High Resolution NDVI Research Articles

Evaluating Generalization of Methods for Artificially Generating NDVI from UAV RGB Imagery in Vineyards

High-resolution NDVI maps derived from UAV imagery are valuable in precision agriculture, supporting vineyard management decisions such as disease risk and vigor assessments. However, the expense and complexity of multispectral sensors limit their widespread use. In this study, we evaluate Generative Adversarial Network (GAN) approaches—trained on either multispectral-derived or true RGB data—to convert low-cost RGB imagery into NDVI maps. We benchmark these models against simpler, explainable RGB-based indices (RGBVI, vNDVI) using Botrytis bunch rot (BBR) risk and vigor mapping as application-centric tests. Our findings reveal that both multispectral- and RGB-trained GANs can generate NDVI maps suitable for BBR risk modelling, achieving R-squared values between 0.8 and 0.99 on unseen datasets. However, the RGBVI and vNDVI indices often match or exceed the GAN outputs, for vigor mapping. Moreover, model performance varies with sensor differences, vineyard structures, and environmental conditions, underscoring the importance of training data diversity and domain alignment. In highlighting these sensitivities, this application-centric evaluation demonstrates that while GANs can offer a viable NDVI alternative in some scenarios, their real-world utility is not guaranteed. In many cases, simpler RGB-based indices may provide equal or better results, suggesting that the choice of NDVI conversion method should be guided by both application requirements and the underlying characteristics of the subject matter.

Long-term reconstructed vegetation index dataset in China from fused MODIS and Landsat data

The vegetation index is a key satellite-based variable used to monitor global vegetation distribution and growth. However, existing vegetation index datasets face limitations in achieving both high spatial and temporal resolution, restricting their application potential. This study revised a machine learning spatiotemporal fusion model (InENVI) to produce a high-resolution NDVI dataset with 8-day temporal and 30 m spatial resolution, covering China from 2001 to 2020. A total of 432,230 Landsat scenes were processed, enhancing data quality and accuracy. The dataset was validated using 255,000 samples across 6 geographical regions, showing strong performance in capturing spatiotemporal NDVI variations. Additionally, the dataset effectively addresses Scan Line Corrector-off stripes in Landsat 7 imagery. This dataset enables reliable annual NDVI estimates for China at a 30-m resolution and is available for reuse through an open data repository.

High-resolution temporal NDVI data reveal contrasting intratidal, spring-neap and seasonal biomass dynamics in euglenoid- and diatom-dominated biofilms

Intertidal microphytobenthos (MPB) are a major contributor to primary production in estuarine ecosystems. While their biomass is highly variable at multiple spatial and temporal scales, the underlying drivers are as yet little understood. Both in situ sampling and remote-sensing techniques often lack the temporal resolution or coverage to simultaneously capture short-term (intratidal to daily) and longer-term (weekly to annual) biomass changes. Our field setup with in-situ NDVI sensors allowed us to study MPB surface biomass variability at high temporal resolution (10 mins) for up to two years in a freshwater euglenoid dominated mudflat, and a brackish and a marine diatom dominated mudflat. MPB biomass showed marked periodicities at multiple temporal scales: seasonal, spring-neap and intratidal. The diatom-dominated MPB community showed a seasonal biomass peak in winter, while the euglenoid-dominated community showed biomass peaks during spring and summer, probably caused by underlying divergent responses to mainly irradiance, temperature and wind-induced resuspension, and macrobenthos grazing. Spring-neap periodicity likely resulted from differential migratory responses of the MPB communities to variation in timing and duration of daylight exposure. In the freshwater community, upward migration only occurred when exposure duration was sufficiently long (≥4 h). In the diatom-dominated community, morning daylight exposure resulted in highest NDVI values. This study highlights the differences in MPB biomass dynamics between MPB communities within estuarine ecosystems, and underscores the great potential of high-resolution temporal NDVI monitoring for more accurate estimates of MPB biomass and primary production.

Synergistic Impacts of Built-Up Characteristics and Background Climate on Urban Vegetation Phenology: Evidence from Beijing, China

Vegetation is an important strategy for mitigating heat island effects, owed to its shading and evaporative cooling functions. However, urbanization has significantly affected regional vegetation phenology and can potentially weaken the cooling potential of vegetation. Previous studies have mainly focused on national and regional vegetation phenology, but local-scale vegetation phenology and dynamic variations in built-up areas remain unclear. Therefore, this study characterized the vegetation phenology in the densely built-up area of Beijing, China over the period of 2000–2020 based on high-resolution NDVI data using Savitzky–Golay filtering and explored its spatiotemporal characteristics and drivers. The results indicate that the vegetation phenology exhibits significant spatial heterogeneity and clustering characteristics. Compared with vegetation in peripheral blocks, vegetation in central urban blocks generally has an earlier start in the growing season (SOS), later end in the growing season (EOS), and a longer growing season length (GSL). However, the overall distribution of these parameters has experienced a process of decentralization along with urbanization. In terms of drivers, vegetation phenology indicators are mainly influenced by background climate. Specifically, SOS and GSL are mainly affected by temperature (TEP), whereas EOS is mainly influenced by annual precipitation (PRE). Additionally, local environmental factors, particularly the percentage of water body (WAP), also have an impact. Notably, the local environment and background climate have a synergistic effect on vegetation phenology, which is greater than their individual effects. Overall, this study extends the current knowledge on the response of vegetation phenology to urbanization by investigating long-term vegetation phenology dynamics in dense urban areas and provides new insights into the complex interactions between vegetation phenology and built environments.

Bioclimatic predictors of forest structure, composition and phenology in the Paraguayan Dry Chaco

Abstract One of the largest remnants of tropical dry forest is the South American Gran Chaco. A quarter of this biome is in Paraguay, but there have been few studies in the Paraguayan Chaco. The Gran Chaco flora is diverse in structure, function, composition and phenology. Fundamental ecological questions remain in this biome, such as what bioclimatic factors shape the Chaco’s composition, structure and phenology. In this study, we integrated forest inventories from permanent plots with monthly high-resolution NDVI from PlanetScope and historical climate data from WorldClim to identify bioclimatic predictors of forest structure, composition and phenology. We found that bioclimatic variables related to precipitation were correlated with stem density and Pielou evenness index, while temperature-related variables correlated with basal area. The best predictor of forest phenology (NDVI variation) was precipitation lagged by 1 month followed by temperature lagged by 2 months. In the period with most water stress, the phenological response correlates with diversity, height and basal area, showing links with dominance and tree size. Our results indicate that even if the ecology and function of Dry Chaco Forest is characterised by water limitation, temperature has a moderating effect by limiting growth and influencing leaf flush and deciduousness.

Constructing a high-precision precipitation dataset on the data-limited Qinghai-Tibet Plateau

The primary constraint hindering a comprehensive understanding of the atmospheric, hydrological, and ecological dynamics on the Qinghai-Tibet Plateau (QTP) is the absence of highly accurate precipitation datasets. Given the limited availability of climate observations and the pronounced spatial disparities in this region, traditional interpolation techniques are ill-suited, necessitating the development of novel methods. In this study, we formulated a random forest model leveraging precipitation data from 209 climate stations and multisource satellite datasets spanning the period from 1960 to 2020. These datasets encompassed low spatial resolution TRMM precipitation products and high spatial resolution NDVI data. This model facilitated the creation of a monthly precipitation dataset at a 1 km resolution for the QTP. Our findings showed that (1) the prevailing precipitation datasets for the QTP exhibited notable deficiencies, characterized by low spatial precision (R² = 0.592, p = 0.015) and incongruent temporal trends when compared to observational data (NSE = 0.398); (2) by incorporating multisource satellite data, we achieved a substantial enhancement in precipitation accuracy. For instance, the introduction of NDVI led to a reduction in precipitation estimation errors by 7.9 mm and 15.4 mm in desert and nondesert regions, respectively. The inclusion of TRMM satellite precipitation data not only improved accuracy in areas with limited climate stations by more than 16 % but also reduced cumulative monthly errors by 46.8 mm, and (3) the precipitation dataset generated through this method displayed a 24.8 % increase in accuracy (R² = 0.84) compared to existing datasets for the QTP, aligning well with actual precipitation trends (NSE = 0.806). This study offers an effective approach for acquiring high-precision precipitation datasets in regions with limited meteorological stations and substantial spatial heterogeneity in precipitation patterns. Furthermore, our method significantly enhances the spatial heterogeneity of precipitation products, potentially serving as a model for the development of datasets in data-scarce areas.

Vegetation cover change and its response to climate extremes in the Yellow River Basin

Extreme climate events have increased in frequency and severity under the background of climate change, with vegetation growth exhibiting a sensitive response to them. By assimilating GIMMS NDVI and MODIS NDVI using the Residual Network, we obtained a long time series and high resolution NDVI dataset of the Yellow River Basin (YRB). The dataset was utilized for examining the spatiotemporal variability of NDVI and analyzing the response of vegetation cover to climate extremes with meteorological data. Our findings reveal the following: (1) A significant rise in NDVI was seen in the YRB, displaying a mean growth rate of 0.019/10a (p < 0.001). However, seasonal differences exist. The mean NDVI of multi-year declines from southeast to northwest, while the overall trend of vegetation cover improves. (2) The NDVI response to extreme temperature exhibits noticeable spatiotemporal differences. Daytime extreme high temperature in the northern YRB is negatively correlated with NDVI, while they are positively correlated in the lower YRB and the southern part of the middle YRB. Nighttime extreme high temperature exhibits a positive correlation with NDVI. Overall, NDVI displays a stronger response to extreme precipitation than to extreme temperature, with a negative correlation with CWD and a positive correlation with PRCPTOT. (3) The NDVI demonstrates a lagged response to climate extremes in the YRB, with a greater lag in response to extreme temperature than extreme precipitation. The research findings can provide scientific support for the future management and planning of vegetation in the YRB, as well as contribute to the promotion of ecological environment regulation and sustainable development.

Comparative Analysis and Comprehensive Trade-Off of Four Spatiotemporal Fusion Models for NDVI Generation

It is still difficult to obtain high-resolution and fast-updated NDVI data, and spatiotemporal fusion is an effective means to solve this problem. The purpose of this study is to carry out the comparative analysis and comprehensive trade-off of spatiotemporal fusion models for NDVI generation and to provide references for scholars in this field. In this study, four spatiotemporal fusion models (STARFM, ESTARFM, FSDAF, and GF-SG) were selected to carry out NDVI image fusion in grassland, forest, and farmland test areas, and three indicators of root mean square error (RMSE), average difference (AD), and edge feature richness difference (EFRD) were used. A detailed evaluation and analysis of the fusion results and comprehensive trade-off were carried out. The results show that: (1) all four models can predict fine-resolution NDVI images well, but the phenomenon of over-smoothing generally exists, which is more serious in high-heterogeneity areas; (2) GF-SG performed well in the evaluation of the three indicators, with the highest comprehensive trade-off score (CTS) of 0.9658. Followed by ESTARFM (0.9050), FSDAF (0.8901), and STARFM (0.8789); (3) considering the comparative analysis and comprehensive trade-off results of the three test areas and the three indicators, among the four models, GF-SG has the best accuracy in generating NDVI images. GF-SG is capable of constructing NDVI time series data with high spatial and temporal resolution.

RESEARCH ON NDVI NORMALIZATION METHOD BASED ON GF IMAGES

Abstract. The existing NDVI products have problems in terms of low spatial resolution and inconsistent values at a large geographical scale. Based on medium and high-resolution multi-source remote sensing data (GF-1 and GF-2 data), this paper normalized NDVI by combining absolute radiation normalization with relative radiation normalization. And the existing relative radiation normalization method, single-scene global linear normalization (SGloLM) method, is improved to adapt to the production of large-range high-resolution NDVI products. Aiming at the problem of obvious mosaic seams when the SGloLM method is applied to multi-scene images, it is mainly improved from two aspects. One is to improve the coefficient solution of the SGloLM algorithm and propose a new method considering the surrounding multi-scene data, the multi-scene global linear model (MGloLM). The other is to incorporate the Maximum Value Composite (MVC) method to synthesize the maximum value of NDVI at different times in a season, to represent the optimal situation of vegetation growth in the current season. In this study, combined experiments of different methods were performed, as well as qualitative and quantitative evaluations. The experimental results show that SGloLM+MVC and the MGloLM+MVC methods can better eliminate the mosaic seams, and their histogram is most similar to the histogram of standard data, and all quantitative evaluation indexes of SGloLM+MVC are optimal (CC=0.7804, MAD=0.0643, RMSE=0.1012).

Combining target sampling with within field route-optimization to optimise on field yield estimation in viticulture

This paper describes a new approach for yield sampling in viticulture. It combines approaches based on auxiliary information and path optimization to offer more consistent sampling strategies, integrating statistical approaches with computer methods. To achieve this, groups of potential sampling points, comparable according to their auxiliary data values are created. Then, an optimal path is constituted that passes through one point of each group of potential sampling points and minimizes the route distance. This part is performed using constraint programming, a programming paradigm offering tools to deal efficiently with combinatorial problems. The paper presents the formalization of the problem, as well as the tests performed on nine real fields were high resolution NDVI data and medium resolution yield data were available. In addition, tests on simulated data were performed to examine the sensitivity of the approach to field data characteristics such as the correlation between auxiliary data and yield, the spatial auto-correlation of the data among others. The approach does not alter much the results when compared to conventional approaches but greatly reduces sampling time. Results show that, for a given amount of time, combining model sampling and path optimization can give estimation error up to 30% lower for a given amount of time compared to previous methods.

Integration of NDVI Imagery and Crop Coverage Registration System for Apiary Schedule

Abstract Beekeepers need to establish migratory apiaries to benefit from pollen and nectar source plants as in order to increase honey yield. Thus, following the flowering seasons of honey source plants has vital importance when deciding the route of migration. In this study, MODIS imagery was used to generate weekly NDVI data between 1st April to 31st August 2018, when beekeeping activities start and end in the study area. Although MODIS images have high temporal resolution, low spatial resolution (250 meters) makes them insufficient when deciding the crop types and plants. While detecting plants in natural plant areas requires high spatial resolution NDVI, Crop Coverage Registration System (CCRS) parcel-based crop coverage records can enrich the NDVI data without increasing spatial resolution in agricultural lands. Thus, the CCRS data were integrated with NDVI images for migratory beekeeping in agricultural areas as an innovation. To generate both high temporal and spatial resolution, NDVI and CCRS data were integrated together with a beekeeping suitability map to generate the apiary schedule. The results were verified with 176 existing apiary locations and production dates retrieved from field studies which revealed the existence of three seasons in the study area as early and late apiaries (in natural plant areas) and apiaries in agricultural lands. Accuracy analysis showed that 82% of the apiaries intersected with suitable locations and that apiaries in agricultural areas were detected five days earlier than in field studies and obtained more accurately than natural plant apiaries.

"Blend-then-Index" or "Index-then-Blend": A Theoretical Analysis for Generating High-resolution NDVI Time Series by STARFM

"Blend-then-Index" or "Index-then-Blend": A Theoretical Analysis for Generating High-resolution NDVI Time Series by STARFM

WETLAND VEGETATION INTEGRITY ASSESSMENT WITH LOW ALTITUDE MULTISPECTRAL UAV IMAGERY

Abstract. The use of multispectral sensors on Unmanned Aerial Vehicles (UAVs) was until recently too heavy and bulky although this changed in recent times and they are now commercially available. The focus on the usage of these sensors is mostly directed towards the agricultural sector where the focus is on precision farming. Applications of these sensors for mapping of wetland ecosystems are rare. Here, we evaluate the performance of low altitude multispectral UAV imagery to determine the state of wetland vegetation in a localised spatial area. Specifically, NDVI derived from multispectral UAV imagery was used to inform the determination of the integrity of the wetland vegetation. Furthermore, we tested different software applications for the processing of the imagery. The advantages and disadvantages we experienced of these applications are also shortly presented in this paper. A JAG-M fixed-wing imaging system equipped with a MicaScene RedEdge multispectral camera were utilised for the survey. A single surveying campaign was undertaken in early autumn of a 17 ha study area at the Kameelzynkraal farm, Gauteng Province, South Africa. Structure-from-motion photogrammetry software was used to reconstruct the camera position’s and terrain features to derive a high resolution orthoretified mosaic. MicaSense Atlas cloud-based data platform, Pix4D and PhotoScan were utilised for the processing. The WET-Health level one methodology was followed for the vegetation assessment, where wetland health is a measure of the deviation of a wetland’s structure and function from its natural reference condition. An on-site evaluation of the vegetation integrity was first completed. Disturbance classes were then mapped using the high resolution multispectral orthoimages and NDVI. The WET-Health vegetation module completed with the aid of the multispectral UAV products indicated that the vegetation of the wetland is largely modified (“D” PES Category) and that the condition is expected to deteriorate (change score) in the future. However a lower impact score were determined utilising the multispectral UAV imagery and NDVI. The result is a more accurate estimation of the impacts in the wetland.

Monitoring Irrigation Consumption Using High Resolution NDVI Image Time Series: Calibration and Validation in the Kairouan Plain (Tunisia)

Water scarcity is one of the main factors limiting agricultural development in semi-arid areas. Remote sensing has long been used as an input for crop water balance monitoring. The increasing availability of high resolution high repetitivity remote sensing (forthcoming Sentinel-2 mission) offers an unprecedented opportunity to improve this monitoring. In this study, regional crop water consumption was estimated with the SAMIR software (SAtellite Monitoring of IRrigation) using the FAO-56 dual crop coefficient water balance model fed with high resolution NDVI image time series providing estimates of both the actual basal crop coefficient and the vegetation fraction cover. Three time series of SPOT5 images have been acquired over an irrigated area in central Tunisia along with a SPOT4 time series acquired in the frame of the SPOT4-Take5 experiment, which occurred during the first half of 2013. Using invariant objects located in the scene, normalization of the SPOT5 time series was realized based on the SPOT4-Take5 time series. Hence, a NDVI time profile was generated for each pixel. The operationality and accuracy of the SAMIR tool was assessed at both plot scale (calibration based on evapotranspiration ground measurements) and perimeter scale (irrigation volumes) when several land use types, irrigation and agricultural practices are intertwined in a given landscape. Results at plot scale gave after calibration an average Nash efficiency of 0.57 between observed and modeled evapotranspiration for two plots (barley and wheat). When aggregated for the whole season, modeled irrigation volumes at perimeter scale for all campaigns were close to observed ones (resp. 135 and 121 mm, overestimation of 11.5%). However, spatialized evapotranspiration and irrigation volumes need to be improved at finer timescales.

An Improved Method for Producing High Spatial-Resolution NDVI Time Series Datasets with Multi-Temporal MODIS NDVI Data and Landsat TM/ETM+ Images

Due to technical limitations, it is impossible to have high resolution in both spatial and temporal dimensions for current NDVI datasets. Therefore, several methods are developed to produce high resolution (spatial and temporal) NDVI time-series datasets, which face some limitations including high computation loads and unreasonable assumptions. In this study, an unmixing-based method, NDVI Linear Mixing Growth Model (NDVI-LMGM), is proposed to achieve the goal of accurately and efficiently blending MODIS NDVI time-series data and multi-temporal Landsat TM/ETM+ images. This method firstly unmixes the NDVI temporal changes in MODIS time-series to different land cover types and then uses unmixed NDVI temporal changes to predict Landsat-like NDVI dataset. The test over a forest site shows high accuracy (average difference: −0.0070; average absolute difference: 0.0228; and average absolute relative difference: 4.02%) and computation efficiency of NDVI-LMGM (31 seconds using a personal computer). Experiments over more complex landscape and long-term time-series demonstrated that NDVI-LMGM performs well in each stage of vegetation growing season and is robust in regions with contrasting spatial and spatial variations. Comparisons between NDVI-LMGM and current methods (i.e., Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM), Enhanced STARFM (ESTARFM) and Weighted Linear Model (WLM)) show that NDVI-LMGM is more accurate and efficient than current methods. The proposed method will benefit land surface process research, which requires a dense NDVI time-series dataset with high spatial resolution.

Detecting mortality induced structural and functional changes in a piñon-juniper woodland using Landsat and RapidEye time series

Piñon–juniper (PJ) woodlands have recently undergone dramatic drought-induced mortality, triggering broad scale structural changes in this extensive Southwestern US biome. Given that climate projections for the region suggest widespread conifer mortality is likely to continue into the next century, it is critical to better understand how this climate-induced change in vegetation structure alters ecosystem function (e.g., productivity, biogeochemical cycling, energy partitioning) in PJ woodlands. Data from satellite remote sensing could potentially help to gain some of this understanding. However, relatively little is known about the suitability of satellite remote sensing for monitoring mortality induced structural and functional changes in PJ woodlands, a semi-arid biome characterized by sparse vegetation that complicates remote sensing of vegetation properties. Here we examined the potential role of satellite remote sensing to better understand structural and functional changes that take place in PJ woodlands as they respond to and recover from climate induced mortality events. We used time series medium (30×30m) and high (≤5×5m) spatial resolution satellite data in concert with eddy covariance-based gross primary productivity (GPP) data collected in a PJ woodland mortality manipulation near Mountainair, NM. Our results showed that both high and medium resolution satellite remote sensing data can provide ecosystem level assessments of overall canopy mortality and subsequent regrowth in these semi-arid woodlands. However, high spatial resolution data allowed detecting the disturbance almost a full year earlier and enabled us to describe the spatially heterogeneous patterns of recovery typical for such complex. The ability of satellite remote sensing to be used to provide information about changes in ecosystem function was conditional on the presence of sufficient soil moisture; when soil moisture was present the strength of the linear relationship between GPP and spectral vegetation indices improved at the manipulated site (e.g. for high resolution NDVI data R2=0.13 during drought conditions, R2=0.46 when wet, P<.05) and less so at a nearby control site (R2=0.09 during drought conditions, compared to R2=0.22 during growing conditions, P<.05). Thus, although satellite time series can be used to monitor and improve our understanding of canopy function in both disturbed and undisturbed semi-arid coniferous biomes, combining in situ eddy covariance data with satellite time series demonstrated that the utility of these datasets is highest during periods of peak growing conditions. This study therefore provides a benchmark for understanding how spatial and temporal variation in vegetation structure and function can be best studied as semi-arid coniferous biomes respond to future climate variability.

Method to construct high spatial and temporal resolution NDVI DataSet-STAVFM

Method to construct high spatial and temporal resolution NDVI DataSet-STAVFM

Geostatistics for Mapping Leaf Area Index over a Cropland Landscape: Efficiency Sampling Assessment

This paper evaluates the performance of spatial methods to estimate leaf area index (LAI) fields from ground-based measurements at high-spatial resolution over a cropland landscape. Three geostatistical model variants of the kriging technique, the ordinary kriging (OK), the collocated cokriging (CKC) and kriging with an external drift (KED) are used. The study focused on the influence of the spatial sampling protocol, auxiliary information, and spatial resolution in the estimates. The main advantage of these models lies in the possibility of considering the spatial dependence of the data and, in the case of the KED and CKC, the auxiliary information for each location used for prediction purposes. A high-resolution NDVI image computed from SPOT TOA reflectance data is used as an auxiliary variable in LAI predictions. The CKC and KED predictions have proven the relevance of the auxiliary information to reproduce the spatial pattern at local scales, proving the KED model to be the best estimator when a non-stationary trend is observed. Advantages and limitations of the methods in LAI field predictions for two systematic and two stratified spatial samplings are discussed for high (20 m), medium (300 m) and coarse (1 km) spatial scales. The KED has exhibited the best observed local accuracy for all the spatial samplings. Meanwhile, the OK model provides comparable results when a well stratified sampling scheme is considered by land cover.

水文・農業分野における高分解能マルチスペクトル衛星画像の利用について

This paper examines the applicability of high-resolution multispectral satellite imagery in hydrological and agricultural areas. The applicability was investigated via three QuickBird multispectral images of Osaka, which were acquired during summer and winter 2005-2006. In a supervised classification, shadows of ground obstacles were occasionally misclassified as surface water. This problem could be avoided using summer images and a filter. Vegetations were easily separated from non-vegetation, but separation of agricultural crops from forests was somewhat difficult. NDVI from QuickBird image well described the vegetation conditions in small plot-size fields in Japan. Thus, high-resolution NDVI should be a valuable source for agricultural land and water managements. A sensitivity analysis indicated that atmospheric correction is recommended if absolute value of NDVI is important, although atmospherically uncorrected NDVI is sufficient for relative analysis within a single image.

Modeling temporal changes in surface spatial heterogeneity over an agricultural site

High temporal frequency remote sensing observations are required to monitor vegetation functioning. These observations are currently provided by moderate resolution sensor (with pixel size ranging from 250 m to 10 km). However, the intra-pixel spatial heterogeneity which may be important at moderate resolution, induces a scaling bias on non-linear estimation processes of land surface variable. A possible strategy to correct this scaling bias consists in using variogram model of high spatial resolution data (e.g. SPOT/HRV 20 m) as a proxy for the spatial heterogeneity within moderate resolution pixel. However, ways have to be found to get prior knowledge of this intra-pixel spatial heterogeneity metric without systematic concurrent high spatial resolution images. This paper aims at proposing a spatio-temporal model of the variogram of high spatial resolution data, which enables us to retrieve the spatial heterogeneity within moderate spatial resolution pixel using a temporal sampling of few high spatial resolution scenes. It capitalizes on variogram modeling of a time series of high spatial resolution NDVI images to quantify and model the temporal changes in landscape spatial heterogeneity over a particular crop site (Fundulea, Romania). We first demonstrate that important temporal variations in surface spatial heterogeneity observed over an agricultural site mainly result from the shift in seasonal trajectories between crop classes (here winter versus summer crops). The mean length scale as measured by the variogram integral range is mainly influenced by the gathering of fields with similar NDVI values. The scene overall spatial variability and the spatial heterogeneity within moderate resolution pixels, as quantified by the variogram sill and the dispersion variance, respectively, increase with the difference in NDVI between winter and summer crops. The influence of surface spatial heterogeneity on the description of land surface processes is thus critical when the phenological variability between crop classes is maximum, suggesting that the number of high spatial resolution scenes should increase at these periods of the seasonal cycle. Then, based on these observations, we build a model describing the temporal course of surface spatial heterogeneity, i.e. the temporal trajectory of the variogram of high spatial resolution NDVI image, as a function of crop seasonality. Once calibrated from a temporal sampling of few high spatial resolution scenes, this model proves to be powerful to predict the variogram at a date at which the high spatial resolution scene is not available and thus to retrieve the spatial heterogeneity within moderate resolution pixels through the seasonal cycle within a mean relative uncertainty of 20%.