ESA Earth Home Missions Data Products Resources Applications
EO Data Access
How to Apply
How to Access
FRINGE '96 Workshop: ERS SAR Interferometry, 30 September - 2 October 1996
Site Map
Frequently asked questions
Terms of use
Contact us



The use of interferometric results with other remote sensing data in the EMAP-project

Manfred Reich Universität Stuttgart Institut für Navigation Keplerstr. 11 70174 Stuttgart Germany
Alois Wehr Universität Stuttgart Institut für Navigation Keplerstr. 11 70174 Stuttgart Germany


The purpose of the project EMAP (ERS-1/ERS-2 SAR DATA FOR MONITORING AGRICULTURAL LAND USE AS A LONG-TERM PROJECT) is to analyze ERS-1/-2 SAR data as a tool for agricultural crop monitoring. The focus is to apply the data for crop identification. This analysis is supported by combining ERS SAR data with optical satellite images and non remote sensing data in a Geographical Information System. The investigations are carried out in 4 different test sites in Germany, where we have intensive agricultural land use but different characteristics in land use. The aim is to detect parameters which significantly support the crop identification and classification process and to integrate these together in an expert system consisting of remote sensing data (optical and SAR) and GIS-information. The information extracted from interferometric processing of the SAR-data are the coherence properties as a function of seasonal changes in land use and the monitoring of agricultural land management activities by differential interferometry. Therefore the interferometric processing of ERS-1/ERS-2 SAR.SLC data can improve the quality of multi-sensoral / multi-temporal classification.
The EMAP-project started in April 1996. First results concerning the detection of changes in agricultural land use are presented. Our first results show that interferometric processing of ERS-1/ERS-2 SAR data and analysis of the resulting coherence images is an important step for the monitoring of the agricultural land use by remote sensing methods: No other remote sensing data can display the farming activities in a similar way. Also coherence information can improve classification accuracy especially in all cases where no data from optical sensors are available.
Keywords: crop classification, land use, coherence, GIS, detection of land-management activites


The project EMAP (ERS-1/-2 SAR DATA FOR MONITORING AGRICULTURAL LAND USE AS A LONG- TERM PROJECT) is a ERS-1 project, which had been accepted by ESA and is financed by Deutsche Agentur für Raumfahrtangelegenheiten (DARA) and Bundesministerium für Ernährung, Landwirtschaft und Forsten (BML).

Principal Investigator of the EMAP-project is Prof. Kühbauch of the Institut für Pflanzenbau, University of Bonn. Project partners are the Institute of Navigation (INS), University of Stuttgart, the research centre for environment and health (GSF-PUC) Oberschleißheim and the Jena-Optronic GmbH (DJO) at Jena.

The purpose of this project is to analyze ERS-1/-2 SAR data as a tool for agricultural crop monitoring. The ERS-1 /ERS-2 satellites allow a continuous multitemporal monitoring of cultivated crop species. Within this project ERS-1/ERS-2 SAR.SLC data acquired in 35 days intervals at four important agricultural sites in Germany are used to identify crops and analyze crop cultivation practices, crop rotation and biomass development. These four sites represent areas with intensive agricultural land use but with different characteristics. For improving crop identification with ERS-1 /-2 radar image data it is necessary to decouple short-term effects on the radar backscattering resulting from non-crop parameters like rainfall, soil moisture or wind from those of the crops itself. For this reason the project is separated into short term and long term multitemporal investigations, which cover the development of a Geographical Information System (GIS) for the four test sites. The analysis is supported by combining SAR-data with optical satellite images and non remote sensing data contained in the GIS. The complementary use of ERS-1 SAR data, optical data and GIS information improves the reliability of the classification with exclusive optical data or only radar data. This is due to the fact that radar data are statistically independent from optical data. The interaction between incoming energy and scattered / backscattered electromagnetic waves by vegetation canopies and soil surface is quite different in the microwave and optical frequencies: Microwave backscattering is dominated by physical ground parameters like geometrical structure, water content, dielectric constants, polarisation etc. Scattering of optical waves is dominated by biochemical effects like photosynthesis. For this reason a combined classification should provide more accurate results with a better separation of different types of land use. The objective of our method for classification is to use only one optical scene per year combined with a multitemporal set of SAR scenes, which can be acquired independent from weather conditions and sun illumination. The high repetition rate of SAR image acquisition improves multitemporal data analysis and investigations.

Interferometric processing of the SAR-data is an important point in the EMAP conception: the main objective is to characterize coherence properties as a function of seasonal changes in land use of the cultivated areas. Tandem-pair scenes acquired at different dates are studied in terms of their coherence properties, which are related to their biophysical variations. Coherence images again provide information, which is independent from the information contained in the intensity images. They are therefore complementary to both radar intensity and optical intensity images. It could be shown that the interferometric results also allow the monitoring of agricultural land management activities. The various types of land management and the timing of the work on the soil is of particular interest for the expert system, which has to be developped by integrating radar and optical remote sensing data and ancillary data in a GIS.


The investigations are performed at four different regions in Germany, which are characterized by intensive agricultural land use with significant different site characteristics:

  1. Test site „Weilerswist" Lat: 50°39' - 50°48' North Lon: 6°45' - 6°55' East. The test site is located near Bonn/Germany. Very flat area with field sizes between 1 - 45 ha.(Test site of Institut für Pflanzenbau)
  2. Test site „Ostalb" Lat: 48°30' - 49°02' North Lon: 10°09' - 10°30' East. The test site is located in Baden-Württemberg/Germany. It consists of different landscapes (hilly and flat).The field sizes are very small between 0.5 - 2 ha. The test site is characteristic for Baden-Württemberg. (Test site of INS)
  3. Test site „Scheyern" Lat: 48°24' - 48°36' North Lon: 11°20' - 11°40' East. The test site is located in Bavaria / Germany between Munich and the Danube river. The size of the farm „Scheyern" is 150 ha. The farm is located within the center of a very hilly area. (Test site of GSF-PUC).
  4. Test site „Buttelstedt" Lat: 51°01' - 51°07' North Lon: 11°16' - 11°' East. The test site is located in Thuringia / Germany. It is characterized by large field sizes.(Test site of DJO)

Each of the four test sites is surveyed by one of the EMAP-project partners as indicated above.

Ground Truth and Geographical Information System (GIS)

A crop survey has been performed on the basis of a 1:5000 scaled map, where the exact boundaries of each field were digitized and with the present crop type had been entered into polygon attribut table of GIS. Extra information as crop damage, soil erosion and other pecularities have also been identified parcelwise. In the case of test site „Ostalb" for example 2030 parcels with a total area of 3200 ha have been surveyed.

For 24 selected fields (6 fields for each of the four main crop types: winter wheat, winter barley, winter rape and corn) besides the above mentioned parameters other relevant parameters are registered and digitized to obtain a complete knowledge about these polygons for further investigations. These parameters can be devided in three groups:

  1. Data registered at each acquisition date: growth stage of plants (phenology, ear angle), plant height, fresh and dry biomass of plants and crop products, soil moisture, leaf area index, moisture on leaf surface.
  2. Data collected through farmer interview: sowing date ,sowing density, row direction, interrow distance, number of plants per squaremeter, pecularites of cultivations (e.g. deseases, soil compaction) practice and dates for tillage, spraying pestizides fertilizing, rotation scheme, yield.
  3. Ancillary information: official digital cadastral map (ALK = Automatisierte Liegenschaftskarte), official soil map (soil classification), meteorological data, digital terrain model.

Additional ground truth data acquisitions in between two consecutive ERS-1 acquisitions have been foreseen during the period of May and July, where we have very rapid changes in the vegetation state.

Most of these data have been entered in the polygon attribut table of GIS. The GIS had been adapted to the regional needs for a sucessful crop monitoring for each of the four testsites separately. Further input parameters are boarders of countries and municipality, official digital topographic information (ATKIS = Amtliches topographisch kartographisches Informationssystem), phenological data of the main crops, meteorological an agro-ecological data.

These data are used for spatial intersection within GIS and satellite imagery for further analysis.

ERS-1 / ERS-2 data selection and pre-processing

ERS-1 / ERS-2 tandem data and ERS-2 data have been ordered in 35-days intervals as SAR.SLC scenes for all 4 test sites between April and November 1996. Similar data requests are planned for 1997 and 1998. (ERS-1 data have only be made available until 3rd of June 1996). Additional SAR-Images already acquired between 1991 and 1994 are available at INS for the test sites „Weilerswist" and „Ostalb". These provide information about crop species cultivated on the same fields in the past years. This is important because the current agricultural management practices use a specific sequence (crop rotation), which could be analyzed using these data.

Several filter methods have been tested to reduce the speckle effect in the SAR-intensity images. A modified GMAP filter (Lopes et. al, 1993) has been selected for final filtering with a filter size of 7 x 7 pixels. The geocoding of the SAR.SLC data as well as for the interferometric image products using digital terrain models of the test site areas is performed by DJO for all EMAP-project partners.

ERS-1 radar intensity signatures and their influences on land use classification

To analyze the radar signatures of agricultural fields in ERS-1 SAR images the mean grey values of more than 560 test fields were analyzed in the case of „Ostalb" test site. Ground truth information about the field area and vegetation type of these fields were taken from the GIS. A mean value of all pixels was calculated for each of these fields, where pixels containing the field edge were eliminated.

The analysis showed large variations of mean grey values of fields belonging to the same vegetation type. Variations in mean grey values of fields belonging to the same vegetation type are due to differences in the biophysical parameters, responsible for the radar response of an agricultural field. These could be differences in soil moisture content and soil roughness, where the vegetation canopy is not dense enough to block contributions of the soil surface to the backscattering characterics of the field. Other biophysical parameters of the vegetation canopy are volumetric density of plant, water content, size and density of the individual scatterers and the height of the vegetation canopy. These parameters change the radar response throughout the development stages of the vegetation. There are of course field to field variations of these parameters because of different growing conditions.

To get a better understanding we used radar backscatter models to simulate the different influences. One important result of these studies was: Due to the very steep incidence angle of the ERS-1 the vegetation canopy of agricultural fields can only block the contribution of the underlying soil, when the vegetation cover is dense enough and not dry.

Obviously this is only the case during a short period before the plants get ripe and dry. Only then the vegetation dominates the intensity of the radar return. However besides these, there are other influences on the radar backscatter intensity:

  1. the influence of the field orientation towards the SAR sensor, i.e. the actual fluctuation of sensor look angle due to the topography. Field slopes reach values up to +/- 6 degrees from the horizontal plain. This could lead to backscatter intensity variations of up to 5 db. Corresponding fluctuations in mean grey values had to be analyzed and corrected.
  2. the influence of the orientation of plant rows relative to the SAR-sensor. In the early development stage of the plants significant variations in grey values could be observed (Müller et. al., 1993) depending on the row direction of the plants, which resulted in higher grey values if the row direction is parallel to the ERS-1 flight path and in lower grey-values for row directions perpendicular to the flight path.

Conventional Maximum Likelihood Classification (ML) has been tested using multitemporal radar data together with one optical scene for validation areas (see Fig. 1) in the southern part of the „Ostalb" test site (Hartl et. al., 1995). The non agricultural areas (red = urban, dark green = forest, light green = grassland) have been masked and the corresponding pixels were not used for the classification.

A separation between the vegetation types is not possible, if only multitemporal ERS-1.SAR images are used for classification.This is the case although the filtered radar intensity images (combined to a multitemporal colour composite) show some structural information about the land use (Fig. 2). We could only separate with an accuracy of about 80 % between two groups: one group consists of winter wheat, summer barley and oat and the other one of winter barley, rape and corn.

However, if one optical scene (Fig. 3) is used together with the 3 multitemporal SAR images shown in Fig 2 in a combined classification, the separation of the six main crop types was possible with limited accuracy. However the accuracy is better than with the optical data alone. A 10 % average increase of separation accuracy for the most important crop types was achieved in comparison to optical classification. Tab. 1 shows the confusion matrix of the results achieved with the combined classification.

Fig 1: Test site „Ostalb": Detailed ground truth derived from inspection and implemented in the GIS is available in several test areas. Sealed areas (red), forest (dark green) and grassland (light green) has been separated by masking.

Fig. 2: Multitemporal colour composite of the 3 SAR-images used for classification

Fig. 3: Optical image of „Ostalb" test site

Fig. 4: Result of Maximum Likelihood classification

Tab. 1: Result of Maximum Likelihood classsification for validation areas

vegetation type pixels WW












winter wheat 51477 56.08 11.30 11.37 18.62 0.77 1.79
winter barley 24015 15.09 57.21 11.29 12.71 1.81 1.82
summer barley 37727 3.87 6.52 67.77 17.15 2.66 1.98
oat 9327 20.75 10.39 26.60 34.75 1.01 6.43
rape 19802 0.30 1.06 6.16 7.17 84.55 0.72
corn 25484 0.45 0.62 2.39 12.78 0.88 82.48
sum 167832 36062 23356 38442 27028 18895 23870

average accuracy: 63.81%
overall accuracy: 65.05%
Kappa coefficient: 57.32%

The classification results are shown in Fig. 4. The results suffered from the extremely small field sizes in some parts of the test site. Other classification methods using neural network classifiers (Benedictsson et. al., 1990),(Foody et. al., 1995) have been tested for the same data. However no significant improvements were detected. Better results could be achieved with all classifiers, if only field sizes of more than 1 ha and 50 m width are taken into account.

It could be shown that post-classification with majority filter based on known geometry of agricultural fields in the validation areas increases the overall classification accuracy about 20 %. Nevertheless the achieved accuracy is disappointing, but having in mind the above mentioned results of the radar backscatter analysis one can understand that the classification accuracy, which can be achieved with multitemporal ERS-1 / ERS-2 intensity must be limited. For that reason the phase information contained in the ERS-1/ ERS-2 images, which could be extracted by interferometric processing of the SAR-data, should be used to improve these results.


Interferometric processing of the SAR-SLC data has been foreseen to analyze the coherence properties as a function of seasonal changes in the land use of cultivated areas in all four test sites. Especially TANDEM-data with only 1 day interval between the acquisitions have been selected for this purpose, because decorrelation due to vegetation changes have often been the reason for poor results in the past. The data of all 4 test sites are processed at INS making use of the INS-ANTIS (Schmidt, 1995) interferometry system, which is based on PCI-EASI-PACE image processing system software. The interferometric processing is an important part of the EMAP-project because of the additional statistically independent information about land use and because farmers activities can be extracted from coherence and relative phase images. To demonstrate these capabilites, a series of 10 consecutive ERS-1.SAR.SLC images of the Bonn test-site, taken during the second ice-phase has been analyzed by INS. These images were sampled in 3 days intervals. We have therefore good coherence between two consecutive images. Coherence images have been produced for the following pairs with acquisition dates:

TAB 2: ERS-1 image pairs used for generation of coherence images

01.03.1994 04.03.1994 COHERENCE IMAGE 1
04.03.1994 07.03.1994 COHERENCE IMAGE 2
07.03.1994 10.03.1994 COHERENCE IMAGE 3
10.03.1994 13.03.1994 COHERENCE IMAGE 4
13.03.1994 16.03.1994 COHERENCE IMAGE 5
16.03.1994 19.03.1994 COHERENCE IMAGE 6
19.03.1994 22.03.1994 COHERENCE IMAGE 7
22.03.1994 25.03.1994 COHERENCE IMAGE 8
25.03.1994 28.03.1994 COHERENCE IMAGE 9

The nine coherence images generated with each of these pairs are quite similar to one another. They show the following general effects:

Forest areas and highways show up as dark areas, i.e. as areas with low coherence. For forest areas thisis quite clear due to decorrelation effects caused by the vegetation. For the highways, no coherence could be detected, because the amplitude of the reflected signal is too low. Agricultural fields appear in general quite bright, but in all images there are always some fields which appear dark, i. e. with low coherence. These are the fields, where we expected that coherence was lost due to farmers activities during the time between the two datatakes of each pair. An interrogation of farmes about their farming activities on the fields which showed low coherence gave us the confirmation: In all cases the reason for the loss of coherency could be identified as some kind of farming activities as ploughing the fields in the time gap between the 3 days interval of the two ERS-1 acquisition of the corresponding image pair.

Fig. 5: Coulor composite of 3 coherence images of the „Weilerswist" test site
red = coherence image 3 (07.03.1994 + 10.03.1994)
green = coherence image 4 (10.03.1994 + 13.03.1994)
blue = coherence image 5 (13.03.1994 + 16.03.1994)

To show up the farmers activities in a very condensed way over a longer period we produced a multitemporal colour composite image of 3 of our 9 coherence images, where the coherence of the pair taken on 07.03.1994 + 10.03.1994 is shown in red, the pair 10.03.1994 + 13.03.1994 in green and the pair 13.03.1994 + 16.03.1994 in blue colour. The result is given in Fig.5. All fields with similar coherence grey-values in all three channels, i. e. with no change in coherency over the whole period from 07.03.1994 to 16.03.1993 appear as non coloured black and white pixels. The different coloured fields of Fig. 5 result from different grey values in the 3 channels. The red have low coherence in the green and blue channel, i. e. the farming activities must have taken place between 13.03.1993 and 16.03.1993 and between 16.03.1993 and 19.03.1993. Yellow fields are the fields with no coherence in the blue channel, i. e. farming activities in this case have taken place after 13.03.1994.

Not only farmers activities, but also meteorological effects or changes in the vegetation can be the reason for decorrelation. However these effects must have been neglectable in our case, because the vegetation canopies in March are still very low and we had a short time interval between two ERS-1 datatakes.

Fig. 6 to 8 show three coherence images of the „Ostalb" test site, where a series of 3 ERS-1/ERS-2 tandem-pairs have been used. Fig. 6 is quite similar to the series of nine coherence images of the „Weilerswist" test site. Agricultural fields appear quite bright, since only few vegatation is present at that timeof the year. Also only very few fields can be detected as fields with farming activities. Fig.7 and even more Fig.8 show a more datailed structure of the agricultural fields with all levels of coherence values between 0.1 and 0.7-0.8. Fig. 8 shows a field structure pattern similar to an optical remote sensing image. In this case the low coherence values for many of the fields are due to the vegetation canopy and not due to the farmers activities.

Fig. 6: Coherence image ERS-1 23.03.96 / ERS-2 24.03.96
Test site „Ostalb"

Fig. 7: Coherence image ERS-1 27.04.96 / ERS-2 28.04.96
Test site „Ostalb"

Fig. 8: Coherence image ERS-1 01.06.96 / ERS-2 02.06.96
Test site „Ostalb"


The additional information extracted from coherence images can further improve the accuracy of land use classification using multitemporal ERS-1/ERS-2 data together with one optical scene. In general coherence is one additional channel in feature space, statistically independent from optical and radar intensity channels. The degree of coherence, if not degraded by systematic effects (large baselines, different zero doppler frequencies in processing the data) is sensitive to the vegetation. High coherence values indicate that no decorrelation due to the vegetation canopy had taken place. This is the case for bare soil conditions and for thin and/or dry vegetation canopies, where the vegetation has only little influence on the radar backscatter. On the other hand low coherence values are not always the result of a dense or wet vegetation canopy, which is blocking the radar returns from the ground. Also farmers activities on bare soil, if they took place between the acquisitions of the corresponding image pair can result in a full decorrelation. One has to be careful when applying coherence images as additional channels for conventional classification methods. Fields, where the low coherence is due to farming activities have to be excluded from the automatic classification using coherence information. However these fields can easily be identified. Also the detection of farmers activities can be helpful for classification, because the knowledge of field activities at a certain time of the year allow the agricultural experts to extract information about the type of vegetation. However to make use of this information a more sophisticated system is necessary rather than conventional ML- or NN- classification methods.

Nevertheless our first results show that interferometric processing of ERS-1/ERS-2 SAR data and analysis of the resulting coherence images are an important step for monitoring of agricultural land use by remote sensing methods: No other remote sensing data can display the farming activities in a similar way. Also coherence information can improve classification accuracy especially in all cases where no data from optical sensors are available. It seems that coherence information extracted from a tandem.SAR acquisition taken in the middle of the vegetation period can substitute the information containecd in an optical remote sensing image.


BENEDIKTSSON J.A., SWAIN P.H., and ERSOY O.K.: Neural network approaches versus statistical methods in classification of multisource remote sensing data. IEEE Transactions on Geoscience and Remote Sensing, 28, 540-551, 1990.

FOODY G.M., MCCULLOCH M.B., YATES W.B.: Classification of remotely sensed data by an artificial neural network: issues related to training data characteristics. Photogrammetric Engineering and Remote Sensing, 61(4), 391-401, 1995.

HARTL PH., GÜTH S., KLAEDTKE H.-G., PALUBINSKAS G., REICH M., WÖRZ K.: Land use classification of agricultural fields with multitemporal ERS-1 SAR data and optical remote sensing data. Proceedings of the second ERS Applications Workshop, London U.K. 6 - 8 December 1995, ESA SP-383.

LOPES A., NEZRY E., TOUZI R., LAUR H.: Structure detection and statistical adaptive speckle filtering in SAR image. International Journal of Remote Sensing, 14(9), 1735 - 1758, 1993.

MÜLLER U., LÖCHERBACH TH., KÜHBAUCH W., FÖRSTNER W.: Suitability of ERS-1 SAR PRI-Data for multitemporal agricultural land -use mapping. Proceedings of the second ERS-1 Symposium, Hamburg Germany, 1993, ESA SP-361.

SCHMIDT J.: Antarktisches Interferometrie System ANTIS Benutzerhandbuch Version 1.0, Internal Report of the Institute of Navigation, University of Stuttgart, Stuttgart,111 pages, 1995.

Keywords: ESA European Space Agency - Agence spatiale europeenne, observation de la terre, earth observation, satellite remote sensing, teledetection, geophysique, altimetrie, radar, chimique atmospherique, geophysics, altimetry, radar, atmospheric chemistry