Adv Remote Sensing - Lec 14 Time Series Analysis

What is a Time Series Analysis. How is it different from Change Detection?

Time series analyses in RS use many temporal data periods to detect change.

When change occured

Temporary or persistent?

Did multiple events occur?

Did veg recover?

VS change detection only answers question of whether a change occured, what the mag was, and where the change occured.

What are the 4 components of TIme Series Analysis?

Trend

Periodic Fluctuations

Abrupt breakpoints

Residual/Noise

What RS variables are tracked for Time Series?

NDVI (greeness and vegtation condition

NBR: forest fire disturbance

EVI: less sensitive version of NDVI

SWIR bands: moisture and disturbance sensitivity

Canopy height: structural change from LiDAR

What are appropriate scopes for time series?

pizel

object

stand

region

What effect does phenology have on Time series analysis in RS?

informs disturbance interpretation (i.e. healthy normal change that is good for productivity i.e. a predictable change)

What does TIme series repsect about a forest?

It’s dynamism. aka growth, harvest, drought disease, partial distubances, small gap disturbances that wouldn’t be reflected in change detection.

What is the reso and data date availability for Landsat?

30 m optics

15m panchromatic

every 16 days

1972 - present (longest running database

What is the reso and data date availability for Sentinel - 2A/B

10m - VIS - NIR

20m - SWIR - red/edge

60m - atmospheric bands

every 5 days (A+B)

2015-present

What is the reso and data date availability for Sentinel - 2A/B

3-5m

daily (global)

2009-present

Is pre-processing neccesary for Time Series? If so, what?

Cloud shadow masking

atmospheric radiometric correction

geometric co-registration

sensorr harmonisation (if multiple sensors are used)

consistent seasonal window

suitable index + var selection

What are the 3 types of Analysis Approaches and examples?

Regression / Harmonic Modelling - CCDC (Continuous Change Detection and Classification)

Trajectory Segmentation - LandTrendr ( Landsat-based detection of trends in disturbance)

Decomposition and Structural Break Detection - BFAST (Breaks for Additive Seasonal and Trends

What is Regression/Harmonic Modelling? What is it good for?

Fitting a model to the data to time observations

Harmonic regression can capture seasonality/ phenology

Changes are detected via persistent deviations from the model

GOOD: continuous forest modelling

Near real-time disturbance alerts

On GEE

What is Trajectory Segmentation? What is it good for?

Dividing Temporal trajectory into meaningful segments

Done through breakpoints, slopes, and changes in mag

Data extracted: Year of disturbance, mag, duration, rate of change, recovery

GOOD: abrupt disturbance

recovery

annual plantation harvests

GEE

What is Decomposition and Structural Break Detection? What is it good for?

Seperates data into seasonal, trend, and remainder components

GOOD: phenology, seasonality, drought/ stress, defoilation, long term veg condition

R