Satellite data

Transcription

Satellite data
PhenoDETECT – Analyzing phenology from satellite
and proximal remote observations
R. Colombo1, L. Busetto1, F. Fava1,2, M. Migliavacca1,3, M. Rossini1, S. Cogliati1, M.
Meroni1,3 E. Cremonese4, M. Galvagno1,4, C. Siniscalco5, U. Morra di Cella4,
(1)
(2)
(3)
(4)
(5)
Remote Sensing of Environmental Dynamics Laboratory, DISAT, Università degli Studi Milano-Bicocca, roberto.colombo@unimib.it
Desertification Research Group, Università degli Studi di Sassari, Sassari, Italy
European Commission, DG-JRC, Ispra, VA, Italy
ARPA Valle d’Aosta, Aosta, Italy
Università di Torino, Italy
PhenoALP project Final Meeting \ 12-14 October 2011
Outline
•
Main objectives
•
Study areas
•
Regional and local analysis
•
Results
•
•
Phenological maps from satellite data
•
Relationships between phenology and primary topographic attributes
•
Relationships between phenology and climate
•
Proximal sensing at site level
Conclusions
PhenoALP project Final Meeting \ 12-14 October 2011
Objectives
Regional analysis:
•
Development and application of method for
phenological monitoring of Larch forests and highaltitude prairies in the Alps using MODIS 250 m
satellite imageries;
•
To analyze spatial and temporal variability of Alpine
grassland and European larch phenology and to
investigate the relationships with topography and
climatic drivers;
Local analysis:
•
To exploit continuous and long-term (hyper)spectral
measurements
based
on
HyperSpectral
Irradiometer, Skye sensors and Webcam images for
the estimation of gross ecosystem productivity and
phenological cycle of grassland and forest.
PhenoALP project Final Meeting \ 12-14 October 2011
Motivation
Regional analysis:
•
Phenological monitoring is traditionally based on field observations. Field
observations are time and work demanding, and difficult to extrapolate on
large areas;
•
Analysis of Vegetation Indices Time Series acquired from high temporal
resolution satellite sensors (e.g., NOAA-AVHRR, MODIS) allow to map the
key phenological dates on large areas, in a fast, economic and repeatable
way;
Local analysis:
•
To gain new insights into the description of the seasonal canopy
development. Calibration, validation of algorithms for upscaling phenology
and GPP using satellite data.
PhenoALP project Final Meeting \ 12-14 October 2011
Study area
•
Aosta Valley and French Alps
Experimental sites for local analysis
Torgnon (grassland)
Torgnon (larch)
Loriaz (grassland)
Extent of the study area for regional analysis
Champorcher (grassland)
PhenoALP project Final Meeting \ 12-14 October 2011
Regional analysis
•
Larch Forests (between 1330-2200 m) and high altitude prairies (above 1900 m) in
Aosta Valley and French Alps;
•
Identified from «Carta Natura» map (Aosta Valley) and CORINE 2006 map (France)
Distribution of Larch and Prairies in the study areas
PhenoALP project Final Meeting \ 12-14 October 2011
Phenological field surveys for regional analysis
•
Conducted on Larch forests in Aosta Valley to provide ground truth for the satellite
estimates;
•
8 European Larch forest stands located in different zones of the region;
•
Phenological observations were carried out to define the dates of Start (SOS) and
End (EOS) of the growing season
Distribution of field monitoring sites
PhenoALP project Final Meeting \ 12-14 October 2011
Phenological field surveys for regional analysis
• Weekly Phenological surveys in Spring and Autumn
(2005 – 2011)
• 3 plots at different heights within each site; 10 trees
randomly selected within each plot.
PhenoALP project Final Meeting \ 12-14 October 2011
Phenological field surveys for regional analysis
Start Of Season
End Of Season
• Score ranging from 1 to 5 (related to phenological phase) assigned to each tree
• Phenological phase of each plot calculated as the mean score of the ten trees
• Site-level SOS and EOS dates computed as the mean of the dates at which spring
score reached 2 (SP2 – Bud Burst) and autumn score reached 1 (AP3 – Complete
Yellowing), in the three plots.
PhenoALP project Final Meeting \ 12-14 October 2011
Experimental sites: intruments for local analysis
• Torgnon grassland:
(hyper)spectral data, eddy data, meteo data, webcam, skye sensors
• Torgnon Larch:
same as Torgnon grassland
• Champorcher:
webcam
• Loriaz:
webcam
HyperSpectral Irradiometer
Metorological station
Webcam
Eddy Covariance
Skye sensors
Example at the Torgnon grassland site
PhenoALP project Final Meeting \ 12-14 October 2011
Spectral data at the experimental sites
Spectral configuration of HSI instruments and sky sensors

HSI hosts a couple of HR4000 spectrometers (OceanOptics, USA). Continuous measurements
Spectrometer
Spectral
range
(nm)
Sampling
interval
(nm)
FWHM
(nm)
1
350-1050
0.25
1
2
700-800
0.02
0.1

Application
Irrad. measurements, ρ and VIs
computation
Sun-induced Chl fluorescence at O2-A
Skye sensors
Sensor
1
2
Channel
Center wl
(nm)
FWHM
(nm)
1
645
50
2
858.5
35
1
531
5
2
570
5
Application
NDVI computation
(same as NASA-MODIS sensor)
PhenoALP project Final Meeting \ 12-14 October 2011
PRI computation
Satellite data
•
Input Data: MODIS TERRA/AQUA Vegetation
Indexes – 16 days -250 m (MOD13Q1) constrained
view angle maximum value composite
•
Spatial Resolution: 250 m
•
Temporal Resolution: 16-days composites
•
Collection: V005, each pixel contains information
on the time of acquisition
•
Equatorial crossing time – TERRA 10.30; AQUA
13.30;
•
Temporal analysis – TERRA (2000-2010), AQUA
(2003-2009).
100 Km
Example of MODIS NDVI image
(1/03/2003 – 17/03/2003 )
PhenoALP project Final Meeting \ 12-14 October 2011
Satellite data
MODIS: MODerate resolution Imaging Spectrometer
NDVI (Normalized Difference Vegetation Index)
NDVI =
ρ NIR − ρ R
ρ NIR + ρ R
• Remote sensing of phenology is tipically based on the analysis of Vegetation Indices Time
Series (e.g. NDVI, EVI) acquired from high temporal resolution/coarse spatial resolution
satellite sensors (e.g., NOAA-AVHRR, MODIS)
PhenoALP project Final Meeting \ 12-14 October 2011
Pre-processing of satellite data
 Reprojection and resize (From SIN GRID to UTM-WGS84)
 a) Savitzky and Golay iterative smoothing (Chen 2004)
PhenoALP project Final Meeting \ 12-14 October 2011
Pre-processing of satellite data
 Reprojection and resize (From SIN GRID to UTM-WGS84)
 a) Savitzky and Golay iterative smoothing (Chen 2004)
PhenoALP project Final Meeting \ 12-14 October 2011
Pre-processing of satellite data
 Reprojection and resize (From SIN GRID to UTM-WGS84)
 Savitzky and Golay iterative smoothing (Chen 2004)
 b) Computation of “Winter NDVI” value (Beck, 2006) and snow correction
PhenoALP project Final Meeting \ 12-14 October 2011
Processing of satellite data
• Curve Fitting and NDVI metrics computation
 a) Fitting of “Snow corrected” NDVI time series with double logistic functions
PhenoALP project Final Meeting \ 12-14 October 2011
Processing of satellite data
• Curve Fitting and NDVI metrics computation
 Fitting of “Snow corrected” NDVI time series with double logistic functions
 b) Extraction of DOYS corresponding to 10 “characteristic points” of the fitted
curve (NDVI metrics)
PhenoALP project Final Meeting \ 12-14 October 2011
Processing of satellite data
• Identification of the “best” NDVI metrics for SOS and EOS estimation
 Comparison of dates corresponding to the different metrics with field
phenological observations to identify the metrics best suited for detection of SOS
and EOS dates
 Metrics S2 and A4 (Zeroes of the 3rd derivative) are the best suited for detection of
Larch SOS and EOS
S2
A4
Busetto et al., 2010
PhenoALP project Final Meeting \ 12-14 October 2011
Processing of satellite data
• Identification of the “best” NDVI metrics for SOS and EOS estimation
 Comparison of dates corresponding to the different metrics with field
phenological observations to identify the metrics best suited for detection of SOS
and EOS dates
 Metrics S2 and A4 (Zeroes of the 3rd derivative) are the best suited for detection of
Larch SOS and EOS
r = 0.87
MAE = 6.91
RMSE = 7.91
N = 25
r = 0.62
MAE = 3.77
RMSE = 4.88
N = 22
Comparison between observed (field) and MODIS SOS and EOS dates
PhenoALP project Final Meeting \ 12-14 October 2011
Generation of phenological maps
• Val d’Aosta
 Metrics S2 and A4 were therefore used to generate yearly SOS and EOS maps for
Larch and Prairies in the two study areas. Yearly maps of the anomalies with
respect to the 2000-2010 mean were also produced.
Larch Yearly SOS maps – Aosta Valley
Larch Yearly SOS Anomaly maps – Aosta Valley
PhenoALP project Final Meeting \ 12-14 October 2011
Generation of phenological maps
• French Alps
Delay
Advance
Prairies Yearly SOS maps – France
Prairies Yearly SOS Anomaly maps – France
PhenoALP project Final Meeting \ 12-14 October 2011
Spatial and temporal variability of phenology
• Start of the growing season
France
Prairies
Larch
Aosta Valley
•
Analysis of MODIS phenological
maps
highlights
large
interannual variations of SOS
dates
•
Variations of more than 20 days
in mean SOS dates were observed
in the different years
•
No significant temporal trend
was observed
Mean SOS dates (± 1 st.dev.)
PhenoALP project Final Meeting \ 12-14 October 2011
Spatial and temporal variability of phenology
• End of the growing season
France
Prairies
Larch
Aosta Valley
lower
•
EOS
dates
show
interannual variations
•
Variations of less than 10 days
in mean EOS dates were observed
in the different years
Mean EOS dates (± 1 st.dev.)
PhenoALP project Final Meeting \ 12-14 October 2011
Spatial and temporal variability of phenology
• Relationships between France and Aosta Valley interannual SOS anomalies
•
Interannual variations in the two
study areas are strongly related
PhenoALP project Final Meeting \ 12-14 October 2011
Spatial and temporal variability of phenology
• Altitude effects on SOS
SOS = -0.045*H + 57.5
•
SOS dates are strongly influenced
by height (Mean delay of about 4.5 days
per 100 m increase, in both Larch and
Prairies)
•
The relationships change in the
different years! (An year -e.g 2001may be warmer at the beginning of Spring
(advance at lower elevation), but colder
afterwards (delay at higher elevation)
Colombo et.al 2009
PhenoALP project Final Meeting \ 12-14 October 2011
Spatial and temporal variability of phenology
• Altitude effects on EOS
EOS = -0.02*H + 335.6
•
Effects of height on EOS dates are
milder (Mean anticipation of about 2 days
per 100 m increase in height).
•
Influence of photoperiod
PhenoALP project Final Meeting \ 12-14 October 2011
Spatial and temporal variability of phenology
• Aspect effects on SOS
Exposure variability of larch and grassland
PhenoALP project Final Meeting \ 12-14 October 2011
Linking phenology with climate
• Relationships between phenology and air temperature variability
•
Mean yearly anomalies of SOS and
EOS were compared with the
mean regional anomalies of air
temperatures derived from data
recorded at 16 meteorological
stations in the Aosta Valley
•
Anomalies
of
monthly,
bimonthly and tri-monthly mean
air temperatures with respect to
the
2000-2010
mean
were
considered
Location of Meteorological Stations
PhenoALP project Final Meeting \ 12-14 October 2011
Linking phenology with climate
• Relationships between phenology and air temperature variability
Larch
Aosta Valley
France
•
Strong
linear
relationships
between
phenological
and
meteorological anomalies
•
Larch and prairies posticipate
SOS of about one week for
each degree of increase in
spring temperature (For Larch the
Prairies
considered temperatures are that of the
march-may period, while for prairies they
are that of the april-june period)
PhenoALP project Final Meeting \ 12-14 October 2011
Linking phenology with climate
• Relationships between phenology and air temperature variability
Aosta Valley
France
EOS variations are milder:
about 2 days for each degree of
September/October
mean
temperature
Prairies
Larch
•
PhenoALP project Final Meeting \ 12-14 October 2011
Spectral data at the experimental sites
Proximal sensing instrument:
HyperSpectral Irradiometer
(HSI)
PhenoALP project Final Meeting \ 12-14 October 2011
Spectral data at the experimental sites
Example
of HSI NDVI
time
series
(2009)
10 June
130 DAYS
PhenoALP project Final Meeting \ 12-14 October 2011
18 October
Spectral data at the experimental sites
2-band sensors (Skye Instrum.)
NDVI and PRI sensors
Installation in 2011 at the larch and
grassland sites
Sensor
Channel
Center wl (nm)
Bandwidth (nm)
Satellite
1
645
50
MODIS
2
858.5
35
MODIS
1
531
5
2
570
5
NDVI
PRI
PhenoALP project Final Meeting \ 12-14 October 2011
Spectral data at the experimental sites
Example: NDVI at the grassland site
165 DAYS
15 April
27 September
SNOW
PhenoALP project Final Meeting \ 12-14 October 2011
Phenology from webcam
Model CC640
Campbell
Scientific and
Nikon D5000
PhenoALP project Final Meeting \ 12-14 October 2011
Phenology from webcam
Greeness Index, GI
GI = G/(R+G+B)
Example at the Torgnon grassland site
PhenoALP project Final Meeting \ 12-14 October 2011
Phenology from webcam
EOS Canon 100D Camera
– ROIs distance > 400m
L_ROI
C_ROI
R_ROI
Example at
the Loriaz
site
GI and GEI indexes computed on the 3 ROIs (May – Sep 2011)
PhenoALP project Final Meeting \ 12-14 October 2011
Conclusions and perspectives
o MODIS 250 m NDVI time series can be effectively used to monitor grassland and
larch phenology in mountain regions;
o The start and end of the growing season can be estimated with good accuracy
from the roots of the third derivative of logistic curves fitted to MODIS TERRA NDVI
time series (RMSE of about 8 and 5 days);
o The interannual variability of phenology, as seen from MODIS data, is closely
linked to regional climatic variability
o The consistency of the results obtained in two distinct geographic regions is
promising for the extension of the methodology at the European Alps scale.
o The unattended optical system succeeded in collecting reliable time series of
hyperspectral optical properties, including spectral vegetation indices and
fluorescence
o Webcam digital camera imagery are a promising tool for monitoring the seasonal
development and interannual variability of the canopy greenness (i,.e. phenology)
PhenoALP project Final Meeting \ 12-14 October 2011
Thank you!