Super typhoon Haiyan makes landfall
Super typhoon Haiyan
Subscribe here to receive special images

Mission - Vegetation System

Icon VGT


Part Three


The main features that have to be defined for each typical use of the VEGETATION mission are :

  • the spectral properties : spectral bands , their location and bandwidth,
  • the radiometric properties : resolution (expressed in terms of Noise Equivalent Difference of Reflectance : NEdR) and accuracy (related to calibration capabilities)
  • the temporal properties : frequency of data useful for a particular use, time and frequency of data acquisition,
  • the geometric properties : spatial resolution, spatial coverage, viewing angles, accuracies for location and registration (between different bands, between different images)
  • the coherence with high spatial resolution data,
  • particular constraints on delivery, processing systems...

Application of the specifications :

As ancillary data could be used to improve the quality of measurements done by the instruments, some level of product must be defined : the following specifications apply to the first level of data that will be made available to users (Products specifications). For this level, both system and ancillary informations are used to determine correction parameters. They include any knowledge that can be obtained from satellite sensors (attitude, orbit for example), from internal calibration systems as well as from calibration procedure that should be performed by the satellite operator. They also include informations that would be obtained from external data, especially for geometric processing where maps, digital elevation models or a database of ground control points could be used.

However, due to the lack of accuracy of some ancillary informations, the specifications might be difficult to meet or the quality of the data difficult to assess. This is particularly true for atmospheric corrections where the uncertainties on atmospheric properties might lead to uncertainies on ground reflectances that will be larger than the specifications. In that case, the quality will be expressed on the ground reflectance as specified but making the assumption that actual atmospheric effects are those which are computed using the atmospheric corrections procedure.


The spectral properties

For each of the main missions, some specific parameters are important and have to be derived from remote sensing data. To keep the measurements as robust as possible, only wide spectral band measurements (&ap50 nm) are considered and the objectives are to characterise the main features of plant canopies : absorption by chlorophyll, water contents and structural properties. The best and minimal set of spectral bands known to fulfil this need is composed of:

  • a red band centered on the absorption peak of the chlorophyll (0.665 µm),
  • a near infrared band corresponding to the maximum vegetation spectral reflectance and principally related to the structural properties of the canopies and to percentage of soil covered by vegetation,
  • a short wave infrared band centered around 1.65 µm where reflectance is related to water content of the canopy components and to its structure.

Considerations for atmospheric effects characterisation or correction should be added : among different possibilities that are under validation, both the use of adapted vegetation indices computed from red and near infrared reflectances (for example see GEMI) and direct or indirect use of additional spectral bands in the blue region (see ARVI) have to be retained. To provide capabilities of computing or characterising the atmospheric state (aerosols) an additional band can be proposed :

  • a blue band (between .45 and .50 µm) where ground reflectance of vegetation cover is minimal and atmospheric aerosol diffusion effects are maximal.

The influence of atmospheric water vapour, which is most important in a wide near infrared band, can be severely decreased by limiting the upper portion of the near infrared band to avoid the .935 µm water vapour absorption band.

The spectral responses, as presented in annex, will have to be as “similar” as possible to the high resolution instrument bands, at least for spectral bands that participate to the same mission : the red, near infrared and short wave infrared bands.

The radiometric properties

Radiometric properties must be described in terms of :

  • radiometric resolution which gives the smallest radiance or reflectance which should be detected : it will be expressed in terms of Noise Equivalent Difference of Reflectance (NEdR) that should be detected within specified ranges of solar elevation angles and reflectances for each spectral bands,
  • radiometric calibration : to insure that measurements taken in the same image, in different spectral bands or at different times are consistent. This will be specified in terms of
    1. intra-image consistency which could also be specified as an equivalent to a radiometric noise. Variations of calibration within short range (about 10x10km corresponding to the elementary zones) can be considered as high frequency radiometric noise and will be specified as part of the radiometric resolution. Low frequency variations of calibration could be specified either as radiometric noise (specifying values of NEdR for large areas and for the entire image) or as intra-image calibration,
    2. inter-band calibration accuracy and
    3. multitemporal calibration accuracy.

Estimation of radiances or reflectances should also be comparable with measurements obtained from other instruments : this should be insured by the absolute calibration accuracy.

Notes :

  1. Calibration accuracy figures will be given as relative accuracy of reflectances : dR/R. For example, an absolute calibration accuracy of 10% means that a reflectance of 0.10 if known with an accuracy of .01.
  2. As the final data that will be used are surface reflectances, the specifications must be given for surface reflectances. For the detailed specifications of the instrument itself, top of atmosphere (TOA) reflectances and radiances should be used. The transformation between surface and TOA reflectance or radiances can be performed using the 5S code.

Considering existing research work on the use of reflectance properties or derived indices (NDVI, SAVI, MSAVI, GEMI,) and taking into account first optimisations done with system engineers, the following values can be proposed, for sun illuminations corresponding to a sun zenith angle of less than 60° :

ranges for surface reflectance (allowing for saturation for some land covers as snow or bright soils in some conditions or spectral bands and for clouds) should be consistent with usual reflectance values:

  1. from 0.0 to 0.5 for the red band,
  2. from 0.0 to 0.7 for the NIR band and
  3. from 0.0 to 0.6 for the SWIR band.
  4. for the blue band, as it is only envisaged as an experimental band for possible corrections of atmospheric effects on soil and vegetation, typical values of surface reflectances on these two land covers are generally less than 0.5.

surface reflectance resolution of the order of 0.001 to 0.003 should be an objective for the specification with some adjustments for the different bands :

  1. for the red band, as vegetation ground reflectances are low (generally less than 0.1) , a target specification of 0.001 for reflectances of up to 0.1 is desirable, at least for analysis on small blocks of pixels corresponding to areas of about 10x10km. The specified value for NEdR could increase linearly up to 0.003 for reflectances values of about 0.5.
  2. for the NIR and SWIR bands, reflectance differences of 0.003 must be detectable for the entire range of reflectances and either for small blocks or the entire image.
  3. for the blue band, as the variation of TOA reflectance for the extreme conditions of atmospheric conditions (from 5km to 23 km visibility) is about 0.035, differences of about 0.003 should be detectable.

Comment : from existing knowledge it can be assumed that, when atmospheric conditions and directional effects can be accurately estimated, variations of the surface NDVI values of about 0.03 can be related to variations of vegetation fraction cover, biomass or intercepted photosynthetically active radiation that are significant (see Gutman, Ormsby and Tucker for example). For conditions corresponding to the beginning of growth of vegetation, when red and NIR reflectances of 0.1 and 0.2 respectively can be encountered , this means that differences of about 0.0045 should be detected in the red and NIR bands. Taking into account the improvement that will occur in the interpretation schemes in the next three or four years, the above figures for noise specifications seems to be of the order of what should be useful for operational programmes using these types of data.

Intra-image consistency : the calibration consistency within an entire image (large areas) should give a NEdR better than 0.005 for any reflectance value.

Calibration accuracy should be comparable to the high resolution instrument specifications that are within a quality range consistent with existing studies and probable needs for the next ten years :

  1. interband and multitemporal calibration accuracy better than 3%,
  2. absolute accuracy better than 5%

Digitization : digitization must be consistent with the radiometric resolution specification.

The time and frequency of data acquisition

It must be related to the evolution rate of the processes to be characterised, taking into account limitations due to observations from space in the solar energy domain, mainly atmospheric disturbances and cloud coverage. These two factors will force an over sampling in time so that accumulation of acquisitions and screening of cloudy measurements can lead to a “useful” acquisition frequency adapted for vegetation studies. The effect of these factors on acquisition reduction can only be known from statistics on cloud coverage and atmospheric optical thickness, which is varying during the day, with the season and with the geographical location.

To get a minimal cloud cover, the best acquisition time is midmorning as many of the sun synchronous satellite remote sensing systems devoted to land applications (Landsat, SPOT).

Existing operational systems are delivering information on vegetation or meteorological conditions with a period ranging from 5 to 10 days. A mean interval between useful acquisitions to measure changes in vegetation growth can be considered to be about one week : high level products should then be generated to provide data with the usual frequency, the VEGETATION system providing sufficient data to derive the final useful information. To achieve this goal, experience from existing systems shows that actual acquisition should be as much as possible with a frequency of one day, to ensure coverage of the entire land areas each day. Even with this strong constraint, cloud screening will, in some regions and for some periods in the year, significantly decrease the useful acquisition frequency (especially in the tropical regions during the rainy seasons). This is probably the greatest drawback of solar energy measurements and any possibility to keep to the one day interval should be reserved.

Frequency acquisition is strongly related to spatial resolution, number of pixels by line of image and field of view of the instrument. Consistent modifications of these parameters should be discussed with the users to provide the best compromise, current values for the specifications being the preferable combination that was accepted today.

The geometric properties

From an instrument point of view, geometric specifications should be expressed in terms of :

  • sampling rate in two directions,
  • Modulation Tranfer Function for the entire instrument (optical and electronic components),
  • Field of view,
  • accuracies of location and registration for each band relative to the others or to a reference image.

From the users point of view, specifications of spatial resolution, sampling rate and accuracies will define the main users characteristics of the system while other instrumental specifications will be adapted during the instrument design phase taking into account physical or technical constraints.

Spatial resolution and sampling

Two strategies can be used to define spatial resolution and sampling : either to consider the VEGETATION instrument alone or to consider the association with the high spatial resolution instrument. A good number of criteria to make the choice of a particular spatial resolution have been defined (see for example Townshend et al). Both the standard deviation of NDVI values at some particular dates and the standard deviation of time differences of NDVI on some selected sites were chosen and represented on the following figures (from Townshend).

Figure 1 Figure 2
Figure 1 : Standard deviation of NDVI on some selected sites at two different dates Figure 2 : Standard deviation of differences between NDVI values at different dates

These two figures clearly show that the information content decreases as spatial resolution increases but that the decrease in information content from Landsat MSS resolution to a resolution of more than 200 m is much more important than between 200m and 1km. Then, if the first strategy is considered, for one instrument that cannot have high resolution , some information at resolutions of about 200m has to be acquired : this is the MODIS case. In the second strategy and to extrapolate from biophysical models that are only established on “homogeneous” land cover, it is preferable to request some capability for sampling studies using spatial resolutions better than the Landsat MSS resolution and allow some flexibility for the low resolution system. As shown on Figure 1, to keep appropriate information on “Richmond” like zones that are very similar to classical European landscape, it seems desirable to specify a spatial resolution of about 1 km : this is coherent with studies on the European countries done for the MARS project where the AVHRR 1km resolution supplemented by high resolution imagery (Landsat TM or SPOT) proved to be satisfactory, with some problems due to lower resolution obtained for off nadir AVHRR pixels (see INRA report).

Then the requirement for spatial resolution and sampling interval is that it should be about 1 km, with preference for systems that would allow as constant as possible resolution in the entire field of view. The same spatial resolution and sampling specifications apply to all spectral bands.

Spatial resolution is also related to the shape of the MTF of the instrument that should be as high as possible up to Nyquist frequency; specifications usually retained for other instruments should be the objective : the MTF value must be no less than 0.3 up to a frequency corresponding to half the sampling frequency.

The field of view would have to be such that the entire globe can be imaged once each day, especially providing adjacent swath at the Equator. However, that requirement might lead to some problems :

  • a complex design would have to be made to keep the radiometric quality within the specifications due to the high deviation from optical axis,
  • spatial resolution might dramatically decrease due to earth curvature,
  • directional effects due to high zenith observation angles might prevent any useful measurement. (For example , with an altitude of 800 km, an off nadir angle of 50° gives a zenith angle of observation of about 60° that represents a maximum angle for which directional effects are becoming much variable and difficult to handle).


Geometric quality must be expressed on the basis of the particular analyses that will be applied on the images : apart the local distorsion, some order of priority for specifications of the different accuracies can be given :

  1. first, the highest priority should be put on the multispectral registration for spectral analyses or use of multispectral indices like the NDVI or new indices that could be generated using the SWIR or blue bands.
  2. then comes the capability to correctly locate the High Resolution pixels acquired simultaneously, relatively to the VEGETATION pixels,
  3. as the temporal evolution will be one of the most important feature analysed from the VEGETATION data, the multitemporal accuracy should be particularly good,
  4. finally, the absolute location accuracy should allow adequate positionning of each time serie on other maps or geographical information.

Local distortion reflects the sampling accuracy within a small area and can be expressed as a quadratic mean of differences between the actual pixel position and a reference regular position. This characteristic should be consistent with the registration and multitemporal accuracies (see below) and such that the deviation for each spectral band be not more than 0.3 pixel,

Figure 3
Figure 3 : Diagram of errors to be considered for geometric specifications

Each individual spectral relative location for one date is represented by B0, B2, B3, B4 (blue, red, near infrared and short wave infrared spectral bands). The " mean multispectral relative location " is defined as the centre of the smallest circle including all spectral measurements relative locations. The multispectral registration error will be measured as the diameter of that circle : xS.

At one particular date, the corresponding High Resolution " multispectral pixel " will be located at H relatively to its computed location X. The colocation error is the distance between H and the mean multispectral relative location : d .

At different dates, the mean multispectral relative locations will be located inside a smallest circle, the center of which is defined as the " mean multitemporal relative location ". The period that should be considered for the definition of that circle is one year. The multitemporal registration error will be measured as the diameter of that circle : xT.

Finally, the absolute location error will be defined as the distance between the mean multitemporal relative location and the computed location : D.

That particular scheme for definition of geometric errors puts a higher priority on the multitemporal registration than on absolute location of each multispectral pixel. However, the specification for the absolute location error of each multispectral measurement (one particular date) can be inferred from the specifications on xT and .

The following specifications for the errors assume non biased errors and are given as the values of two standard deviations S2() (corresponding approximately to a probability of 5% to have errors larger than the specified value in case of gaussian distribution) :

  • multispectral registration specification : S2(xS) should be significantly less than 0.3 km, with an objective value of 0.1 km,
  • colocation specification : S2(d) should be less than 0.3km,
  • multitemporal registration : S2(xT) should be less than 0.5 km with an objective value of 0.3km,
  • absolute location specification : S2(D) should be less than 1 km with an objective value of 0.5 km.

For the blue band, as it should be used mostly for atmospheric corrections, at least for the first flight model of the instrument, the specifications for xS, d, xT, and Dcan be relaxed to be of the order of 1km.

These specifications apply for the first level of products defined in the Product Definition document.

Coherence with high spatial resolution data

As it is required that high spatial resolution data be used for specific studies together with the VEGETATION data, some constraints have to be defined for the inter-instrument quality :

Collocation of pixels

See above.

Spectral bands consistency

The spectral bands of the two instruments should be as similar as possible, the relative difference between measurements on the same object being not more than 3%. This specification has to be insured on the typical spectral reflectance variations that can be found on bare soils and vegetation canopies.

Spatial coverage

Considering the importance of different areas of the globe both for scientific or applicative project, all the land areas should be imaged by the instrument at any time. Radiometric quality must be met as soon as the solar zenith angle is less than 60°. However, as described in the mission section, some areas could be excluded for the global mission : they are mainly areas covered by snow or ice (Antarctica and Greenland) for which the secondary mission could be performed using local receiving stations specifically installed for that purpose if there is no station dedicated to the vegetation mission itself.

To get daily coverage and then adjacent swath for the lowest possible latitudes, the system will be designed such that some overlap exist for high latitudes : geographical points at high latitudes might be imaged more than once a day but with different sun and viewing geometry. As this difference might give information on the directional properties of the surface, they should not be eliminated at any step of the acquisition-transmission-archiving chain.

Operation modes.

Central receiving system

The entire system should be designed to allow a centralised access to data over the entire globe. Users should find in a single facility :

  • informations on all data acquired and processed to standard levels of processing, especially on their quality (cloud cover). Some information on High Resolution data acquired simultaneously to VEGETATION data should be made available.
  • capability to process and deliver standard products,
  • information related to the use of VEGETATION data sets.

See companion document on Products Specifications for a full description of the products (Ref. 19).

Local receiving stations

As described in the mission section, local receiving capacity should be possible with stations that are “affordable” to small organizations. Some continuity with the existing receiving stations should be provided, taking into account the new technical possibilities (changes in transmission bands, compression...).

As the quality of the data is strongly dependant from system informations on radiometry or geometry, these informations should be made available to the local stations for local processing up to the same data product as in the centralised archiving centre. It is also recommended that a standard preprocessing system be specified and provided for these stations.

Each local station will be responsible for other products processing and delivery as well as for their delivery time.

Degraded modes of operations

In case of failure of part of the system, every effort should be done to associate users through a structured entity similar to the International Users Committee and elaborate a new operating mode. Priorities should be established to provide as much compatible service as possible with the nominal mission. Some guidelines for priorities can already be given :

  • the global coverage feature should be a high priority, leading for example to establish a network of local receiving stations in case of failure of one of the components of the system, allowing centralized access to the entire continental data set,
  • data products similar to elaborate data sets already available from existing systems on the globe should be made avalaible, even if adaptability to particular needs cannot be achieved : for example synthetic data sets over periods of about a decade should be always available, possibly with only one sampling grid or on a predefined partition of the continents,
  • the raw data, together with information on system parameters that should be used to process them, should in any case be archived because historical studies will have utmost importance for any analysis of changes at regional or global scales.

Up Up to Contents Up Back to Part One Up Back to Part Two Continue Continue to Annex