Exercise

Preprocessing

Creating a raster brick and cleaning the MODIS data using the reliability layer

Option 1: detect break at the end of the time series with BFAST Monitor

Now we apply the bfastmonitor function using a trend + harmon model with order 3 for the harmonics (i.e. seasonality modelling): ==Time of break== ==Magnitude of change plot==

Option 2: detecting all breaks in the middle of a time series with BFAST Lite

BFAST monitor用于检测时间序列末尾的第一个中断。如果您需要检测多个中断，则需要使用不同的算法：BFAST 或 BFAST Lite。 让我们对前面步骤中的数据运行函数 bfastlite() 。与默认值 LWZ 相比，将参数 breaks 设置为 BIC 可以更自由地检测中断。

Seasonality monitoring using harmonics

PPT Part1

引言

Why time series analysis?

• understand ==long-term changes== and ==dynamics==
• ==Near real-time== change detection

  Satellite-based time series analysis

  Urban Sprawl Sea level rise Annual Forest change» selective logging

  研究现状(radar 狠狠增长)

  Global dense satellite time series imagery, at 10 – 30m spatial resolution now available openly:

• Optical: Sentinel-2, Landsat,
• Radar: Sentinel-1

  Microsatellites provide high resolution (1 – 5 m) data

• Optical: Planet Labs
• Radar: IceEye, Capella Space

  Content

  Bi-temporal versus time series

  two points vs series

  Bi-temporal Change Detection

  e.g.

• image differencing
• delta classification

  Advantage

• conceptually simple and quick
• easy to interpret

  Disadvantage

• no accurate timing of changes or understanding of dynamics
• inaccurate estimate of change magnitude Bi-temporal versus time series

  (complete)Time Series

  Advantage:

• accurate monitoring of forest change dynamics and timing
• analysing change (e.g. change magnitude) and post-change parameters (e.g. recovery speed, follow up land use)

  Disadvantage:

• conceptually（概念上） more advance
• mainly pixel-based

  Preparing for time series analysis

  Important steps: ==1. Image acquisition»2. Pre-processing»3. Analysis==

  1.Image acquisition

• Trade-off（权衡） between spatial and temporal resolution
• High spatial and temporal resolution E.g. Sentinel 1/2 sensors; 5 daily, 10-20m Planet: daily, 4 m (PlanetScope) or 5-daily, 1 m (SkySat)

  2. Pre-processing(optical satellite data)

  目的Maximizing the signal to noise ratio

  1. Geometric correction
  2. Radiometric correction
  3. Atmospheric correction
  4. Cloud Masking
  5. Compositing 影像合成
  • Derivation of vegetation indices
  • Advantage: extract cloud and atmospheric effects
    1. Derivation of vegetation indices ==Pre-processing advantage with time series»extract cloud and atmospheric effects==

       3. Analysis(Time series methods)

       Break-detection

       BFAST: Breaks for Additive Season and Trend CCDC(Continuous Change Detection and Classification)

       Segmentation

       LandTrendr

       Probabilistic approaches

       Bayts

       Conclusions

• Time series analysis
• Understand long-term changes and dynamics
• Near real-time change detection
• Selection of time series methods depends on application
• Dense & high resolution Sentinel-1 (radar) and -2 (optical) time series available

  PPT Part2

  Disturbance monitoring using BFAST-type algorithms扰动监测

  BFAST family(Breaks for additive seasonal trend)

• BFAST: find ==all breaks== in the time series
• BFAST Lite: faster, but less detailed version
• BFAST Monitor: find one break ==at the end of the time series==
• BFAST01: find the single ==biggest break== and classify its type

  BFAST(Breaks For Additive Seasonal Trend)

  find all breaks in the time series

  Components of a time series:

• Trend
• Seasonality
• Noise 三个加起来是Time series

  BFAST principles

(1).Decomposition 分解

• using stl()
• Season and Trend decomposition using LOESS (locally estimated scatterplot smoothing) into components

  (2).On each component

• Piecewise linear regression(线性相关)
• Optimise to minimise model’s residual sum of squares (RSS)（最小残差和）
• Choose number of breaks based on Bayesian Information Criterion (BIC)（break的次数）»>BIC越小，表示模型在拟合数据上的表现越好。在时间序列分析中，如果考虑断点的个数，可以尝试不同的断点数量，计算每个数量下的BIC值，然后选择BIC值最小的数量作为最优的断点数量。

  Linear regression of seasonality: harmonics

  This gives all breaks in the time series, separately for season and trend

  BFAST Lite

  ==updating historical land cover maps globally==

• Concept: faster, but less detailed version»> new development, like BFAST, but skips (1) and goes straight into (2)
• Advantage
  • significantly faster
  • handles missing values
  • Does not distinguish between breaks

    BFAST Monitor

• Near real-time change monitoring
• The end of time series abnormal or not==(deforest)==
• Temporal perspective»>Using the history to determine what is normal
• Methodology: 3 steps for near real-time monitoring
  • Identify a stable history period
  • Model the stable history period
  • Do new observations in the monitoring period conform with the expected behaviour of the history sample

    BFAST01

    find the single biggest break and classify its type