GoConqr has a new look! Let us know what you think.

Sign up free

Flashcards

Geog 2007 part 2

Pippa Holman English 163 cards

What is the anatomy for a supervised classification?

Cards in this deck (163)

  1. 1What is the anatomy for a supervised classification?What is the anatomy for a supervised classification?
  2. 2What are the stages for supervised classifiers?1. User defined land cover classes
    2. Training site selection
    3. Generation of statistical parameters from the training sites
    4. Classification
    5. Accuracy assessment
    6. Output stage
  3. 3What is meant by training site selection for supervised classifiers?*Good quality training data are needed to 'train' the computer to recognise similar patterns in the imagery
    *Represents small sampling of the entire image/ region to be classified
    *Training data can be selected based on;
    1. Field sites
    2. High spatial resolution data
    3. Previous maps
    4. Investigator knowledge
    *The goal is to select multiple areas for each cover type throughout the image
  4. 4What is the aim for training site selection and what has to be thought of?- To provide quantitative descriptions of the appearance of each thematic class of interest in the image
    - First step is to define the classes (land cover types) of interest and should be based on the aim of the study
    1. NUMBER OF TRAINING PIXELS - Number of pixels to make up a sample depends on variability and distinctiveness of the spectral response of the class. (rule of thumb - >100 per class or at least 10x number of spectral bands in the image)
    2. TRAINING AREA SIZE AND LOCATION - It is best to acquire the sample size number from many small areas around the image rather than from one/ two areas (allows for possible variations in the image to be accounted for)
  5. 5How are evaluating spectral signatures a use of spectral statistical parameters (step 3)?- An important and time consuming step
    - The objective is to have a set of signatures that describe the spectral response of the land cover classes
    - Has class separability
    - Selects homogenous training sites
  6. 6How is estimating statistics a use of generating statistical parameters (step 3)?- Statistical measure of distance between two signatures
    - Separability can thus predict the result of the algorithms based on distance measures
    - The standard deviation should be low
  7. 7What is meant by the classification stage (number 4)?- Rule(s) are needed to decide into which class a given pixel is put into
    - Numerous mathematical approaches to a spectral pattern recognition have been developed
  8. 8What is the parallelepiped classifier?1. Assigns boundaries around the spread of a class in feature space (takes account of spectral variance)
    2. All pixels in the image with values within the designated parallelepiped will be classified as that spectral class
    3. Bounds (ranges of acceptable values in each band) are usually determined from the training sets
    4. The standard deviation in each band is determined and used to calculate the bounds of the cluster
    (Pixels are tested in order of class and as soon as the test is passed, the pixel is classified and the algorithm moves to the next pixel)
  9. 9What is the biggest problem with the parallelepiped classifier (BOX)?*The overlapping bounds*
    -This happens when the bounds from one or more bands from one class overlap with the bounds from another class
  10. 10What is the minimum distance to means (MDM) method?*Calculates the euclidean distance from a candidate pixel to means of each of the classes*
    1. Calculate the mean spectral value for each training set in each band
    2. Put every unknown pixel into the nearest cluster
    3. Compute the distance between the value of the unknown pixel and each of the category mean
    4. Define a limit beyond which a pixel remains unclassified
    NB. This is slower than a parallelepiped classifier and is the same approach as the unsupervised classification clustering method
    NB. If the pixel is close to feature space of two training data, it assigns it to the closest matching cluster mean and is insensitive to variance within spectral value of the training data
  11. 11What is needed for each pixel to be classified in the maximum likelihood classifier (ML)?- For each pixel to be classified...
    1. Assumes dta in a class are (unimodal) Gaussian (normally) distributed
    2. The probability of classification is calculated for each class
    3. The pixel is classified as the class with the largest probability
    4. This is the slowest of the classifiers discussed
    5. Theoretically the best classification
  12. 12What is the maximum likelihood classifier?- Uses probability rather than distance in feature space
    - Ask which class is each pixel 'most likely' to belong to...
    - Much more sophisticated, but is computationally expensive compared to the distance methods
  13. 13What are the advantages and disadvantages of supervised classification (as opposed to unsupervised image classification)?ADVANTAGES - 1. Analyst has control
    2. Processing is tied to specific areas of known identity
    3. Not faced with the problem of matching categories on the final map with field info
    4. Operator can detect errors and often remedy them
    DISADVANTAGE - 1. Signatures are forced (training classes are based on field identification, not spectral properties)
    2. Training data selected by the analyst may not be representative (heterogeneity in classes is common)
    3. Training data is time consuming and costly
    4. Unable to recognise and represent special or unique categories not represented in the training data
  14. 14What is an accuracy assessment?GOALS - Assessing how well the land cover map represents reality/ understanding in how to interpret usefulness of someone else's classification
    STEPS - 1. Collect reference data 'ground truth' (determine class types at certain locations - through high resolution satellite/ airborne imagery/ GPS field visit/ GIS layers)
    2. Compare the reference to classified map (class type on classified map = class type determined from reference data)
  15. 15What are the issues posed and the corresponding solutions for accuracy assessments?1. CHOOSING THE REFERENCE SOURCE - Make sure reference source is useful for assessing classification schemes (e.g don't use aerial photos for determining different individual grasses etc.)
    2. DETERMINING SIZE OF REFERENCE PLOTS - Match spatial scale of reference plots and remotely- sensed data
    3. DETERMINING POSITION AND NUMBER OF SAMPLES - Make sure to adequately sample the landscape, taking a variety of spatial sampling schemes. The greater number of reference samples the bettie (need to balance this with the cost (time/resources) required
  16. 16What are the different sampling methods for accuracy assessment reference data?1. Simple random sampling - observations randomly placed
    2. Stratified random sampling - minimum number of observations are randomly placed in each category
    3. Systematic sampling - Observations are placed at equal intervals according to a strategy
    4. Systematic non- aligned sampling - a grid provides even distribution of randomly placed observations
  17. 17What happens after reference data and locations are chosen for accuracy assessments?- Reference data are compared with the classified map, creating an ERROR MATRIX
    - Total accuracy = number of correct pixels/ total number of pixels
  18. 18What does an error matrix look like?Diagonals - Represent pixels classified correctly according to the reference data
    Off- diagonals - Mis-classified pixels    What does an error matrix look like?
  19. 19What is the problem with total accuracy calculations for error matrices?- The summary value is an average
    - It doesn't reveal if an error was evenly distributed between classes or if some classes were really bad/ some really good
    - One badly classified landcover type will have an effect on total accuracy (It is therefore necessary to include producers/ users accuracy too)
  20. 20What is meant by the users accuracy when included in an error matrix?- Corresponds to the error of commission (inclusion)
    - Have to find how many of the pixels on the map are actually what they say they are
    Calculated as:
    1.Number of pixels correctly identified in a given class/ number of pixels claimed to be in that class
    2. Number of correctly classified pixels in a class/ row total for that class
  21. 21What is meant by the producers accuracy when included in an error matrix?- Corresponds to the error of omission (exclusion)
    Calculated as:
    1. Number of pixels correctly classified in a given class/ number of pixels in the reference data in that class
    2. Number of correctly classified pixels in a class/ the column total for that class
  22. 22What is mapping land cover type based on?The fraction of solar energy reflected by the surface
  23. 23Why is there need to map land cover types?1. To understand regional to global scale environmental phenomena
    2. Environmental degradation and climate change have become major issues, especially through the suspected human activity
    3. Aim is to identify the nature, extent and severity of degradation processes to provide the necessary information to better manage the global environment
    4. The quantitative information can be put in to a model to produce better predictions of future events
  24. 24What is the cycle shown for removal of forest canopies in relation to the importance of not only being able to define the land cover type (forests), but the amount to which it is degrading?What is the cycle shown for removal of forest canopies in relation to the importance of not only being able to define the land cover type...
  25. 25Why is it important to estimate vegetation properties?1. Inter and intra annual global vegetation monitoring on a periodic basis
    2. Global biogeochemical, climate and hydrological modelling
    3. Net primary production (NPP) and carbon balance
    4. Detecting anthropogenic and climate change
    5. Agricultural activities (plant stress, harvest yields, precision agriculture)
    *Can measure; quantity/structure/ chemical composition (pigment content of chlorophyll etc.)*
  26. 26What is the importance of having composite bands?- In one band it can be hard to distinguish land surface features
    - Composites provide a clearer picture of the scene - more than one band is then used to get better information
  27. 27What could a ratio of wavebands be used for when looking at vegetation properties?- Two objects of identical material will appear different if they are receiving different amounts of solar illumination
    - The ratio between matching pixels in each band will remain the same
    - The ratio of two wavebands can reduce the 'noise' in the data
  28. 28What are vegetation indices (VI)?- Have better sensitivity than individual spectral bands for detection of biomass
    - There are more than 50 common vegetation indices
    - The ideal VI should be particularly sensitive to vegetative covers and insensitive to noise (soil background/ atmospheric and environmental effects/ solar illumination geometry and sensor viewing conditions
    - SR (simple ration) and NDVI (normalised difference vegetation index) are commonly used VI
    - They are designed to monitor for specific conditions (EVI - enhanced vegetation index or MTCI - MERIS terrestrial chlorophyll index)
  29. 29What is the soil line?- A linear relationship between the NIR and red reflectance of bare soil as characterised by slope and intercept parameters
    - Vegetation indices use soil line parameters extensively in crop growth analyses
  30. 30What is the simple ratio?- The NIR to red simple ration is the first true vegetation index
    - SR = NIR/red
    - It is unbound (has no upper limit)
    - SR is close to 1 for soil and for vegetation, SR is generally above 1
  31. 31What are the disadvantages to SR?1. Any sensor noise that is localised in a particular band is amplified by the ratio calculation
    2. It has a very big range and values range from decimal values less than 1 (for NIR less than RED) to values much greater than 1 (for NIR greater than RED)
    3. This range of values posed some difficulties in interpretation, scaling and contrast enhancement
  32. 32How is the normalised vegetation index (NDVI) calculated?How is the normalised vegetation index (NDVI) calculated?
  33. 33What is the normalised vegetation index (NDVI)?- Healthy, dense vegetation has high NDVI stressed/ sparse vegetation produces lower NDVI
    - Bare rock and soil have NDVI near 0, snow produce negative NDVI values and clouds produce low to negative values of NDVI
  34. 34What are the disadvantages for the normalised difference vegetation index (NDVI)?1. Saturates at high biomass/ dense vegetation
    2. Any factor that unevenly influences the red and NIR reflectance will influence the NDVI - atmospheric path radiance/ soil wetness
    3. Very sensitive to background variation (therefore lots of different indices were developed)
  35. 35What is the soil adjusted vegetation index (SAVI)?- Introduced by Huete in 1988
    - Designed to remove the dependency of the NDVI on the brightness of canopy background reflectance
  36. 36What is the equation for SAVI?What is the equation for SAVI?
  37. 37What is the advantage of spectral indices?- Have better sensitivity than individual spectral bands for detection of vegetation properties
    - There are many different Vis because they are affected by different environmental factors and but the spectral wavebands and satellite sensors
    - Vegetation indices can be used to manage, monitor and understand our environment
  38. 38What vegetation indices have been designed to monitor specific vegetation properties?1. EVI (Enhanced Vegetation Index) - designed to estimate LAI (leaf area index) and canopy structure, providing information on vegetation amount (the one sided green leaf area per unit ground area) - a board MODIS
    2. MTCI (MERIS Terrestrial Chlorophyll Index) - designed to estimate chlorophyll content in the vegetation canopy, providing information on the vegetations condition
  39. 39What is the equation for EVI?What is the equation for EVI?
  40. 40What are the advantages of EVI?1. Has a variety of wavebands, including a blue band that allows for residual atmospheric contamination
    2. Minimises soil background effect
    3. Sensitive to high biomass area - saturates at higher biomass level less than NDVI (NDVI saturates at high biomass)
    4. LAI values go up to 8 (for v.v.v. dense rainforest)
    NB: Landsat has similar wavebands to MODIS (could make a type of EVI)
  41. 41What is special about the Red and NIR wavebands?- They are very contrasting
    - Good for measuring vegetation
    - Phenology will affect NDVI
    - A negatie NDVI (anything below 0) is likely to be water
    - Sandy backgrounds can affect NDVI (as seen in the greening of the Sahara)
  42. 42Why do we need information on chlorophyll content?- Vegetation conditions include
    1. Green colour
    2. Photosynthetic rate
    3. Physiological status
    - Environmental understanding includes
    1. Nutrient cycling
    2. Carbon budgets
    3. Productivity
  43. 43What is the MTCI?- Uses spectral information in the red edge, as this is most sensitive to changes in chlorophyll (3 different regions)
    - With an increase in chlorophyll content, the difference between band 10 and band 9 increases and between band 9 and band 8 decreases
    - The information from this has been put in climate models
  44. 44What problems are there with vegetation indices?- NDVI data can be different as sensors can have marginal differences between the wavebands they use in the Red and NIR channels
    - NDVI from different sensors cannot be compared
  45. 45What are the advantages of vegetation indices?1. Simplicity - no assumptions and its not modelled
    2. Relationships developed in the field translate directly (or scale-up easily) to the satellite data sets
    - Relationship between LAI and VI can be extrapolated to a larger area (NB. these are empirical relationships (no direct link between the parameter, but evolves in a similar way) and only valid at the site and time they were developed
    3. Widely used from land cover, climate models and operational applications
  46. 46What is plant phenology?- The study of the timing of recurring life cycle events
    - For plants these events are bud burst/ leaf unfolding/ blossoming/ fruit ripening/ leaf colouring
    - You can see more defined areas in agricultural regions in NDVI as the plants will have a similar phenology
    - Higher temporal resolution is important to characterise the seasonal cycle
  47. 47Why is it important to monitor plant phenology?1. Tracks biotic response to climate change
    2. Estimates biological productivity
    3. Understands the land - atmosphere interactions
    4. Models vegetative inputs into biogeochemical cycles
  48. 48What are the components of phenology?1. Time of start of season (SOS) / onset of greens - beginning (end) of measurable photosynthesis
    2. Length of the growing season - duration of photosynthetic activity
    3. Time of maximum NDVI - time of maximum photosynthesis
    4. NDVI at end of season/ end of greeness - level of photosynthetic activity at EOS
    5. Seasonal integrated NDVI - photosynthetic activity during growing season
    6. Rate of greenup (senescence) - speed of increase (/decrease) in photosynthesis
  49. 49What has NDVI been used to achieve?- Despite its limitations, its been extensively used in remote sensing for;
    1. Primary productivity
    2. Identifying eco regions
    3. Monitoring phenological pattern - phenology is unique to each vegetation and can show when areas are stressed
    4. Assessing the season length
    5. It can be estimated from most of the spacebourne sensors (global data since 1981)
  50. 50How can NDVI curves be used to identify key phenological dates?- Inflection points are areas where the curve changes and indicate a changing period
    - NDVI and phenology are used to estimate yields
    - Observations tend to be noisy - smoothing the curve makes it easier to analyse and identify the phenological parameters
    -Can see the 'growing degree day' as it shows the day it is warm enough for first growth
    - Anomalies show how far a value is departed from the expected value
  51. 51What are biophysical parameters?- Physical properties of terrestrial surfaces due to geosphere- biosphere- atmosphere interactions, including;
    LAI/ biomass/ leaf chlorophyll content/ fractional vegetation cover/ Net primary productivity/ snow cover/ lake ice cover/ surface spectral albedo
    (These parameters are often used to assess changes in habitat condition/ vegetation distribution/ C sequestration)
  52. 52How has NDVI been used to make for biophysical parameter estimation?- Grain yields from reflectance readings of NIR and Red can indicate fertiliser recommendation based on the predicted yield
  53. 53What happened during the Amazon droughts?- Two severe droughts in 2005 and 2010
    - Used VI to study the response of vegetation
    - Used EVI data to findthe amazon drought plants were not light limited and some plants were able to grow very quickly
    - In different areas of the world, plants have different limiting factors
    - Lack of correspondence between drought years and negative anomalies
    - The years 2010 and 2005 were not the years of maximum decline for the majority of the forested areas
    - There exists an overall declining trend in vegetation greeness over the 2000 to 2010 period
  54. 54What is the satellite data availability?1. SATELLITE DATA: Unprocessed (radiance and reflectance, which includes processed (data products) and can be from space agencies
    2. DATA PRODUCTS (where land cover maps are already made) :
    1. Land - Surface temperatures/ land cover type/ LAI
    2. Atmosphere - Aerosols/ cloud types
    3. Ocean - SST, chlorophyll concentration
    *Can be collected on a daily/ 8-daily/ monthly and annual temporal basis
    *Image, tile or global spatial coverage
  55. 55What happens to NDVI at the rate of green up?* It decreases
    *Phenology looks into the plant lifecycle
    *MTCI is the VI used to estimate chlorophyll leaf content
  56. 56What is change detection useful for?*For before and after events
    1. Mapping Arctic sea ice
    2. Showing change of land cover type
    3. For natural disasters
  57. 57What is important to look at to make use of after mapping land cover types and using reflectance measurements to quantify surface characteristics?- The temporal dimensions
    - Temporal information is important for the scales and they cycles (desertification gradual/ deforestation very fast)
    - Characteristics can then be looked at at different spatial scales
  58. 58What is needed to be known for looking at change detection?1. Where and when has the change taken place in the landscape
    2. How much change has occurred and what type is it?
    3. What are the cycles and trends in the change?
    - Change detection involves identifying the differences in the state of an object/ feature or phenomenon and quantitatively analyses the temporal dynamics of the phenomenon or object/ feature
  59. 59What are examples of phenomena/ objects that can lead to land surface change?1. Urbanisation
    2. Deforestation
    3. Changing agricultural land use patterns
    4. Desertification
    5. Natural disaster
    6. Geomorphic processes
    7. Plant community changes
  60. 60What types of change are there?1. Short term - synoptic weather events
    2. Cyclic - seasonal phenology/ diurnal surface temperatures
    3. Directional - Urban development (one land cover to another)
    4. Multidirectional - Deforestation and then regeneration (one land cover to another and back)
    5. Event - wildfires/ disasters
  61. 61What is meant by change detection?It is the process of identifying the differences in the state of an object or phenomenon by observing it at one or more different times
    (Takes two images at Time 1 (T1) and time 2 (T2) and uses multi temporal analysis to make a change map)
  62. 62What does change detection assume?- Change in two images is due to change in landcover amount/ condition/ type and NOT external factors - attention needed for changes in reflectance due to differences in
    1. Vegetation phenology - esp. in agricultural areas, where different crops can be used over the year
    2. Soil moisture - especially in floods
    3. Atmospheric conditions
    4. Solar angle (time of day/ year) - anniversary images
    5. Sensor characteristics
    *This is part of image selection and pre- processing and needs to use data from the same satellite, which is matched to the temporal scale*
  63. 63What is important to consider for change detection?- Important to have same spectral, spatial and radiometric resolution
    - Imagery to be used for sensors with different spatial scales have to be resampled to sample resolutions
    - There has to be accurate image to image registration (pixel offsets will increase change detection errors)
    - Atmospheric correction/ image to image normalisation (minimising pixel value differences)
  64. 64What is meant by mosaicing images?- In larger spatial scales, images to be mosiaiced joins can be prominent as the solar angle could be different, so images have to be normalised, by deriving a relationship between different pixel reflections
    - Image to image normalisation
    - Identifies psuedo- invariant targets (roads/ roof tops)
    - Collects pixel values for these targets in both areas
    - Derives a linear transformation between these to the data set
  65. 65What are the specific considerations for the impacts of clouds?- There might be a change in the image which might not be the type of change that is needing to be mapped/ detected
    - The cloud is to be masked in both images
    - Clouds will obscure the surface and masking means loosing data but the wrong change isn't detected
  66. 66What is the process for being able to create a change detection map?What is the process for being able to create a change detection map?
  67. 67What are the pixel based change detection methods flow chart?What are the pixel based change detection methods flow chart?
  68. 68What is meant by image differencing?Du = DNu (t2) - DNij (t1)
    - Imagery of one date is subtracted from the imagery of another
    - Imagery can be of reflectance wavebands or band rations
    - Positive and negative values are interpretable
    - Pre - processing makes the images as close together as possible
    - The ideal no change is 0
  69. 69What are the pros and cons to image differencing?PROS
    1. simple to apply
    2. Easy to interpret the results
    3. Fairly robust
    CONS
    1. requires atmospheric correction for correctly identifying
    2. The method offers no way of assessing change from one class of land cover to another
    3. The same value may have a different meaning depending on the starting class (two sets of different absolute values may yield the same difference values)
  70. 70What is meant by image ratioing?Ru = (DNu (t2)) / (DNij (t1))
    - Variation in viewing condition can confound identification of surface properties (shadows (on mountains/ trees/ buildings) and seasonal reflectance differences
    - Reduce these effects by image ratioing
    - Image ratio = image1/ image2
    - No change = 0
    - Change is > or < than 1
    - Change will be in the tails of the image histogram
    - Image histograms show change at the extremes - Standard deviation is used to show the extent of the degree of change
  71. 71What happens when features don't have the expected value from the ratio?- Due to external factors, stable features may not have an expected value of unity from the ratio
    - The critical step is to determine the threshold value to differentiate change/ no change (which is often empirical)
    - No. from- to information on change detection is available
    - Change detections is about applying method and then using the threshold to show the degree of change (mainly through the standard deviation)
  72. 72What is meant by post classification analysis?- Its the most common method for change detection
    - Involves independently classifying images from two dates
    - An algorithm is used to compare the classified images to produce a map indicating areas of change
    - Provides qualitative information on the nature of change (change of land cover X to land cover Y)
    - Areas of the change area can be calculated by counting the number of pixels changed (e.g. a landsat pixel is 30mx30m)
    - This can create a change decision matrix
  73. 73What does post classification analysis require?1. Accurate classifications
    2. Errors will propagate and magnify (e.g. 0.8 x 0.8 x 100 = 60% classification accuracy in the change detection output)
    3. Accurate co-registration (or gets errors at class boundaries)
    4. Misses changes within a class
  74. 74What is the Coastal Change Analysis Program case study?- National coastal land cover and change mapping programme
    - Satellite based map products
    - Standardised data and methods
    - Designed to help improve understanding the linkages between land cover and the environment
    - Some land clearing became evident in preparation for residential development
    - Much of the forest canopy had been defoliated due to hurricane and continued residential development
  75. 75What are the steps for creating change detections?1. Obtaining images
    2. Pre- processing
    3. Classification
    4. Creating change detection matrices
    - The diagonal elements show stable areas and off - diagonal areas can then show the changes
  76. 76What is the summary for different change detection methods?What is the summary for different change detection methods?
  77. 77What is meant by an error matrix?- Each type of accuracy estimate yields different information
    - If we focus on only one, there can be an erroneous sense of accuracy
    - The diagonals look at measurements that are correct (total error)    What is meant by an error matrix?
  78. 78What is the Kappa statistics?*Reflects the difference between actual agreement and the agreement expected by chance
    *Shows the extent to which the correct values of an error matrix are due to 'true' vs. 'chance' agreement
    *Ideal case is c.a. = 0, o.a. =1 and Kappa =1
    *A kappa of 0.85 means there is an 85% better agreement than chance alone
  79. 79What is the equation for Kappa?What is the equation for Kappa?
  80. 80What are the other uses of the Kappa statistic and other measures of accuracy?Kappa
    1. Compares two error matrices
    2. Weight cells in the error matrix according to severity of misclassification
    3. Provide error bounds on accuracy
    Other measures of accuracy
    1. Z statistics
    2. Normalised accuracy
    3. Conditional kappa
  81. 81What is the difference between land cover and land use?Landcover - The observed (bio)physical cover on the Earth's surface (water/ forest/ wetland)
    Land use - Characterised by the arrangement, activities and inputs people undertake in a certain land cover type to produce, change and maintain it (agriculture/ commerce)
  82. 82Why does land cover matter?- Land cover is a principal factor controlling the exchange of energy, water, gases and nutrients within the Earth's system
    - For use in Global change modelling (boundary conditions for GCM/ in global biogeochemical and hydrological models)
    - Land cover/ use change impacts carbon, water and energy cycling at all spatial scales
    - For biodiversity, resource management, fire and disaster monitoring
    - European Space Agency has made it one of the things to monitor climate change - an ECV - Essential Climatic Variable
  83. 83What does the nature of a landcover map depend on?1. The purpose of the study (what information is needed - defining land cover maps is not straightforward and need to consider how to define boundaries)
    2. The data collection method employed (technical capabilities and constraints)
    3. Cost (data collection and processing are time consuming and expensive - the more detailed it is, the more time/ information and expense)
  84. 84What should landcover maps normally be?1. MUTUALLY EXCLUSIVE - Meaning there is no taxonomic overlap of any classes (deciduous forest/ evergreen forest as distinct classes)
    2. EXHAUSTIVE - All land cover classes present in the landscape are counted for and there are no omissions
    3. HIERARCHICAL - Sub level classes are hierarchically combined into a higher- level category that can make more sense. There are different levels of increasing complexity
  85. 85What is meant by being hierarchical?What is meant by being hierarchical?
  86. 86What is the relationship between detail and cost/ resolution?- Detail and cost are negatively related (more expensive for each level of feature detail)
    - Detail and resolution are positively related - smaller pixel is higher resolution
  87. 87What is the relationship between detail and cost/ resolution in a graphic form looking at different levels of data?What is the relationship between detail and cost/ resolution in a graphic form looking at different levels of data?
  88. 88What are the different land cover mapping schemes and how do they differ?- Provides different definitions for landcover classes
    - Use different data sets
    - Defined for a specific purpose
    E.g. 1. USGS Land use/ land cover scheme
    2. International Geosphere Biosphere Programme (IGBP)
    3. Global Landcover 2000 landcover database
  89. 89What is the USGS Land use/ land cover scheme ?- Provided the first generalised landcover map
    - First produced in 1976
    - Used landsat data and aerial photos
    - Based on Anderson's classification scheme
    * 4 levels of classification
    *Level one has 9 classes (Urban/ built up land/ agricultural land/ rangeland/ forest land/ water/ wetland/ barren land/ tundra/ perennial snow/ ice
  90. 90What is the IGBP land cover classification?- A 1-km resolution global land cover (one now being created of 30mx30m using a global landsat classification)
    - AVHRR data spanning a 12 month period (April 1992 - March 1993)
    - Unsupervised classification (K means)
    - 17 landcover classes
    - One level of classification
  91. 91What does the IGBP land cover classification map look like?What does the IGBP land cover classification map look like?
  92. 92What is the Global Landcover 2000 landcover database?- Produced by joint research centre, Italy in collaboration of over 30 research teams
    - Prepared from data required in 2000 by the VEGETATION sensor on the SPOT satellite
    - Spatial resolution of 1km
    - 27 landcover classes
    - Uses slightly different algorithms for each continent on the landcover base
  93. 93What is GlobCover and its data volume?- A whole year of data, using spectral and temporal characteristics to classify (using phenological timings) they bring one out every few years
    - Sensor is MERIS at 300m
    - Made from May 2005 and April 2006
    - Land cover areas between 82' N and 56' S
    - Total volume (9.56 Gb x 15 orbits x 365 days/ 3) = 17 TB
    (1,150km x 17,500 km for a full orbit)
  94. 94How was GlobCover used to create LCCS land cover maps?How was GlobCover used to create LCCS land cover maps?
  95. 95How is the GlobCover classification method used?How is the GlobCover classification method used?
  96. 96What is the comparison for the different land cover projects?What is the comparison for the different land cover projects?
  97. 97What did the land cover map of the UK (2007) show?- Produced by the Centre for Ecology Hydrology (CEH)
    - Follows on from the LCM 1990, LCM 2000
    - Habitat based classes
    - Based on summer and winter satellite data from Landsat TM5, IRS- LISS3 and SPOT sensors
    - Used external datasets for knowledge based enhancement (KBE)
    - Uses landsat data of 20-30m spatial resolution
  98. 98What is meant by object based classification?- Groups pixels together to segment the image and then classifies it
    - This was used for the UK landcover map for validation with field data and comparison with the countryside survey (an overall accuracy of 83%)
  99. 99What is meant by aggregation?- Aggregation up to a grid cell degrades the data down to fewer land cover types
  100. 100What was the data product for the 2007 UK landcover map?What was the data product for the 2007 UK landcover map?
  101. 101What is meant by landcover conversion?- A complete alteration in the cover of a tract of land
    (e.g. forest to agriculture)
    - Globally, 2.3 million sq km of forest were lost during the last decade and 0.8 million sq km of new forest was gained
    - In the 1990s, almost 70% of deforested areas were changed to agricultural land. In Latin America, most conversion was large scale, whereas in Africa it was small- scale agricultural enterprises dominated
  102. 102What is the importance of remotely sensed data for land cover conversion?- Remotely sensed data are extremely valuable tools for monitoring both human induced and natural changes on the surface of the Earth.
  103. 103What is meant by landcover modification as a type of landcover change?- A change in the character or condition, but not type of landcover (e.g. closed forest to open forest)
  104. 104What is the issue with landcover modification with respect to changing closed forest to open forest?What is the issue with landcover modification with respect to changing closed forest to open forest?
  105. 105What is the importance of urban landcover maps?- Can be used to understand urban heat island intensity
    - Urban heat island is found to be more intense at night and in the winter
    - UBI has several effects (direct and indirect) including ozone concentration, air quality, convective activity downwind of city, precipitation, cloud cover enhancement
  106. 106What is needed to be known for landcover monitoring and what does change detection involve?What to know:
    1. Where and when has change taken place in the landscape
    2. How much change and what type of change has occurred?
    3. What are the cycles and trends in the change?
    Change detection;
    1. Identifying differences in the state of an object/ phenomenon
    2. Quantitatively analysing the temporal effects of the phenomenon
  107. 107What type of sensors does thermal radiation use?- Passive sensors that measure emitted energy
    - Thermal complements the visible part of the spectrum
  108. 108In which ways can energy be transferred to one object to the next?1. Conduction - Occurs when one body (molecule/ atom) transfers its kinetic energy to another by colliding with it
    2. Convection - Kinetic energy of bodies transferred from one place to another by physically moving the bodies
    3. Electromagnetic Radiation - Can take place in a vacuum and be sensed remotely
  109. 109At what temperatures will objects emit energy?- All objects (inc. rocks/ soil/ vegetation) emit thermal infrared energy
    - So long as they are above -273K (this even can include ice caps...)
  110. 110What does the graph for atmospheric absorption of thermal insulation look like?In the thermal IR, there are many wavelengths with absorption features due mainly due to atmospheric water vapour, CO2 and O3
    (The grey areas show where there is little emmitance)
    *Thermal atmospheric windows at 3.5 microns and 8.14 microns (where absorption is minimal) is where data is used for remote sensing    What does the graph for atmospheric absorption of thermal insulation look like?
  111. 111What is the importance for clouds?- Thermal can see 'beneath the clouds'
  112. 112What is meant by a blackbody?- An ideal surface that absorbs and then remits ALL energy incident upon it
    - Real objects emit only a fraction of the energy emitted from a black body at the same temperature
  113. 113What is meant by emisitivity?- The ratio between the radiant existence emitting from a real world object and that from a blackbody at the same temperature
    - Amount of energy emitted: amount of energy absorbed
    - Albedo is the opposite, measuring how much energy a surface reflects at all wavebands
  114. 114What are the 3 laws to describe thermal properties of objects?1. Plancks law - Describes the distribution of energy with wavelength as a function of temperature
    2. Stefan- Boltzman Law - Describes the relationship between the total energy emitted and temperature
    3. Weins displacement Law - Describes the spectral distribution of energy as a function of temperature
  115. 115What is Plancks Law?- Shows that energy distribution with wavelength depends on temperature
    - Has a characteristically shaped graph of frequency distribution
    *Steep rise at short wavelengths
    *Long tail at long wavelengths
  116. 116What is the Stefan Boltzmann law?- The amount of energy (overall wavelengths) that an object radiated can be expressed by the Stefan boltzmann law
    - The exponent of ^4 shows a small increases can create an exponential emmitance of energy (Area under the curve is the total area) - total energy emitted increases rapidly with increased temperature
    - The law is expressed for an energy source that behaves like a blackbody
    * As the Earth is cooler than the sun, it emits a fraction of the energy and the area under the curve is much smaller
  117. 117What is Weins displacement law?- Just as total energy varies with temperature, the spectral distribution of energy also varies
    - The dominant wavelength at which a blackbody radiation curve reached a maximum is related to temperature by Weins law
    - Wavelength at peak energy emission (can then work out the peak wavelength)
  118. 118What satellite sensors are there with Thermal InfraRed?TIR (8-14 microns)
    1. Landsat TM/ETM (band 6)
    2. MODIS
    3. ASTER
    4. SLSTR
    MIR (3-5 microns)
    1. MODIS
    2. SLSTR
    * NB. SEVIRI and GOES the meteorological satellites have both
  119. 119What is the importance of cloud detection?- Cloud detection is a function of contrast between the target and the background. Contest can be;
    1. Spatial - textural information to sperate cloud from the surface
    2. Temporal - Clouds detached in a sequence of images if the clouds are moving (cloud motion vectors allow detection of wind direction/ speed)
    3. Spectral - Determined by the radiative properties of the cloud and surface (contrast due to 1. (in visible) Albedo 2. (in IR) Temperature differences
  120. 120Why should clouds be identified?*Clouds affect atmospheric radiative transfer in the...
    1. Optical domain - albedo influenced by nature/ size/ shape of cloud particles
    2. Thermal - (water a strong absorber/ emitter of energy) higher clouds at the top with cooler the particles and lower the rate of emission
    *An important 'mask' for other terrestrial satellite data products (e.g. NDVI) if interested in observing the ground
  121. 121How can clouds be detected?- Clouds appear very bright white in a visible image
    - Its difficult to distinguish between cloud heights
    - High atmosphere clouds are easier to spot
    - The properties of the cloud affect their reflectance;
    1. Bright white = v. cold temps and high cloud
    2. Black = v. warm temps and low/ surface cloud
  122. 122What is the consequence of clouds?- If pixels are large, the whole pixel can be 'cloud contaminated', which can massively affect the spectral information
    - Developing cloud masks can shield the interference
    - MODIS develops a cloud mask map
    - Landsat hasn't made one, so you have to do it yourself
  123. 123What is the Urban Heat Island?- Urban areas are warmer, like an 'island' of heat surrounded by cooler rural areas (as much as 4'C per km...)
    - Pattern of urban temperature distribution influenced by land uses (open areas/ lakes/ impervious surfaces etc.)
    - Very tall buildings make for higher temperatures
  124. 124What is meant by the intensity of the urban heat island?- Depends on the urban - rural temperature difference
    - Elevated temps/ impact of energy demand/ air pollution (O3)/ rainfall/ heat- related illness and mortality have called for the need of COOL TECHNOLOGIES (Reflective/ green roofing and more green space) - lakes and parks in urban areas can cool the area
    - Surfaces in the shade are 30- 50' hotter than paved areas
    - Urban areas are expanding rapidly, so this is only going to increase (54% of worlds pop live in urban areas in 2014 - rise to 66% by 2050)
  125. 125How is thermal remote sensing used for monitoring volcanoes?- Crater lakes on volcanoes act like calorimeters, absorbing he heat given off by intruding magma
    - Approximately 12% of the worlds 700 or so active volcanoes have crater lakes
    - Satellites remote sensors is an important early warning system for volcanic eruptions (lava and pyroclastic flows and ash clouds) - can predict/ monitor where flows are going.
    - MODIS looks at things on a daily basis
  126. 126What changes climate?Variations in
    1. Sun's output
    2. Earth's orbit
    3. Drifting continents
    4. Volcanic eruptions
    5. Greenhouse gases
    These can be seen through biomass burning
  127. 127What is the Keeling curve?- Shows the change in CO2 concentrations - atmospheric measured in Mauna Loa, Hawaii
  128. 128How does CO2 impact on radiative forcing?- Energy absorbed and reflected
    - Energy absorbed gets reemitted in thermal wavelengths
    - Increasing GHGs traps more heat
    -
  129. 129What is biomass burning an important primary source of?1. CO2 - 12-36ppm of increase between 1850 and 2000. 1.6GtC/yr in 1990s, now 6.4 from fossil fuels
    2. CO - around 1/4 of the total source
    3. Hydrocarbons (inc. CH4)
    4. NOx - around 15% of total source
    5. OC - around 20-60%
    6. BC - around 2/3 (black carbon)
    nb. lots of measurements come with big uncertainties
    OUT OF 13 RADIATIVE FORCING AGENTS, 8 ARE IMPACTED BY FIRE
  130. 130What is the terrestrial vegetation cycle?- Fluxes between the biosphere and the atmosphere set the seasonal patterns in atmospheric CO2
    - Through respiration, photosynthesis and decomposition
    - Combustion is essentially photosynthesis in reverse (oxygen and carb --> CO2, water and energy)
  131. 131What are the natural and anthropogenic causes of fire?NATURAL = volcanoes and lightning (particularly in boreal regions (canada/ Siberia)
    ANTHROPOGENIC= land clearance/ deforestation (accounts for 40% of Amazonian fires)/ agricultural clearance/ pest control/ land management/ arson and accident - discarded cigarettes out of car windows
  132. 132What is the importance of fire?1. Major carbon cycle and land cover change component - alters the land surface optical properties (spectral albedo) , important in radiative budget and can change vegetation characteristic and type
    2. Land - atmosphere flux (emissions)
    - CO2/ CH4/ CO and aerosols, accounts for 30% of cloud condensation nuclei, perturbs precipitation efficiency
    3. Changing climate may increase fire incidence - warmer and drier conditions increase the risk
    4. A natural hazard with large societal costs and impacts - economic damage (expensive to put out and evacuate too) and impact of air quality and health
    5. Fire occurs in atmosphere, not just surface - peatlands at high risk
    6. Can be carbon neutral if there's regrowth, but risks invasive species and organic soil can sometimes be burnt
  133. 133Why should wildfires be monitored?- To develop management practices
    - Early warning systems
    - Understand implications for biophysical processes and atmospheric composition
    - Fires are very dynamic and not much is known about them
  134. 134How does biomass burning affect human health?- Peat fires in Russia in 2010 very bad because of the aerosols
    - Smaller aerosols travel further through the body <10 microns go into airways, <5microns through bronchus, <2-3 into alveoli, <0.1 = alveolar diffusion
    - Estimated around 230,000 premature deaths a year are due to smoke emissions
  135. 135How much of the CO2 growth rate is attributed to biomass burning?- around 2/3, spikes in 2003 and 1998 correlate to years of high fire activity - both over 2ppm CO2
  136. 136How can earth observation instruments provide information on biomass burning?1. Burned area mapping - reflectance observations and detecting burnt areas days or weeks after the fire (the longer the fire takes hold, the more complete the combustion and the whiter the ash). Light burns are the most difficult to measure
    2. Active fire detection - thermal observations in emitted energy (heat), detects actively burning fires at time of satellite overpass
  137. 137How do different degrees of burn look in remote sensing images?How do different degrees of burn look in remote sensing images?
  138. 138How are burnt areas mapped using remote sensing techniques?Uses wavelengths between 0.4 and 2.5 microns (MODIS wavebands good for this)
    1. Detect spectral change in the surface reflectance, characterised by the removal of vegetation and alteration to the structure, deposition of char and ash and the exposure of the soil layer
    2. Spectral change depends temporally and spatially due to;
    1. the type of vegetation burnt
    2. Amount burnt
    3. Rate of dissipation of char/ ash by wind and rain
    4. Rate of post fire vegetation regeneration
  139. 139What indicie is used to detect burned areas?- Normalised burn ration (NBR)
    - (NIR - SWIR)/ (NIR + SWIR)
    - 1.6 microns SWIR band responds differently to soil and vegetation moisture
  140. 140How are active fires detected?- Thermal radiations and smoke emitted
    - Fires have very high temps compared to ambient surroundings, these high temps result in very intense radiant energy emissions particularly in the MIR (3-5micron) spectral region
    - Peak wavelength in MIR for fire detection as is very sensitive to temp and fire
    - Uses MIR and TIR (11 microns), exploiting the high and low sensitivity of MIR and TIR channel to thermal emissions
    - Much less energy is emitted by fires >400'C at 11 microns than 4 microns
    - Subtraction of TIR from MIR allows for clearer identifications of active fires
  141. 141What is the distribution of fires?- Seasonal cycle of fires changes globally and temporally
    - Peak time for a fire is 2pm local time
    - Aerosols moved around the atmosphere with weather patterns (meteorologists can warn towns and cities of poor air quality)
  142. 142What are the major pools of the carbon cycle?-Earth's crust 1billion Pg C
    - Oceans 38,000 Pg C
    - Atmosphere 750 Pg C
    - Terrestrial ecosystems - 560 PgC Veg+ 1500 PgC soil
    Carbon cycle is the exchange of carbon between pools. Carbon is the building blocks of life and influences global warming - an important part in Earth System Science
  143. 143What are the components of the terrestrial carbon cycle?1. Photosynthesis - use of solar energy in a chemical reaction to convert CO2 and water to carbs
    2. Respiration - Breakdown of complex organ substances to relate energy, CO2 and water
  144. 144What is Vegetation primary production and what ways are used to measure it?Vegetation primary production - Threat at which radiant energy is made into and stored as organic substances (kgC/m2/yr)
    1. GPP - Gross Primary Productivity - rate at which plants capture and store energy over a given time period per unit area. Doesn't consider the cost (respiration in photosynthesis)
    2. Net Primary Productivity - difference between GPP and total energy lost due to plant respiration over the same period
    BIOMASS - a state variable. Standard mass within a plant - everything stored within it
  145. 145What is the importance of vegetation primary productivity?- Entry point of atmosphere CO2 to terrestrial ecosystems
    - Carbon storage and sequestration
    - Climate change feedback (+ve/ -ve)
    - Provision of food and fibre, sustaining life
  146. 146How is primary productivity modelled and measured within remote sensing?1. Allometric methods - Standard method with well established protocols - mostly measuring biomass
    2. Flux tower measurements- Based on eddy covariance/ exchange of air in the atmosphere/ biosphere boundary layer
    3. VI -empirical approaches - Developing relationship between vegetation indicies and productivity
    4. LUE based models - Based on Monteith's light use efficiency concept - common EO based approach
    5. Process based models - Accurate representation of biophysical processes such as photosynthesis using mathematical functions
    (6. Field measurements involve measuring the girth of the tree/ cutting it down to burn in lab)
  147. 147What is the Allometric method?- Field based method for estimating AGB
    - Can be upscaled using remote sensing data (LIDAR and SAR)
    - Depends on landcover
  148. 148What is the flow chart for the Allometric method?What is the flow chart for the Allometric method?
  149. 149How are flux towers used?- Measure the CO2 (and other gases) exchange rates between terrestrial ecosystems and the atmosphere
    - Spatial coverage is limited
    - Used as an important validation of EO estimates
  150. 150How are vegetation indicies used?- VIs based on the reflectance measured in the visible (red) and NIR spectrum
    - Link the VIs to GPP/ NPP - e.g. integrated into NDVI/EVI for annual productivity
    - Relationships can be site dependent/ specific
    - Saturation in most VIs can affect relationships
    - Site specific - high levels of leaves doesn't have the same level of increase for VIs
  151. 151What is LUE?- Estimating primary productivity using RS is mainly done via LUE
    - Uses the photosynthetically active range (PAR) energy beyond the range that plants can't use
    - Is the efficiency of how plants convert CO2 to dry matter
    - the efficiency will normally be less than the maximum possible duet limiting factors
  152. 152What is the LUE concept?- First noted by Monteith (1972) that in healthy annual crops, GPP was directly related to absorbed solar radiation in the PAR range (400-700 nm) and LUE
    - APAR depends on amount of PAR received during the growing season/ area of green leaves in the plant canopy
  153. 153What is the general LUE based model equation?GPP = PAR*FARPAR*LUE(emax)*Scalars
    Where FAPAR = fraction of absorbed PAR by the canopy and Scalars = temp, Ppt, VPD
    - FAPAR shown by the near - linearly related to vegetation indicies (NDVI)
    - PAR, Scalars can be derived from remote sensing data too
    - LUE term constant at plant functional type level (coniferous forest/ deciduous forest etc.)
    - Most RS based models are driven by this equation
  154. 154What is meant by estimation of LUE?- Different types of vegetation have different LUE coefficients making it a main source of uncertainty
    - Approaches under exploration: to derive LUE directly from remote sensing
    - Use of VI and developing means of estimating Chlorophyll Fluorescence which has been shown to be linearly related to LUE-subject of the new ESA's FLEX explorer mission to be launched in 2022
  155. 155What are the problems in estimating LUE using PRI?The photochemical reflectance Index (PRI)
    - Most operational sensors don't measure data in the 570nm waveband
    - Reflectance difference between 531nm and 570nm very small
    - PRI is sensitive to sun/ sensor geometry
    - PRI = R531 - R570/ (R531 + R570)
  156. 156How is FAPAR estimated?- FAPAR can be derived from RS data related to VIs
    - Canopy vs green/ photosynthetic FAPAR
    - Can be estimated using NDVI
  157. 157What is the MOD 17 model?What is the MOD 17 model?
  158. 158What is the SCARF model?- Based on the LUE concept, but uses quantum yield for C3 and C4 plants
    - Instead of canopy FAPAR, uses photosynthetic/ green FAPAR
    - C3 plants involve fixation of carbon and harvesting of light occurring in the same cells, C4 do different jobs in different cells (more efficient)
    - Seems to be more effective and closer to the flux tower values than MOD17 GPP
  159. 159What is the equation for the SCARF model?What is the equation for the SCARF model?
  160. 160How is the scarf model implemented?How is the scarf model implemented?
  161. 161What is the LAI?Leaf Area Index
    - Used to characterise land surface in the models
    - Active surface (photosynthetically active)
    - Related to canopy structure
    - Related to leaf function (absorption of PAR)
    - LAI one sided leaf area m2 of leaf per m2 of area of the ground, amount of leaf and a unitless
    - Process models not based on LUE robust in describing biophysical processes such as photosynthesis, but require a number of datasets to parameterise
  162. 162What is the issue with LUE models?- How to disentangle whole canopy FAPAR
    - Flux tower sites to validate models not in the tropical regions - where sequestion occurs
  163. 163—

Information

Description
Flashcards on Geog 2007 part 2, created by Pippa Holman on 27/10/2016.
Created by
Pippa Holman
Views
2
Rating

More on Remote Sensing

Feedback

What is it about?

What do you think of the new look?

How do you like it?