Beyond Spectral and Spatial data: Exploring other domains of information
description
Transcript of Beyond Spectral and Spatial data: Exploring other domains of information
Beyond Spectral and Spatial data: Exploring other domains of
information
GEOG3010 Remote Sensing
and Image Processing
Lewis
RSU
Multitemporal information
• Background
– The reflectance / scattering properties of earth's surface change over time
Multitemporal information
• Background – May be due to factors such as:
• vegetation growth / senescence cycles
• de/reforestation / fires
• variations in soil moisture
• variation in (size of) water bodies
• built environment changes
• coastal erosion
Multitemporal information
• Background – Changes occur
• at a range of temporal scales
• over a range of spatial scales
Multitemporal information
• satellite EO appropriate to range of dynamic monitoring tasks – repeated coverage – consistent instrumentation – accurate – non-intrusive – variety of spatial and temporal scales
Multitemporal information
• satellite EO appropriate to range of dynamic monitoring tasks – monitoring vegetation dynamics over course of
a year – link to (crop) growth models to provide yield
estimates – distinguish cover types (classification)
dynamics
Anomalies
Issues
• temporal sampling – reconcile requirements of monitoring task with
sensor characteristics and external influences• repeat cycle of sensor
• spatial resolution of sensor
• lifespan of mission / historical data
• cloud cover effects on optical / thermal data
Issues
• discriminating surface changes from external influences on RS data – Viewing and illumination conditions can change
over time • Viewing:
– wide field of view sensors
– pointable sensors
• Illumination: – variations in Sun position
• variations in atmospheric conditions
Issues• cloud cover
Issues
• sensor calibration – degradation over time – variations between instruments
• Coregistration of data – effects of misregistration (practical)
Issues
• Quantity of data – can be large (TB) – preprocessing requirements can be very large– move towards formation of databases of RS-
derived 'products' (EOS, CEO)
Dealing with issues
• Vegetation Indices (VIs)– measured reflectance / radiance sensitive to
variations in vegetation amount– BUT also sensitive to external factors – want contiguous data (clouds) – Typically take VI compositing approach
Use of VIs
• direct: – attempt to find (empirical) relationship to
biophysical parameter (e.g. LAI)
• indirect:– look at timing of vegetation events (phenology)
VI Issues
• VI can still be sensitive to external factors (Esp. BRDF effects)
• no one ideal VI - NDVI used historically
• empirical relationships will vary spatially and temporally
VI Issues
• IDEAL:– Attempt to make VI sensitive to vegetation
amount but not to external factors: • atmospheric variations
• topographic effects
• BRDF effects (view and illumination)
• soil background effects
– SAVI, ARVI etc.
VI Issues
• PRACTICE:– VIs maintain some sensitivity to external
factors – Be wary of variations in satellite calibration etc.
for time series
VI Issues
VI Issues
VI Issues
Examples/Techniques
• multitemporal SAR data for crop classification
– varying growth / senescence between crops used to distinguish crop type
– can attempt to use standard classification algorithms
Examples/Techniques
• multitemporal SAR data for crop classification
– noise issues with SAR (practical)– image segmentation (detect fields) and classify on
field-by-field basis – smooth ('despeckle') data prior to use of pixel-by-
pixel classification
Examples/Techniques
• land cover change detection
• Vegetation Indices eg:– change in VI - infer change in vegetation state – NDVI variation in Mozambique (UN World Food
Programme)
Examples/Techniques
NDVI variation Mozambique
Classification
• Change in area covered by various classes– eg. forest cover to investigate variations in
global / regional Carbon budgets
Forest cover 1973
Forest cover 1985
Forest cover change
Examples/Techniques
• land cover change detection – Methods:
• characterise trajectories to models (phenology)
• analysis of time trajectories of NDVI / thermal data
• Principal Components Analysis
Examples/Techniques
• phenology
NDVI image sequence over Colorado 1990-1996
Examples/Techniques
NDVI time series
Examples/Techniques
Examples/Techniques
Examples/Techniques
Time of greenness onset
Duration of growing season
Examples/Techniques
• land cover change detection – Methods:
• characterise trajectories to models (phenology)
• analysis of time trajectories of NDVI / thermal data
• Principal Components Analysis
Examples/Techniques
• Lambin, E. F. and D. Ehrlich (1996), The surface temperature -- vegetation index space for land cover and land-cover change analysis, International Journal of Remote Sensing 17(3):463-487.
LAI, cover
dryness
Examples/Techniques
• land cover change detection – Methods:
• characterise trajectories to models (phenology)
• analysis of time trajectories of NDVI / thermal data
• Principal Components Analysis
PCA
• Rotation and scaling along orthogonal directions of maximum variance
PC1
PC2
Consider multitemporal NDVI:
Expect high degree of correlation
but also deviations from this
use PCT...
Monthly NDVI - Africa
96.68% of variance in PC1
Loadings very similar for all months …average
Monthly NDVI - Africa
2% of variance in PC2
Dec-March minus
April-Nov
Seasonality - ITCZ movement
PCA
• Information on – state (PC1)– dynamics (seasonality, longer term trends)
Summary
• Basis: dynamics/change
• Methods:• classification … change
• phenology-based description / classifications
• NDVI / thermal data - temporal trajectories
• Principal Components Analysis