % (34 %) within +/- max(10 %, 0.05) FAPAR uncertainty requirements. The gap filling and temporal filters applied in GEOV2-AVHRR proved effective in improving the completeness (only 1 % of missing data) and temporal precision (smoothness) of LAI and FAPAR time series as compared to MODIS. The intra-annual consistency of GEOV2-AVHRR highly agree with GEOV2-CLMS, indicating it is mostly driven by the algorithm. On the contrary, the inter-annual consistency of LAI and FAPAR datasets appears to be very sensitive to the consistency of the input surface reflectance. GEOV2-AVHRR showed high stability as evaluated with MODIS LAI/FAPAR and improves the stability of GEOV2-CLMS. Some residual inter-annual inconsistencies from the transition to sensors are observed for GEOV2-AVHRR as well as for other long term AVHRR datasets (i.e. GIMMS, GLASS and C3S). GEOV2-AVHRR shows overall greening trends in similar to 70 % (similar to 50 % significant at p < 0.05) of land pixels, and the magnitude and spatial pattern of trends highly agree with those of GIMMS.

Vegetation canopy water (VCW) plays one connecting role in the coupling of terrestrial carbon-water cycles, and together with soil moisture, identifying the main changes of the terrestrial ecosystem. With regard to the remote sensing technologies, microwave-based VOD (vegetation optical depth) has been widely used as the VCW proxy. The feature of coarse resolution especially for microwave passive as well as mixing of vegetation water and biomass together would limit its more precise application. In spite of some efforts for the hyperspectral thermal and Global Navigation Satellite Systems (GNSS) limited in regional areas, as well as optical indices and initial efforts for AVHRR and SNAP from optical remote sensing, there are still no global operational and mature VCW product in the science community. To bridge the research gap, this study proposed the unified VCW retrieval algorithm for optical satellites, by improving the methodology developed with some first attempts (e.g., machine learning trained on PROSAIL radiative transfer model simulations). The improvements were implemented by comprehensively parametrizing the VCW related variables (i.e., leaf traits and soil background) in PROSAIL model, based on the largest open integrated global plants (TRY) and soil spectral (OSSL) databases, respectively. In PROSAIL, VCW is expressed as the product of green/leaf area per horizontal ground area (LAI, cm2/cm2) and leaf water content per green area (Cw, g/cm2). In the proposed algorithm, we bridge the quantitative relationship between VCW (LAI *Cw) and simulated TOC reflectance using the machine learning model. The algorithm was assessed for Landsat8 and Sentinel-2, using the ground measurements distributed over diverse climate and biome types worldwide. The results indicate that the developed VCW exhibits satisfactory performance, with R of 0.731 and unbiased RMSE (ubRMSE) of 0.055 g/cm2. Moreover, the proposed VCW achieves reasonable spatial patterns and seasonal changes over diverse vegetation types. The developed VCW product in this study is expected to provide new insights for monitoring global or regional vegetation water variations from optical satellites. With the strength of high spatial resolution compared to the microwave ones in the remote sensing community, the developed VCW would further facilitate the better hydro-ecological applications, especially for the terrestrial carbon-water couplings through vegetation, drought monitoring etc.

Accurate prediction of phenotypes across genotypes and environments is crucial for accelerating crop improvement. Process-based crop growth models (CGMs) can capture complex genotype-by-environment interactions, but their use is limited by labor-intensive genotypic parameter measurements. Here, we developed a faster data assimilation pipeline integrating high-throughput phenotyping (HTP) observations with the SiriusQuality wheat model to efficiently estimate key genotypic parameters and predict genotype performance. Using time-series RGB imagery from a ground-based Phenomobile, we assimilated intercepted photosynthetically active radiation (fIPAR), heading date, and final grain yield to jointly assimilated to calibrate twelve genotypic parameters governing phenology, canopy development, light interception, biomass accumulation, and grain filling. Two data assimilation strategies—a Bayesian DREAM(zs) algorithm and a lookup table (LUT) inversion—were compared through both in silico experiment and eight years of multi-environment field trials of nine durum wheat cultivars. The LUT method demonstrated superior computational efficiency, with prediction accuracy comparable to Bayesian inference on real field data. Multi-year field trials showed that two environments (year / site) were sufficient to reliably characterize genotypic parameters and predict performance across environments. By combining time-series HTP data with ecophysiological modeling, our data assimilation pipeline offers breeders a powerful tool for genotype characterization. It streamlines the process of capturing environmental variance and phenotypic stability, reducing time and effort in crop improvement.

The combination of Sentinel-2 multispectral instrument (MSI) and Landsat 8 operational land imager (OLI) creates a virtual constellation of decametric sensors with high revisiting frequency. However, the differences in the spectral characteristics of the two sensors cause inconsistencies in downstream applications. This study proposed a multiband constraint spectral harmonization method called HARMU. In comparison to existing methods, HARMU uses all the spectral bands in the source sensor to predict the reflectance of the targeting sensor and so fully exploits spectral linkage among different bands. HARMU was specifically implemented by Gaussian process regression (GPR), with training data collected from the spatiotemporally representative BEnchmark Land Multisite ANalysis and Intercomparison of Products 2.1 (BELMANIP2.1) sites. We reproduced the top of the canopy reflectance at both common bands of OLI and MSI and also reflectance at red-edge (RE) bands that are only equipped on MSI. The results indicated that HARMU performed satisfactorily with R(2 )larger than 0.91 and Rel-Bias less than 0.19 for all bands over BELMANIP2.1 sites. HARMU offered similar performances as the widely used Harmonized Landsat and Sentinel-2 (HLS) products: average R(2 )slightly improved from 0.86 for HLS to 0.88 for HARMU for the common bands as evaluated over ground-based observations for validation (GBOV) sites, and additionally, it well reconstructs the missing RE band in HLS-based OLI ( R-2>0.81 and Rel-Bias <0.15). HARMU will substantially contribute to generating spatiotemporally continuous time series of decametric data from the MSI-OLI virtual constellation and monitoring vegetation dynamics in large-scale and long-time sequences.

Multispectral, multi-lens cameras, which acquire spectal images from different individual cameras equipped with different optical filters, are among the most widely used multispectral cameras available on the market. However, their use for close-range sensing is limited by the lack of registration algorithms capable of handling the strong parallax effects observed on scenes with non-negligible relief. In this paper, we propose a method based on stereo camera calibration and disparity estimation to register a close-range multispectral image while retrieving the corresponding 3D point cloud. The method takes advantage of the rigidity of these cameras and the synchronized capture of multispectral bands, both of which are thus compulsory. The algorithm is three-fold. First, the optimal combination of band pair alignments is found. Then, the semi-global matching stereovision algorithm combined with a robust matching cost function are used to align these band pairs and to compute the point cloud. Finally, a pixel filling step that exploits the spectral covariances of the different classes of materials in the image is implemented to limit the number of missing pixels, e.g., due to occlusions. The method was tested on Airphen multispectral images of four plant crops (wheat, sunflower, cover crops and maize) acquired at a distance to the ground ranging from 1.5 to 3 m, thus encompassing a large variability in 3D structure and parallax effects. The results demonstrate that the proposed method achieves better registration performance than six state-of-the-art existing methods, while maintaining a reasonable processing time. Further, the point cloud provides accurate information on the 3D structure of the imaged scene, as shown by the centimetric plant height estimation accuracy. As the point cloud is aligned with the registered multispectral bands, the method provides a 4D (spectral and spatial) description of the scene with a single image, i.e., a multispectral point cloud. This opens up interesting prospects for several applications in close-range sensing including, but not restricted to, vegetation characterization.

Green area index (GAI), leaf chlorophyll content (LCC) and canopy chlorophyll content (CCC) are key variables that are closely related to crop growth. Concurrent and continuous monitoring of GAI, LCC and CCC is critical to keep consistency among variables and make decisions for field precision managements. Previous studies have developed several instruments and algorithms to monitor continuous GAI, while the autonomous monitoring of three variables simultaneously has been lacking. This study presents a novel algorithm to retrieve daily GAI, LCC and CCC from continuous directional observations acquired by a fixed and economic affordable multi-band spectrometer (6 bands covering red, red-edge and near infrared domains) and a photosynthetically active radiation (PAR) sensor in the field. It is composed of three main steps, corresponding to three crucial questions when retrieving variables under natural environments using multi-band spectrometer installed on a near-surface platform: diffuse fraction in each spectral band, radiometric calibration and diurnal sun variation of daily acquisitions. First, we estimated diffuse fraction in each spectral band from the relationship with PAR diffuse fraction based on simulations of the 6S atmospheric radiative transfer model. Second, we computed the relative value of each band to the reference of mean of measurements on all six bands from near-surface measurements, in place of absolute radiometric calibration to limit the influence of changing illumination conditions. In the third step, we combined PROSAIL canopy radiative transfer model and kernel-driven models to retrieved GAI, LCC and CCC from artificial neural network using above spectral diffuse fraction and diurnal multi-angle relative observations. The algorithm was evaluated over 43 IoTA (Internet of things for Agriculture) systems that were installed in 29 wheat fields in France from March to May 2019. Results showed that our method provides good estimates of GAI with root mean square error (RMSE) of 0.54, relative RMSE (RRMSE) of 26.95%, R2 of 0.86, LCC (RMSE = 12.06 mu g/cm2, RRMSE = 33.34%, R2 = 0.52) and CCC (RMSE = 0.23 g/m2, RRMSE = 24.58%, R2 = 0.93). This study shows great potentials for concurrent estimates of GAI, LCC and CCC from continuous ground measurements. It will be useful over other vegetations or other near-surface platforms for simultaneous estimations of biophysical variables.

Monitoring crops with high spatio-temporal resolution satellites provides valuable observations to ensure food security in the global change context. This study focuses on estimating the Green Area Index (GAI) to monitor wheat crops with a spatial resolution of 3 m and daily satellite observations from the SuperDove constellation. With an easier access to large training datasets of ground GAI measurements, and the improvement of the realism of radiative transfer model simulations, the choice of the optimal approach (data -driven or model -driven) constitutes a key question when retrieving GAI from satellite observations. This study compares a data -driven and a model -driven approach to estimate GAI from the SuperDove satellites. Both approaches are based on Gaussian Process Regression (GPR) machine learning techniques. The datadriven approach uses over 300 ground GAI measurements collected from 12 sites in China and France, each with 20 to 51 contrasting plots. The model -driven approach uses 10,000 simulations of top of canopy reflectance and the corresponding GAI values generated by the LESS radiative transfer model applied to 3D scenes built with the ADEL-Wheat (Architectural model of Development based on L -systems) model. Results confirm that the SuperDove reflectance are reliable and consistent with Sentinel -2 values. When estimating GAI using GPR with SuperDove top of canopy reflectance, the model -driven approach (R 2 = 0.83, RMSE = 0.80, Accuracy = 0.01 and Precision = 0.80) generally outperforms the data -driven approach (R 2 = 0.80, RMSE = 0.88, Accuracy = -0.13 and Precision = 0.87), except for small GAI values. In-silico experiments show that the uncertainties in the ground -measured GAI and the size and diversity of the training datasets limit the data -driven approach. In contrast, the model -driven approach is mostly constrained by the realism of the reflectance simulations, particularly for low GAI values. Two ensemble solutions based on the weighted average of the two previous approaches are then proposed: the global ensemble solution (R 2 = 0.86, RMSE = 0.75, A = -0.06 and P = 0.74) where the weight is assumed independent from the GAI values, and the adaptive ensemble solution (R 2 = 0.85, RMSE = 0.76, A = -0.08 and P = 0.76) where the weight depends on the GAI values. Both solutions perform similarly, improving both datadriven and model -driven approaches. Finally, applying both solutions to monitor wheat plots along the growth cycle allows clear differentiation of nitrogen modalities and cultivar effects. However, a minimum plot size of 12 m x 12 m (4 x 4 pixels) is recommended to minimize the co -registration errors and increase estimate precision.

Allometric rules provide insights into the structure-function relationships across species and scales and are commonly used in ecology. The fields of agronomy, plant phenotyping and modeling also need simplifications such as allometric rules to reconcile data at different temporal and spatial levels (organs/canopy). This paper explores the variations in relationships for wheat regarding (i) the distribution of crop green area between leaves and stems, and (ii) the allocation of above-ground biomass between leaves and stems during the vegetative period, using a large dataset covering different years, countries, genotypes and management practices. Our results show that the relationship between leaf and stem area was linear, genotype-specific, and sensitive to radiation. The relationship between leaf and stem biomass depended on genotype and nitrogen fertilization. The mass per area, associating area and biomass for both leaf and stem, varied strongly by developmental stage and was significantly affected by environment and genotype. These allometric rules were evaluated with satisfactory performance, and their potential use is discussed with regard to current phenotyping techniques and plant/crop models. Our results enable the definition of models and minimum datasets required for characterizing diversity panels and making predictions in various G × E × M contexts.

Accurate image segmentation is essential for image-based estimation of vegetation canopy traits, as it minimizes background interference. However, existing segmentation models often lack the generalization ability to effectively tackle both ground-based and aerial images across a wide range of spatial resolutions. To address this limitation, a cross-spatial-resolution image segmentation model for rice crop was trained using the integration of in-situ and in silico multi-resolution images. We collected more than 3,000 RGB images (real set) covering 17 different resolutions reflecting diverse canopy structures, illumination conditions and background in rice fields, with vegetation pixels annotated manually. Using the previously developed Digital Plant Phenotyping Platform, we created a simulated dataset (sim set) including 10,000 RGB images with resolutions ranging from 0.5 to 3.5 mm/pixel, accompanied by corresponding mask labels. By employing a domain adaptation technique, the simulated images were further transformed into visually realistic images while preserving the original labels, creating a simulated-to-realistic dataset (sim2real set). Building upon a SegFormer deep learning model, we demonstrated that training with multi-resolution samples led to more generalized segmentation results than single-resolution training on the real dataset. Our exploration of various integration strategies revealed that a training set of 9,600 sim2real images combined with only 60 real images achieved the same segmentation accuracy as 2,400 real images (IoU = 0.819, F1 = 0.901). Moreover, combining 2,400 real images and 1,200 sim2real images resulted in the best performing model, effective against six challenging situations, such as specular reflections and shadows. Compared with models trained with single-resolution samples and an established model (i.e., VegANN), our model effectively improved the estimation of both green fraction and green area index across spatial resoultions. The strategy of bridging real and simulated data for cross-resolution deep learning model is expected to be applicable to other crops. The best trained model is available at https://github. com/PheniX-Lab/crossGSD-seg.

Allometric rules provide insights into structure-function relationships across species and scales and are commonly used in ecology. The fields of agronomy, plant phenotyping, and modeling also need simplifications such as those provided by allometric rules to reconcile data at different temporal and spatial levels (organs/canopy). This study explores the variations in relationships for wheat in terms of the distribution of crop green area between leaves and stems, and the allocation of above-ground biomass between leaves and stems during the vegetative period, using a large dataset covering different years, countries, genotypes, and management practices. The results showed that the relationship between leaf and stem area was linear, genotype-specific, and sensitive to radiation. The relationship between leaf and stem biomass depended on genotype and nitrogen fertilization. The mass per area, associating area and biomass for both leaf and stem, varied strongly by developmental stage and was significantly affected by environment and genotype. These allometric rules were evaluated and shown to have satisfactory performance, and their potential use is discussed with regard to current phenotyping techniques and plant/crop models. Our results enable the definition of models and minimum datasets required for characterizing diversity panels and making predictions in various genotype × environment × management contexts.

This study aimed to estimate the plant density of early straw cereal crops with spectral reflectance for high-throughput phenotyping. Spectral reflectance was collected in microplot experiments from different sites, between 1-leaf to 3-leaf growth stages and with different density treatments. Plant density was estimated indirectly from spectral reflectance using green fraction (GF) as a proxy. The GF values were extracted from RGB images captured at the same time as the spectral measurements. The results show that the estimation with a 45° observation angle with local calibration on different sites and different growth stages had better accuracy (RMSE=82 plants/m2, rRMSE=0.33) than the others. The band-choosing process showed that using only 4 bands as input resulted in a better balance between model accuracy and simplicity compared to using hundreds of bands.

Essential vegetation variables including leaf area index (LAI), fraction of absorbed photosynthetic active radiation (FAPAR) and fraction of green vegetation cover (FCover) are produced and distributed in the Copernicus Global Land Service. We describe here the algorithmic principles, consistency and improvements of GEOV2, Version 2 of LAI, FAPAR and FCover products derived from SPOT/VGT (1999–2013) and PROBA-V data (2014–2020) at 1 km resolution, as compared to the earlier version GEOV1. GEOV2 is based on neural networks first trained with CYCLOPES and MODIS products to estimate LAI, FAPAR and FCover from daily top of canopy reflectance. Temporal techniques are then applied to filter, smooth, fill gaps and get a composited value every 10 days. Results show that GEOV2 products keep a high consistency with GEOV1 (90% of residuals within ± max(0.5, 20%) LAI, and 80% within ± max(0.05, 10%) FAPAR / FCover) and improves in terms of product completeness (<1% of missing data), temporal consistency, consistency across variables and accuracy.