Difference between revisions of "Remote Sensing"

From Agri4castWiki
Jump to: navigation, search
Line 58: Line 58:
| align="center" style="background:#f0f0f0;"|'''[[NOAA-AVHRR]]'''
| align="center" style="background:#f0f0f0;"|'''[[NOAA-AVHRR]]'''
| align="center" style="background:#f0f0f0;"|'''[[METOP-AVHRR]]'''
| align="center" style="background:#f0f0f0;"|'''[[METOP-AVHRR]]'''
| align="center" style="background:#f0f0f0;"|'''[[SPOT-   VGT]]'''
| align="center" style="background:#f0f0f0;"|'''[[SPOT-VGT]]'''
| align="center" style="background:#f0f0f0;"|'''[[TERRA-MODIS]]'''
| align="center" style="background:#f0f0f0;"|'''[[TERRA-MODIS]]'''

Revision as of 10:29, 9 December 2013


The remote sensing component of the MCYFS basically involves four actions:

  • Data collection: Systematic acquisition of the raw imagery of a number of earth observation (EO) satellites, typically with low resolution but high repetitivity.
  • Pre-processing: Correction of the raw scenes for radiometric, atmospheric and geometric effects, and composition of all the corrected tracks to 10-daily (dekadal) synthesis images (S10).
  • Post-processing: Extraction from the pre-processed S10-syntheses of value-added products useful for vegetation and crop monitoring.
  • Analysis: Use of this information in the final analyses and decision processes concerning crop monitoring.

Over the years, the systematic ingestion and processing of all the EO-data gave rise to a number of “time series” of images, with continental coverage and dekadal frequency. In the MCYFS, the series are systematically extended and every dekad a new set of products is added in near-real time (NRT). The MCYFS uses the image series in three different ways:

  • Qualitative analysis: The mere display of the imagery immediately gives an overview of the general state of the vegetation in a certain area and period. This information is often useful to confirm, adjust or refute the decisions based on the growth modelling approach.
  • Image-derived indicators from the statistical years and the actual information can be included in the statistical yield forecasting process.
  • Remote sensing can directly provide some data needed as inputs by the growth model. For instance, the daily solar radiation can be derived from MSG, and the start of the growing season can be assessed by inspection of annual image series.

The description of the remote sensing component of the MCYFS is split up in three topics:

  • Sensors: An overview of the sensors used and the processing steps from raw data to final products.
  • Products and algoritms: A description of all sensor-independent algorithms and procedures.
  • Output information: A technical desciption of the output products.

Sensors and general processing approach

The 5 EO-Systems used by the MARS project

In line with the objectives of JRC-MARS, the remote sensing data must cover Europe and have to be updated at least every ten days to allow the monitoring of the relatively fast crop growth dynamics. In the MCYFS, all remote sensing products focus on this ten-daily or dekadal step. Currently, only two types of EO-systems can fulfil these requirements:

  • Near-polar orbiting satellites equipped with panoptic wide-swath sensors, such as NOAA/METOP-AVHRR, SPOT-VEGETATION and TERRA-MODIS. These systems provide daily global coverage, at a course resolution of about 1 km (250m for MODIS).
  • Geostationary satellites, which are fixed above a certain point on the equator, at a distance of about 36 000 km. MSG (Meteosat Second Generation) hangs above the point with zero longitude and systematically scans the exposed part of the earth surface, mainly Africa and Europe. The image frequency is high (15-30 minutes) but due to the distance, the spatial resolution is very low (3 km sub-nadir).

In practice the MCYFS exploits the data of the five different EO systems, listed in the table below.

The processing of the EO-data is organised in two parts, namely the pre- and the post-processing.

The pre-processing involves all steps between acquisition of the raw registrations (often called "segments") and the delivery of fully corrected composite images, with a daily (S1) or most often 10-daily (S10) time step. In great lines, this involves the following operations:

  • Geometric corrections: definition of each pixel’s geolocation and remapping of the entire segment image to a common projection.
  • Radiometric corrections: calibration and atmospheric correction to obtain surface reflectances (visible channels), brightness temperatures (thermal bands).
  • Masking: detection of bad observations (clouds, cloud shadows, snow/ice and observation errors) and labelling of the involved pixels in a dedicated image, the “status mask”.
  • Compositing: combining all image segments within a pre-defined period to create a synthesis image according to a compositing rule (e.g. maximum value) and by excluding the bad observations.

The pre-processing software is run on LINUX machines. The concatenation of individual modules into operational 'chains" is realised by the commercial software AppWorx. The post-processing ingests the S1/S10 composites, delivered by the pre-processing, and creates a number of more specific, value-added products in the form of Images, Quicklooks and databases of regional unmixed means (see products and algorithms). The post-processing is performed on Windows PC's with the GLIMPSE software (Global Image Processing Software), and it ends with the delivery of the final products to JRC.

The processing of the data is discussed per sensor in the links below.

Products and algorithms

Temporal:Since April 98 Aug. 81 Jan. 08 Jan. 00
Main periodicities S10 / S30 S10 / S30 S10 / S30 S10 / S30
Spatial:Resolution 1km 1km 1km 250m
Surface reflectances: BLUE + +
RED + + + +
NIR + + +
SWIR + + +
Brightness temperature MIR + +
TIR1 + +
TIR2 + +
Angles: Sun/View (\thetas, \varphis, \thetas, \varphi) + + + +
Quality:Status mask + + + +
Time Grid/Segment_ID + + + +
Nr. good obs. for composite + + + +
Geolocation quality (+DB) + + +
TS Surface Temperature + +
NDVI Normalized Difference VI + + + +
fAPAR Fraction of absorbed PAR + + + +
DMP Dry Matter Productivity + + + +
C. DIFFERENCES to previous year
TS AD / RD + +
NDVI AD / RD + + + +
fAPAR AD / RD + + + +
DMP AD / RD + + + +
D. HISTORICAL Min, max, µ, \sigma, N
TS +
NDVI (+ deciles for VPI) + + +
fAPAR + + +
DMP + + +
E. DIFFERENCES to historical year
TS AD / RD +
NDVI AD / RD / VCI / VPI + + +
fAPAR AD / RD + + +
DMP AD / RD + + +
Phenology: start/end of growing season(s) + + +
Most similar year compared to actual one + + +
Cluster classifications + + +

This section describes the data processing algorithms which are common to several sensors. The algorithms are organised in pre-and post-processing routines.

The table below summarizes the products that are derived from remote sensing for the different near polar-orbiting sensors.


  • AD = Absolute difference.
  • RD = Relative difference.
  • VCI = Vegetation Condition Index (relative position of actual NDVI between historical Min/Max).
  • VPI = Vegetation Productivity Index.

The products are grouped thematically in six categories. The first concerns the image data (A.), which are the result of the pre-processing. The algorithms to derive these products are often sensor-dependent and are treated only in the section of the concerned sensor. The procedures that are common are listed here and they can be consulted through the links below.

  • Geometric correction
    • Chip matching: A procedure to georeference images to a reference images.
    • Quality of the geometric correction: A number of metrics that are derived to assess the quality of the geometric correction.
  • Radiometric operations
    • fAPAR estimation and atmospheric correction: The method of Gobron et al. (2006) for fAPAR retrieval based on uncorrected reflectances applied on SPOT-VGT and TERRA-MODIS data.
    • Land surface temperature
    • Number of frost days
  • Compositing to 10-daily composites

The categories B. to F. in the above table list products that are derived from the 10-daily composites after some preliminary (optional) adaptations, such as flagging and smoothing. The algorithms for the different sensors are identical, but not all procedures are applied on all sensors (e.g. due to the lack of a long time series, no long term average is calculated for METOP-AVHRR). The following list of procedures are explained in the links below.

  • Preliminary adaptations
    • Flagging: Application of the information in the status mask on the images.
    • Smoothing: A method to reduce the noise in the time series per pixel.
  • Vegetation indicators procedures
    • fAPAR: A method based on 10-daily composites of surface reflectances (Weiss et al., 2010) used for NOAA-AVHRR and METOP-AVHRR.
    • fAPAR: A method based on uncorrected reflectances (Gobron et al., 2006) applied on SPOT-VGT and TERRA-MODIS.
    • Dry matter productivity (DMP): A variant of the Monteith approach to estimate the DMP from fAPAR and meteo information.
  • Time domain indicators and operations
    • Monthly composites: The combination of the 10-daily composites into monthly images.
    • Long term statistics: The calculation of the historical year based on the observations of all previous years.
    • Difference images: Differences to the previous year and to the long term average.
  • Time series analysis
    • Similarity analysis: A method to assess for each pixel time series the most similar year.
    • Cluster classification: The identification of pixels showing a similar evolution of a vegetation indicator during the season.

The products are delivered in three different forms: ENVI Raster Images, Quicklooks and databases with Regional Unmixed Means (RUM). The procedure to calculate these RUM-databases is also described in this section, while their format is provided further in Output information.

The products from MSG-SEVIRI are somewhat different and are listed in the following table. The description of the processing algorithms is provided in the section MSG-SEVIRI.

Products from MSG-SEVIRI

Output information

This section provides more details on the region of interest, the map projection, data formats, flagging information and filename conventions. It ends with a summary of all products delivered in the remote sensing component of the MCYFS.