From Agri4castWiki
Jump to: navigation, search

Main principles

The satellite platforms TERRA and AQUA are part of the American Earth Observation System (EOS). Both follow a standard, near-polar, sun-synchronous orbit at a 705 km altitude and a frequency of about 14.6 cycles per day, with the following differences:

  • TERRA: launched in Dec 1999, morning orbit (10h30’ at equator, descending node).
  • AQUA: launched in May 2002, afternoon orbit (13h30’ at equator, ascending node).

Both platforms carry the same MODIS-sensor (MODerate resolution Imaging Spectrometer), which simultaneously operates in 3 different modes:

  • 1 km resolution, with 29 bands covering the full spectral range from Visible to Thermal Infrared (400 – 14 000nm).
  • 500 m resolution, with 5 bands from the Visible to the Shortwave Infrared (450 - 2200nm).
  • 250 m resolution, with 2 bands: RED (620-670nm) and NIR (841-867nm).

Characteristics of the spectral bands of MODIS

Characteristics of the spectral bands of MODIS

Thanks to its large swath width of 2330 km, MODIS provides global coverage every 1 - 2 days. The MODIS programme consists of different Discipline Groups for Atmosphere, Oceans, Land and Calibration all providing different MODIS products ranging from absolutely raw “granules” to completely processed composites. The compositing periods used by the MODIS programme are defined on 8 and 16 days for most of the products. More information on MODIS can be found on the MODIS website (see Links).

Input data for MCYFS

Since the compositing steps of 8/16 days is not compatible with the practices of the MCYFS, an alternative processing chain was set up starting from the Level1 segments at 250m spatial resolution providing 10-daily composite data, fully compatible with the other data used in MCYFS. For MCYFS, only the RED, NIR, SWIR and BLUE bands are processed, together with the geolocation data for remapping, and the cloud mask (see table). All data are from Collection 5, provided by NASA-LAADS.

The MODIS files needed by the processing chain are summarised in the table below.


Further details on the used datasets:

  • MOD02QKM:
- dataset EV_250_RefSB: The RED and NIR data are converted to TOA reflectance values with error codes being replaced by a -1, which is the ”no value” code adopted for the chain.
  • MOD02HKM:
- dataset EV_250_RefSB: as for the 250 m data; RED, NIR and BLUE band and -1 for “no value”
  • MOD03:
- datasets Longitude and Latitude: The 1 km resolution (Lon,Lat) grid is “resampled” to 250 m which is the resolution of the radiometric data.
- datasets SensorZenith and SensorAzimuth: Azimuth values are converted to values in the [0°..360°] range and resampled to 250 m.
- datasets SolarZenith and SolarAzimuth: Azimuth values are converted to values in the [0°..360°] range and resampled to 250 m.

Note that the resampling takes into account that the input data consists of swaths of 40 lines, by not interpolating data associated with different swaths, i.c. at the bottom of each swath there are 3 lines being extrapolated rather than interpolated (see shaded area in the figure below). Furthermore, the MODIS interband registration (see Figure below) is correctly taken into account by resampling, while a ‘best fit’ registration is used by explosion (i.e. the top-left corners of the first 1km pixel and the ‘virtual’ 250m pixel (-1,0) are assumed to coincide).


The data are received through a data subscription for MODIS data. Every day, NASA makes the data subset within the requested ROI available for ftp download and then sends an e-mail with the information needed to access the data. An automated procedure at VITO then processes these e-mails, downloads the data and reorganizes them in separate directories for each day. The number of images received and possible missing products are written to a log file which can be used by an operator to check if the data are being downloaded correctly.

Processing of the imagery of TERRA-MODIS (since March 2000). The Level-0 products are acquired from NASA-LAADS (collection 5).


During the preprocessing, the input files are imported and converted to ENVI files. These are then used to calculate TOC reflectances and fAPAR using a method published by Gobron et al. (2006) (JRC-IES GEM unit). The results are then remapped to the geographical Lat-Lon system aka Plate Carree projection. In a final step the results are composited to dekadal products. These steps are explained below in the following sections:


The import step extracts the necessary data from the HDF-EOS files and converts these to ENVI files. The products received from NASA use internal data compression. As compression is part of the HDF standard this has no implications for the user whom can open and read from these files as if they were uncompressed. However it has a significant drawback in terms of processing time. The files are therefore uncompressed before the actual import step using “hrepack”, a HDF internal compression/uncompression and/or chunking tool.

Geometric correction

The MODIS input data are delivered with Latitude and Longitude planes for geometric correction.

By means of the Lon/Lat-planes obtained from the MOD03 products, the “unpacked” images are remapped to the WGS84-Geographical Lat/Lon system (“unprojected”), with a step of 1°/112/4. In this way, the framing of the MODIS pixels is compatible with the other sensors, thus each VGT or AVHRR pixel is exactly covered by 4x4 MODIS pixels. A “backward” remapping approach is used, i.e. for each pixel position in the result space the “nearest neighbour” pixel in the input space is searched using the available (Lon,Lat) grid and all the data associated with this pixel is then copied to the result space.

The algorithm used has been reviewed several times in order to find the most effective way to handle:

  • bow-tie effects
  • the fact that successive swaths partly ‘overlap’ each other (i.e. they partly contain data for the same region on earth)

The results of the geometric correction consist of geo-corrected RED and NIR reflectance data, angle data, an updated status map and a coverage map.

Radiometric operations

Radiometric calibration

The MODIS input data consist of calibrated granules.

Atmospheric correction

The atmospheric correction of RED and NIR is done simultaneously with the estimation of the fAPAR using the method of Gobron et al. (2006). When applying this method, so-called rectified reflectance values for the RED and NIR bands are obtained. These rectified bands are surface reflectance values corrected for atmospheric and angular disturbances, and are used to calculate the NDVI. The present procedure generates the rectified RED and NIR reflectances at 250m spatial resolution using ‘rectification factors’ based on MODIS data in the BLUE, RED and NIR bands at 500m. More information on the method of Gobron for the estimation of fAPAR can be found here. The SWIR band is treated separately and atmo-corrected with the classical SMAC-approach (Rahman and Dedieu, 1994). The BLUE-band is not atmo-corrected, so its reflectances remain at TOA-level.


The cloud detection result of MODIS is used (MOD35_L2). It was decided not to use the 250m cloud flags in bytes 5 and 6, but the 1km flags in byte1 with both “cloudy” and “uncertain” being output as “cloud” in the generated status map. The reason is that the 250m cloud flags are only based on RED and NIR tests and do not contain information about the presence of ice/snow.


All layers are composited to 10-daily composites using a constrained maximum fAPAR approach. The constraints used are:

  • Cloudy and snow observations are avoided
  • Viewing zenith angle
  • Solar zenith angle

For the MODIS compositing special attention was paid to selection of the view zenith angle (VZA) criterion. Experimental evidence indicated that it is not advisable to increase the VZA criterion to values in excess of 35°. As auxiliary data a Land/sea mask is needed, which is derived from GLC2000 map.

As quality indicators, a number of additional images are created:

  • the number of good observations per compositing period
  • the time grid including segment ID
  • geolocation quality

The latter is realized by applying a chip matching procedure using fixed geolocation points for which the distortions are calculated. More information is provided on the page quality of the compositing. A detailed description of the compositing method can be found in the section 10-daily composites A new status map is created after the compositing. More information on the meaning of the bits can be found below.

Meaning of the status mask bits



All algorithms and procedures of the post-processing are explained in detail in the concerned section. A short description of the TERRA-MODIS post-processing products is provided here.

Preliminary adaptations


The information contained in the status mask, derived from the 10-daily compositing, is applied on all composite layers, such that the flag information is easily available for further processing. The procedure is described here.


Smoothing reduces the noise in a pixel’s time series and results in time profiles that are more close to reality instead of containing e.g. undetected clouds. The operation is optional and can only be applied on vegetation indicators, because they have a gradual seasonal evolution. Reflectances cannot be smoothed. The method is described in the section Smoothing.

Vegetation indicators


In the case of TERRA-MODIS, the fAPAR was calculated during the pre-processing, simultaneously with the atmospheric correction (see Radiometric corrections). The procedure is described in the section fAPAR. The fAPAR is also used in the estimation of the DMP after smoothing.


Dry Matter Productivity (DMP) expresses the primary productivity (expressed in g DM/ha/day) in a 10-daily time step. The method uses smoothed fAPAR and meteo parameters (from Weather Monitoring), and is based on the Monteith equation explained in detail in the section DMP.

Time domain indicators and operations

Monthly composites

The smoothed and non-smoothed vegetation indicators (NDVI, fAPAR and DMP) are further aggregated to monthly composites. A different method is used than for the 10-daily composites and is described here.

Differences to previous year

Two difference operators (absolute and relative) are used to calculate the difference between the current composite and the corresponding one from the previous year. The difference operators are discussed in the section Difference Images.

Historical year

The historical year is a kind of climatology derived from the entire time series of vegetation indicators (NDVI, fAPAR and DMP) derived from TERRA-MODIS since 2000. The historical year is updated every two years. It contains the average value, the standard deviation, minimum , maximum and deciles per 10% per composite. More information is provided in the section Long Term Statistics.

Difference to historical year

The same difference operators are applied between the current composite and the historical year for the vegetation indicators NDVI, fAPAR and DMP. For NDVI, two additional differences are calculated, i.e. the Vegetation Condition Index (VCI) and the Vegetation Productivity Index (VPI). The methods are described in the section Difference Images.

Time series analysis

Similarity analysis

The purpose of the similarity analysis is to identify the most similar year compared to the current year. This analysis is performed on the time series evolution per pixel since last October or last March. Only pixels from the classes ‘arable land’, ‘pasture’ and ‘rice’ are considered. The similarity analysis is applied on the smoothed time series of NDVI and fAPAR. The method is described in the section Time series analysis.

Cluster classification

The purpose of the cluster classification is to group pixels which have a similar evolution during the current season. pixels from the classes ‘arable land’, ‘pasture’ and ‘rice’ are considered. It is applied on the smoothed time series of NDVI and fAPAR. The method is described in the section Time series analysis.

Databases with Regional Unmixed Means

Some of the above described products are also delivered in the form of Regional Unmixed Means (RUM) which are then added to the database of the Marsop viewer. Briefly described, RUM-values are derived by averaging (part of) the pixels of a certain administrative region which belong to a certain land cover class. In this way, the derived database can be ingested in a GIS-system. The method to derive the RUM-values is described in the section Databases with Regional Unmixed Means (RUM-values).

Products overview

The table below provides an overview of all 10-daily and monthly composite products that are generated for MCYFS from TERRA-MODIS input data.

Ten-daily and monthly composited data are generated in the form of images (I), quicklooks (Q) and databases with RUM-values (R).


More information on the image format, headers and rescaling can be found in the section Images. The column V[D] refers to the filename suffix, and is explained in the section on Filename conventions.