Adv Remote Sensing

What are the four components of Change Detection Analysis?

Identification (where the change has happened)

Quantification (magnitude of difference, extent, rate of change)

Characterization (what type of change, forest loss, regrowth, fire, land cover conversion)

Accuracy Assessment (how reliable, reference with management plans, GEI, other reference data)

What is the difference between Land Cover and Land Use?

Land Cover is the biophysical component of what Remote Sensing sees (i.e. vegetation, what is actually on the ground).

Easy to detect and more straightforward.

Land Use is how the land is used by humans (i.e. pasture, plantation, residential, industrial). Often needs extra info like zoning and management records, on top of RS.

What are the two types of temporal analysis?

Change Detection

Time Series

What is Precision and Recall?

Precision involves the False Positives, i.e. out of the pixels that were mapped, how many were actually changed? P = TP/ (TP+FN). High precision means there were not a lot of false positives, aka user accuracy is high

Recall: out of all the pixels that changed, how many were we able to detect? High recall means that there were little to missed detections aka false negatives AKA Producers Error.

What is overall Accuracy?

Total True (+) + (-) over everything.

What is the issue with overall accuracy? How does Precision and Recall account for this?

Can be misleading because if there is a small change in an environment where change is rare, the algorithm may not detect change, but overall accuracy will still be high.

Precision and recall account for this …. not sure

What is an F1 Score and how does it incorporate precision and recall?

harmonic mean of precision and recall.

F1 = 2x (P+R)/(P-R)

Values 0-1

1 is perfect aka good balance of precision and recall.

What are the overarching steps for a change detection workflow?

Data Acquisition - satalliete imagery, spectral data, ancillary data (land use)

Pre-Processing (atmospheric correction, geometric co-registration, cloud masking)

Change Detection Variables (spectral bands, vegetation indices (NDVI, EVI, NBR), classification)

Change Detection Strategies (visual detection, image differencing, Post classification)

Validation (OA, Precision, Recall, F1 Score

What is visual Detection

Manual Delination of change (very repetitive)

good for large and obvious changes

x subjective

x not good for subtle changes at the pixel level

x time and labour intensive

x difficult to scale

What is Image Differencing?

subtracting pixel values, (e.g. NDVi before and after)

(-) vegetation loss

0 stable

(+) vegetation growth

easy to compute, preserves magnitude

x can’t tell what type of change is occuring

x sensitive to seasonal/ phenological changes or atmospheric differences

What is Post Classification Comparison

Turning pixels to graphs

classifying 2 images at different time points

Compare thematic changes

LAND COVER ALREADY MADE

Og land cover → new land cover (very explicit))

x errors are propogated from each classification to final map

x ignores subtle changes within same class.

What is a False Positive?

False Negative?

True Positive?

True Negative?

FP = map detected change but reference material didn’t

FN = no change detected on map but reference material did

TP = both map and reference detected change

TN = both map and reference did not detect change.

What are apparent changes in Change Detection?

Cloud/shadow masking

sensor differences

image misregistration

seasonal/ phenological variation

drought in one year

transient objects (cars)

What values does RS use in Change Detection? Think of NDVI and bands

reflectance change

temperature change

texture change (how smooth pixel is)

structure change (LIDAR, active sensors)

What is Pixel Based Detection?

each pixel is individually analyzed

changes involve changes in spectral values, indicies, class labels

Simple and widely used w/ Landsat and sentinel-2 data

x sensitive to image misregistration and pixel level noise

very popular

What is Object Based Detection. What are it’s advantages?

pxels are grouped into meaningly objects

changes are seen as changes in shape size, spectral values, texture, spatial context.

useful for high reso imagery (pixels are on cm scale)

useful for highly defined features.

What is Change Detection in RS?

identifying and quantifying differences in land surface conditions bt observations at discrete time points.