Universität zu Köln

Crop Growth Monitoring by Hyperspectral and Microwave Remote Sensing

Koppe, Wolfgang (2013) Crop Growth Monitoring by Hyperspectral and Microwave Remote Sensing. PhD thesis, Universität zu Köln.

[img]
Preview
PDF - Accepted Version
Download (19Mb) | Preview

    Abstract

    Methoden und Techniken der Fernerkundung fungieren als wichtige Hilfsmittel im regionalen Umweltmanagement. Um diese zu optimieren, untersucht die folgende Arbeit sowohl die Verwendung als auch Synergien verschiedener Sensoren aus unterschiedlichen Wellenlängenbereichen. Der Fokus liegt auf der Modellentwicklung zur Ableitung von Pflanzenparametern aus fernerkundlichen Bestandsmessungen sowie auf deren Bewertung. Zu den verwendeten komplementären Fernerkundungssystemen zählen die Sensoren EO-1 Hyperion und ALI, Envisat ASAR sowie TerraSAR-X. Für die optischen Hyper- und Multispektralsysteme werden die Reflexion verschiedener Spektralbereiche sowie die Performanz der daraus abgeleiteten Vegetationsindizes untersucht und bewertet. Im Hinblick auf die verwendeten Radarsysteme konzentriert sich die Untersuchung auf Parameter wie Wellenlänge, Einfallswinkel, Radarrückstreuung und Polarisation. Die Eigenschaften verschiedener Parameterkombinationen werden hierbei dargestellt und der komplementäre Beitrag der Radarfernerkundung zur Wachstumsüberwachung bewertet. Hierzu wurden zwei Testgebiete, eines für Winterweizen in der Nordchinesischen Tiefebene und eines für Reis im Nordosten Chinas ausgewählt. In beiden Gebieten wurden während der Wachstumsperioden umfangreiche Feldmessungen von Bestandsparametern während der Satellitenüberflüge oder zeitnah dazu durchgeführt. Mit Hilfe von linearen Regressionsmodellen zwischen Satellitendaten und Biomasse wird die Sensitivität hyperspektraler Reflexion und Radarrückstreuung im Hinblick auf das Wachstum des Winterweizens untersucht. Für die optischen Daten werden drei verschiedene Modelvarianten untersucht: traditionelle Vegetationsindices berechnet aus Multispektraldaten, traditionelle Vegetationsindices berechnet aus Hyperspektraldaten sowie die Berechnung von Normalised Ratio Indices (NRI) basierend auf allen möglichen 2-Band Kombinationen im Spektralbereich zwischen 400 und 2500 nm. Weiterhin wird die gemessene Biomasse mit der gleichpolarisierten (VV) C-Band Rückstreuung des Envisat ASAR Sensors linear in Beziehung gesetzt. Um den komplementären Informationsgehalt von Hyperspektral und Radardaten zu nutzen, werden optische und Radardaten für die Parameterableitung kombiniert eingesetzt. Das Hauptziel für das Reisanbaugebiet im Nordosten Chinas ist das Verständnis über die kohärente Dualpolarimetrische X-Band Rückstreuung zu verschiedenen phänologischen Wachstumsstadien. Hierfür werden die gleichpolarisierte TerraSAR-X Rückstreuung (HH und VV) sowie abgeleitete polarimetrische Parameter untersucht und mit verschiedenen Ebenen im Bestand in Beziehung gesetzt. Weiterhin wird der Einfluss der Variation von Einfallswinkel und Auflösung auf die Bestandsparameterableitung quantifiziert. Neben der Signatur von HH und VV ermöglichen vor allem die polarimetrischen Parameter Phasendifferenz, Ratio, Koherenz und Entropy-Alpha die Bestimmung bestimmter Wachstumsstadien. Die Ergebnisse der Arbeit zeigen, dass die komplementären Fernerkundungssysteme Optik und Radar die Ableitung von Pflanzenparametern und die Bestimmung von Heterogenitäten in den Beständen ermöglichen. Die Synergien diesbezüglich müssen auch in Zukunft weiter untersucht werden, da neue und immer variablere Fernerkundungssysteme zur Verfügung stehen werden und das Umweltmanagement weiter verbessern können.

    Item Type: Thesis (PhD thesis)
    Translated abstract:
    AbstractLanguage
    Timely monitoring of crop growth status at different scales is crucial for improving regional crop management decisions. The main objective of the recent study is a model development to predict and estimate crop parameters, here biomass, plant N concentration and plant height, based on different earth observation systems that provide complementary information. Among these are the hyperspectral sensor Hyperion and the multi-spectral sensor ALI based on EO-1 satellite, ASAR on Envisat and TerraSAR-X. Based on the characteristics of the different systems, methods are analysed regarding crop parameter estimation and crop growth status monitoring on a regional level. Factors which are taken into account for optical data are bandwidth and centre of the wavebands, spectral reflectance response from visible and near infrared and the performance of vegetation indices based on the two sensors for semi-empirical parameter estimation. Concerning the SAR sensors Envisat ASAR and TerraSAR-X, parameters such as wavelength, incidence angle and polarization were analysed in terms of crop parameter retrieval and crop status determination. For this, two test areas, one for winter wheat in Northern China Plain (Huimin County) and one for rice in the Northeast of China (Jiansanjiang) were selected. At both sites, intensive ground data collection during the vegetation periods in 2006, 2007 for Huimin and 2009 and 2011 for Jiansanjiang were performed. Concerning the winter wheat crop monitoring based on hyperspectral, multi-spectral and C-band SAR data, the study was conducted in Huimin County, Shandong Province of China in the growing season of 2005/2006 involving three large winter wheat fields each year managed by different farmers. Winter wheat growth parameters including aboveground biomass, plant N concentration, LAI, and plant height were collected at different growth stages. Three different prediction models were investigated: traditional vegetation indices calculated from broad and narrow bands, and Normalized Ratio Indices (NRI) calculated from all possible two-band combinations of Hyperion between 400 and 2500 nm. The results indicated that TVI performed best among the tested vegetation indices using either broad bands (R2 = 0.69, 0.32 and 0.64 for biomass, N concentration and plant height, respectively) or narrow bands (R2 = 0.71, 0.33 and 0.65 for biomass, N concentration and plant height, respectively). The best performing Normalized Ratio Index (NRI) selected through band combination analysis were significantly better than TVI, achieving R2 of 0.83, 0.81 and 0.79 for biomass, plant N concentration and plant height, respectively. The different NRI models use wavebands from the NIR (centred at 874, 732, and 763 nm) and the SWIR (centred at 1225 and 1305 nm) spectrum with varying bandwidth between 10 and 190 nm. The result of this part proved that vegetation indices derived from NIR- and SWIR-Hyperion spectrum are better predictors of plant aboveground biomass, nitrogen concentration and plant height than indices derived from only visible spectrum. In addition, Envisat ASAR VV polarization data were related to winter wheat crop parameters. Bivariate correlation results from this study indicate that both multi-temporal EO-1 Hyperion as well as Envisat ASAR data provides notable relationships with crop conditions. As expected, linear correlation of hyperspectral data performed slightly better for biomass estimation (R2 = 0.83) than microwave data (R2 = 0.75) for the 2006 field survey. Based on the results, hyperspectral Hyperion data seem to be more sensitive to crop conditions. Improvements for crop parameter estimation were achieved by combining hyperspectral indices and microwave backscatter into a multiple regression analysis as a function of crop parameter. Combined analysis was performed for biomass estimation (R2 = 0.90) with notable improvements in prediction power. For the rice monitoring in the Sanjiang Plain, Northeast China, the main objective was the understanding of the coherent co-polarised X-band backscattering signature of rice at different phenological stages in order to retrieve growth status. For this, multi-temporal dual polarimetric TerraSAR-X High Resolution SpotLight data (HH/VV) as well as single polarised StripMap (VV) data were acquired from the test site. In conjunction with the satellite data acquisition, a ground truth field campaign was carried out. The backscattering coefficients at HH and VV of the observed fields were extracted on the different dates and analysed as a function of rice phenology to provide a physical interpretation for the co-polarised backscatter response in a temporal and spatial manner. Then, a correlation analysis was carried out between TerraSAR-X backscattering signal and rice biomass of stem, leaf and head to evaluate the relationship with different vertical layers within the rice vegetation. HH and VV signatures show two phases of backscatter increase, one at the beginning up to 46 days after transplanting and a second one from 80 days after transplanting onwards. The first increase is related to increasing double bounce reflection from the surface-stem interaction. Then, a decreasing trend of both polarizations can be observed due to signal attenuation by increasing leaf density. A second slight increase is observed during senescence. Correlation analysis showed a significant relationship with different vertical layers at different phenological stages which prove the physical interpretation of X-band backscatter of rice. The seasonal backscatter coefficient showed that X-band is highly sensitive to changes in size, orientation and density of the dominant elements in the upper canopy. Overall, the study demonstrated successfully the estimation of crop status by multi-sensoral remote sensing data. The use of different sensor systems to acquire timely information is especially important for agricultural decision support systems. Thus, as many different systems are available in the future, the combination of different satellite sources is gaining more importance.English
    Creators:
    CreatorsEmail
    Koppe, Wolfgangwolfgang.koppe@web.de
    URN: urn:nbn:de:hbz:38-55629
    Subjects: Geography and travel
    Uncontrolled Keywords:
    KeywordsLanguage
    SAR, Remote Sensing, Hyperspectral, crop monitoring, Radar, PolarimetryEnglish
    Faculty: Mathematisch-Naturwissenschaftliche Fakultät
    Divisions: Mathematisch-Naturwissenschaftliche Fakultät > Geographisches Institut
    Language: English
    Date: November 2013
    Date Type: Publication
    Date of oral exam: 21 January 2014
    Full Text Status: Public
    Date Deposited: 25 Sep 2014 13:50:00
    Referee
    NameAcademic Title
    Bareth, GeorgProf. Dr.
    Schneider, KarlProf. Dr.
    URI: http://kups.ub.uni-koeln.de/id/eprint/5562

    Actions (login required)

    View Item