Preliminary data analysis: Gauteng and Limpopo

Get Complete Project Material File(s) Now! »


The main aim of this chapter is to introduce two datasets. All the hypertemporal techniques presented in Chapter 4 are applied to these two datasets. The two datasets are discussed in Section 2.8. The first few sections of this chapter however introduce the basic principles of remote sensing, as well as the MODIS sensor. These introductory sections are needed to understand the content of the datasets in Section 2.8. Remote sensing is the science of converting data about the earth’s surface, recorded with remote (distant) sensor platforms, into usable information. The remote sensors archive how the earth’s surface reflects or transmits electromagnetic energy at different wavelengths and thus record an electromagnetic spectral signature of the earth’s surface [33].


It can be argued that the moment in time which gave birth to photography was in fact also the starting point of spaceborne remote sensing. Photography was invented in 1839. Those early photographs were created by the photographic processes of Nicephore Niepce, William Henry Fox Talbot, and Louis Jacques Daguerre [33]. The first aerial photograph was taken (of Bievre, France) by a Parisian photographer named Gaspard Felix Tournachon (from a balloon). The earliest existing aerial photograph was taken from a balloon over Boston in 1860 by James Wallace Black and immortalised by Oliver Wendell Holmes [33]. The First and Second World War sparked the widespread use of aerial photography as a surveillance tool. The use of aerial photography for environmental purposes only became popular after the Second World War [10]. The term “remote sensing” was first coined by Evelyn Pruit after recognising that “aerial photography” no longer accurately described the different images recorded, as some images (at that point) were recorded by using wavelengths outside the visible spectrum [34]. The next step in the evolution (starting in the 1960s) of remote sensing was when humans started using spaceborne platforms to house remote sensing sensors. The space era, which is still continuing, can be discussed under four headings, namely military reconnaissance satellites, manned space flight, meteorological satellites and earth resource satellites [35].

Military reconnaissance satellites

Before 1960, the United States of America (USA) and the former Union of Soviet Socialist Republics (USSR) used aerial photographs to keep track of each other’s military capability. However, at the Surprise Attack Conference in Geneva (in 1958) it was proposed for the first time to use satellites to gather military information [35]. CORONA was one of the first military programmes under which satellites were launched into space to perform military reconnaissance (active during the 1960s). These missions were usually very short in duration, typically no longer than one or two weeks. These early systems were constrained, since they could only carry a limited amount of film. The film canister was ejected and picked up as it descended to earth [35]. Later systems could store images in digital format and transported the data to the earth via telemetry.

Manned space flight

On April 12, 1961 Yuri Gagarin became the first person to orbit the earth. Although no photos were taken during this mission, it became apparent that spaceborne earth observation had great potential. The USA also started manned space missions in the 1960s, culminating in the moon landing in 1969 during the Apollo programme [35]. The Mercury (1961-1963), Gemini (1965-1966) and Skylab (1973-1974) programmes were some of the manned American programmes that captured pictures of the earth. The Russians also conducted their own manned missions, which included the Vostok and Voskhod programmes, which were analogous to the Mercury and Gemini missions [35].

First generation

The idea of a civilian satellite that could be used for scientific earth surveillance was proposed in 1965 by William Pecora, director of the United States Geological Survey (USGS), and was inspired by the photographs taken on the Mercury, Gemini and Apollo missions in the 1960s. Unfortunately this idea was met with heavy criticism from the Bureau of Budget (BOB) and the Department of Defense (DOD), since the BOB thought high-altitude aircraft would be better suited to the task and the DOD was concerned that earth resource satellites would jeopardise its military reconnaissance missions. In 1966 NASA felt pressure from the Department of the Interior (DOI), after USGS convinced the DOI to announce that the DOI would be starting its own earth resource satellite programme.
This forced NASA to accelerate the building of an earth resource satellite. Unfortunately, a limited budget and sensor disagreements between application agencies again delayed the satellite construction process. Finally, by 1970 NASA received authorisation to build a satellite. The first earth resource satellite, Landsat-1, was launched on July 23, 1972 by NASA; at that time the satellite was known as the Earth Resources Technology Satellite (ERTS) [33]. Landsat-1 carried the Return Beam Vidicon (RBV) and Multispectral Scanner (MSS) systems. Seven satellites were launched in the Landsat series, some of which are still functioning today. Landsat-4, launched on July 16, 1982, carried the Thematic Mapper (TM), another predecessor of MODIS [33].

Second generation

In the early 1980s, there was a merger between the human spaceflight missions and the earth science missions of NASA, which was termed System Z. System Z fostered the idea of having one giant spacecraft carrying a variety of sophisticated earth observation sensors, including radar. System Z changed into the EOS in 1983, after scientists realised that multiple small missions would lead to better results [37, 38]. In the beginning the System Z earth observation system was to consist of two large (15-ton) platforms called EOS-A and EOS-B. After a reduction in size, the original sunsynchronous system EOS-A was renamed to EOS-Terra to emphasise its main function, namely to make land observations. EOS-Terra was launched on December 18, 1999. EOS-B was renamed to EOS-Aqua, as EOS-B would focus on ocean observation, and was launched on May 4, 2002. EOS-Terra carries the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), the Multi-angle Imaging SpectroRadiometer (MISR), MODIS, the Measurements of Pollution in the Troposphere (MOPITT) and the the Clouds and the Earth’s Radiant Energy System (CERES). EOSAqua carries the Advanced Microwave Scanning Radiometer-EOS (AMSR-E), the Atmospheric Infrared Sounder (AIRS), the Advanced Microwave Sounding Unit (AMSU), CERES, the Humidity Sounder for Brazil (HSB) and MODIS. Many EOS missions have been launched over the last decade [37, 38].

Third generation

During the mid 1990s a series of decisions were taken that NASA’s EOS programme would only be regarded as a proof of concept and that NOAA would eventually be responsible for developing systems to study climate change. At about the same time the USA government decided to combine the low earth orbiting satellite programmes of NOAA and the DOD into the National Polar Orbiting Environmental Satellite Series (NPOESS). The responsibility of administrating NPOESS was assigned to the newly created Integrated Program Office (IPO) consisting of NASA/NOAA/DOD [37,38]. The NPP satellite was originally developed by the IPO, until DOD participation in the project was dissolved. The NPOESS Preparatory Project (NPP) satellite is intended to bridge the gap between old (Terra) and new systems (still to be launched) and was launched on October 28, 2011.


  • CHAPTER 1 Introduction
    • 1.1 Settlement detection
    • 1.2 Hypertemporal approaches
    • 1.3 Data selection
    • 1.4 Sequential approaches
    • 1.5 Inductive simulation
    • 1.6 Problem statement
      • 1.6.1 Existing hypertemporal techniques
      • 1.6.2 Contribution to hypertemporal classification
      • 1.6.3 Contribution to hypertemporal change detection
    • 1.7 Publications and related work
    • 1.8 Layout of thesis
  • CHAPTER 2 Remote sensing
    • 2.1 History of remote sensing
      • 2.1.1 Military reconnaissance satellites
      • 2.1.2 Manned space flight
      • 2.1.3 Meteorological satellites
      • 2.1.4 Earth resource satellites
    • 2.2 A typical remote sensing system
    • 2.3 Electromagnetic radiation
      • 2.3.1 Electromagnetic spectrum
      • 2.3.2 Propagation of electromagnetic radiation
      • 2.3.3 Radiation units
      • 2.3.4 Blackbody radiation
    • 2.4 Atmospheric interactions
      • 2.4.1 Atmospheric absorption
      • 2.4.2 Atmospheric scattering
    • 2.5 Surface interaction
      • 2.5.1 Reflection, absorption and transmission
      • 2.5.2 Albedo
      • 2.5.3 Bidirectional Reflectance Distribution Function
      • 2.5.4 Spectral signature of vegetation, soil and water
    • 2.6 Remote Sensing Platforms
      • 2.6.1 Ground-based, airborne and spaceborne platforms
      • 2.6.2 Passive and active remote sensing systems
      • 2.6.3 Resolution of remote sensing sensors
      • 2.6.4 Signal-to-noise ratio
    • 2.7 Moderate Resolution Imaging Spectroradiometer
      • 2.7.1 History of MODIS
      • 2.7.2 MODIS sensor characteristics
      • 2.7.3 MODIS products
      • 2.7.4 MODIS design
      • 2.7.5 The MCD43A4 product
    • 2.8 Dataset description
    • 2.9 Conclusion
  • CHAPTER 3 Sequential analysis
    • 3.1 Neyman-Pearson
    • 3.2 Kullback-Leibler divergence
    • 3.3 Hypothesis testing: Wald’s formulation
      • 3.3.1 The OC and ASN functions of the SPRT
      • 3.3.2 Wald’s approximations
      • 3.3.3 Exact computation
      • 3.3.4 Simulation
      • 3.3.5 Example: Gaussian random variable
      • 3.3.6 Example: Bernoulli random variable
    • 3.4 Hypothesis testing: Bayesian formulation
      • 3.4.1 On the structure of the minimal cost function
      • 3.4.2 Bayesian versus Wald’s formulation
      • 3.4.3 Example
    • 3.5 Bayesian quickest detection
    • 3.6 Non-Bayesian quickest detection
      • 3.6.1 Lorden’s performance measure
      • 3.6.2 Pollak’s performance measure
    • 3.7 Conclusion
  • CHAPTER 4 Hypertemporal techniques
    • 4.1 Simulation
      • 4.1.1 Noise-free inductive models
      • 4.1.2 Proposed simulator
    • 4.2 Classification
      • 4.2.1 Literature review
      • 4.2.2 Minimum distance classifier
      • 4.2.3 Time-varying maximum likelihood classifier
      • 4.2.4 Support Vector Machine
    • 4.3 Change detection
      • 4.3.1 Literature review
      • 4.3.2 Lunetta et al.’s scheme
      • 4.3.3 Cumulative Sum
      • 4.4 Conclusion
  • CHAPTER 5 Results
    • 5.1 Preliminary data analysis: Gauteng and Limpopo
      • 5.1.1 Yearly ensemble mean
      • 5.1.2 Temporal Hellinger distance
      • 5.1.3 CSHO model parameters
      • 5.1.4 Noise correlation
      • 5.1.5 Spatial correlation
    • 5.2 Simulator results: Gauteng and Limpopo
      • 5.2.1 Simulator validation
      • 5.2.2 Preliminary validation results
      • 5.2.3 Discussion of metric selection
      • 5.2.4 Class metrics
      • 5.2.5 Pixel metrics
      • 5.2.6 Discussion of simulator results
    • 5.3 Classification results: Gauteng and Limpopo
      • 5.3.1 Classification accuracy metrics
      • 5.3.2 Structure used for accuracy metrics
      • 5.3.3 Preliminary benchmark classification results: Gauteng
      • 5.3.4 Preliminary SVM classification results: Gauteng
      • 5.3.5 Classification results: Gauteng
      • 5.3.6 Classification results: Limpopo
      • 5.3.7 Important classification conclusions
    • 5.4 Change detection results: Gauteng and Limpopo
      • 5.4.1 Change detection accuracy metrics
      • 5.4.2 Results of Lunetta et al.’s scheme: Gauteng and Limpopo
      • 5.4.3 Temporal dependence and the CUSUM threshold
      • 5.4.4 Results of the CUSUM test: Gauteng and Limpopo
      • 5.4.5 Important change detection conclusions
    • 5.5 Conclusion
  • CHAPTER 6 Conclusion
    • 6.1 Main conclusions
    • 6.2 Summary of work
    • 6.3 Future work
    • APPENDIX A Mathematical Background
    • A.1 Stochastic calculus
    • A.2 Gaussian quadrature
    • A.3 Cholesky factorisation
    • A.4 Lagrange multipliers
    • A.5 Kernel Density Estimation


Related Posts