Scale in RS and Relation to OBIA

Question 1

OBIA

Answer 1

• Object Based Image Analysis
• Partitioning RS imagery into meaningful image-objects, and assessing their characteristics through spatial, spectral and temporal scale

Question 2

Science is?

Answer 2

• Method of learning about the physical universe by applying principles of the scientific method, which includes making empirical observations, proposing hypotheses to explain those observations, and testing those hypotheses in valid and reliable ways

Question 3

Pattern recognition = ?

Answer 3

• Process understanding

Question 4

Components of Scale in RS

Answer 4

• Grain

- Extent

Question 5

Grain

Answer 5

• Spatial resolution

- Function of IFOV of sensor

Question 6

Extent

Answer 6

• Function of sensor swath

Question 7

Trade-off of Scale in RS

Answer 7

• Trade-off: increase grain = decrease extent

Question 8

Scale challenges in RS

Answer 8

• Landscape patterns change depending on scale of observation
• No unique or optimal single scale for defining geographic entities of different size and shape
• Results and conclusions made at one scale may not be valid at other scales
• RS suffers from MAUP

Question 9

What is MAUP? What does it represent? How does it arise?

Answer 9

• Modifiable Areal Unit Problem
• Represents Sensitivity of analytical results to the definition of data collection units
• Arises b/c number of different ways by which a study area can be divided into areal units
• 2 components

Question 10

What are the 2 components of MAUP?

Answer 10

• Scale problem

- Aggregation problem

Question 11

MAUP scale problem

Answer 11

• Variation in results when areal units are aggregated into fewer and larger units for analysis
• e.g resampling same image band represented using larger pixel sizes (10m to 10m, etc.)
• Ex. crop health index single pixels = 132 and 81 while 2 pixel aggregation = 106.5 or 116.5 depending on if vertical or horizontal aggregation

Question 12

What can happen to r2 in linear regressions with MAUP issue?

Answer 12

• Aggregation of pixels to a different scale can increase r2
• Not a proper strengthening of data relationship
• More input data (no aggregation) with lower r2 has more explanatory power even though r2 is lower

Question 13

Aggregation/zoning problem in MAUP?

Answer 13

• Variation in results generated by alternate zoning schemes at the same resolution
• ex. aggregate vertical vs. horizontal pairs

Question 14

Statistical effects of MAUP

Answer 14

• Aggregating to larger and fewer units leads to smoothing of data
• Smoothed decreases variance
• Less variance = stronger correlation coefficients, r2, in most cases

Question 15

Model effects of MAUP

Answer 15

• Model inputs vary according to aggregation schemes

- Model results vary according to inputs

Question 16

Solutions to MAUP

Answer 16

• Abandon traditional statistics
• Conduct a sensitivity analysis (report variation in results w/ change of scale and aggregation
• Derive an optimal spatial resolution
• ID basic entities or objects (abandon pixels for object base approach)

Question 17

Optimal spatial resolution

Answer 17

• No simple solution, application dependent (classification, target detection, modelling, etc.)
• Requires understanding of spatial scale in relation to phenomena under investigation
• Different Phenomena may have different optimal resolutions even in one scene (ex. image classification)

Question 18

What are the 3 basic types of image classifications using statistical rules to assign classes to pixels?

Answer 18

• Supervised, unsupervised, and hybrid

- Accuracy is measure of success

Question 19

Classification Feature Space

Answer 19

• Graph DN’s of one band against DN’s of another band

- Spectral classes are groups of DN’s b/c relate to wavelength bands, give info on classification

Question 20

Mixed pixels

Answer 20

• pixels rarely contain pure signature
• Contain average brings of elements
• DN’s may not represent any true element in the image (composite signature)

Question 21

Which is best, high or low spatial resolution for classification? How does this relate to accuracy?

Answer 21

• Increase spatial res = competing effects
• Less proportion of mixed pixels in high spatial res = greater accuracy
• But also increased within-class spectral variability which = decreased accuracy
• Total accuracy is based on combined result of the 2 competing factors

Question 22

High spatial resolution in an image results in what to the variation?

Answer 22

• Increased variance

- Class separability is sensitive to variance (think of it in classification feature space)

Question 23

What is the problem when a size, shape, and orientation of a grid is superimposed over an area that we are trying to understand?

Answer 23

• MAUP

Question 24

What should the scale for a grid be matched to?

Answer 24

• The phenomena under investigation, often not possible

- Be aware of scale dependence and impact on chosen scale

Question 25

What is the clearest way out of a scale problem?

Answer 25

• Identify objects that relate to real world geographic entities
• Use non-arbitrary representation of space (pixels are arbitrary)
• Called Object-Based Image Analysis

Question 26

While pixels represent an arbitrary representation of geographic phenomena at a given scale, objects provide?

Answer 26

• Objects provide structure to the scale issue and offer solutions to the MAUP