modis ndvi time series classification in envi for … · 2012-06-28 · modis ndvi time series...
TRANSCRIPT
ENVI User Group Spring MeetingVIS li 2012VISualize 2012
MODIS NDVI Time Series Classification in ENVI for CropClassification in ENVI for Crop
Mapping and Performance pp gEstimates Tatiana Nawrocki
June 18 - 19, 2012
h d d b l l d l f fThe process discussed below relied on ENVI classification ofMODIS NDVI time series (multi‐band imagery datasetsassembled from single‐band raster datasets). Unlike inassembled from single band raster datasets). Unlike inconventional multi‐spectral imagery, the bands consisted oftemporal layers. ENVI ISODATA classifier treated temporalstacks as multi‐spectral datasets. The classification revealedlocations of crops, which displayed similar seasonaldevelopment curves (“temporal” signatures) Digitaldevelopment curves ( temporal signatures). Digitalnumbers (NDVI values) in each temporal “band” in aseasonal stack provided crop performance metrics atspecific vegetation stages of each classified crop and servedfor comparing their performance between differentseasonsseasons.
Normalized Difference Vegetation Index (NDVI)
8%50%
0 72g
NDVI = (NIR — VIS)(NIR + VIS)
0.72
0.14
(0.40 — 0.30)/(0.40 + 0.30) = 0.14 (0.50 — 0.08)/(0.50 + 0.08) = 0.72
Synthetic “spectral” signatures of a multiband NDVI image stack simulate NDVI dynamics during crop vegetation cycle
( ) ( ) ( ) ( )
y g p g y
MODIS/NDVI Time Series Database: Crop Explorer
Compare seasonal cropseasonal crop conditions by countries andcountries and
regions
Blended performance metrics of different cropsof different crops
Assemble MODIS NDVI temporal stacks representing seasonal conditionsrepresenting seasonal conditions
Band Start Date End Date1 17-Nov 2-Dec2 3-Dec 18-Dec2 3 Dec 18 Dec3 19-Dec 3-Jan4 1-Jan 16-Jan5 17-Jan 1-Feb6 2-Feb 17-Feb
2008 2009 2010 2011
6 2-Feb 17-Feb7 18-Feb 5-Mar8 6-Mar 21-Mar
Classify seasonal NDVI image stacks in ENVI
Non‐supervised classification within
Review band statistics and select crop types for comparative
cropland maskp yp p
seasonal crop performance analysis
ClVI Classes
ND
R i t ti ti f l t d l
Temporal bands
Review statistics for selected classes
Review temporal band statisticsReview temporal band statistics
2007/2008 2008/2009
2009/2010 2010/2011
Compare peak period histograms2007/2008 2008/2009
2009/2010 2010/20112009/2010 2010/2011
Temporal Band 6 – Early FebruaryTemporal Band 7 – Late February‐March
Crop performance varied by locationsCrop performance varied by locations
(winter season)( )
Seasonal crop production could be related to spatial i f ll d l f iratios of well and poorly performing crops
Compare spatial patterns of seasonal crop performanceperformance
Compare crop areas by different p p yperformance classes, 2007 – 2011 Rabi
crop seasons Pakistancrop seasons, Pakistan
Crop performance in 2011/12
Pakistan MODIS NDVI time series analysis showed inferior f d d d f ll f iperformance and reduced areas of well performing crops
in 2011/12 compared to previous season
MODIS NDVI time series geospatialseries geospatial analysis and classificationclassification methods were
explored with theexplored with the goal to identify
locations of majorlocations of major crops in the
European part ofEuropean part of Russian Federation in 2011 seasonin 2011 season
Similar agro‐climate conditionsClassification was more successful
g
within a smaller window of three most important winter grain regions
(Krasnodar, R i l Rostov, and
Stavropol).
Regional mask
Review spectral profiles seasonal crop f f i i f l d i lperformance of winter grain for selected pixels
Image band16‐day NDVI dataset 16‐day NDVI dataset
l b dTemporal bandImage band
Start date End date Image band Start date End date
Band 1 6‐Apr 21‐Apr Band 5 9‐Jun 24‐Jun Band 2 22‐Apr 7‐May Band 6 25‐Jun 10‐Jul B d 3 8 M 23 M B d 7 11 J l 26 J l
Temporal band
Band 3 8‐May 23‐May Band 7 11‐Jul 26‐JulBand 4 24‐May 8‐Jun Band 8 27‐Jul 11‐Aug
Classified crops,
Southern Russia
Locations of major winter grains bywinter grains by performance categories in
Krasnodar, Rostov, and Stavropol regions,
iRussia, 2011
0 80
0.90
0 40
0.50
0.60
0.70
0.80
0.20
0.30
0.40
6‐Apr 21‐Apr 6‐May 21‐May 5‐Jun 20‐Jun 5‐Jul 20‐Jul 4‐Aug
Good Moderate
onficatio
Verif
REGION NAME Reported winter wheat, thousand ha, 2011
Classified winter wheat, thousand ha
Agreement%
Stavropol'skiy kray 1724.2 1213.3 70.4%Rostovskaya oblast' 1974.7 1584.6 80.2%Krasnodarskiy kray 1306.7 1082.7 82.9%Respublika Adygeya 81.6 60.7 74.4%
Classifications of reduced temporal stackspCrop performance curves obtained during a reduced b i i d ll dobservation period allowed identifying well performing
winter crops (class 9), however moderatelyhowever, moderately
performing winter crops (classes 8 & 7) overlaid with
summer crops.summer crops.
REGION NAME Well performing crop area, thousand ha Moderately performing crop area, thousand ha
5-band 9-band 5-band 9-bandStavropol'skiy kray 928.2 854.7 559.8 358.6
Rostovskaya oblast' 325.5 446.4 1402.6 1138.2
Krasnodarskiy kray 1021 6 930 8 758 0 151 9
8‐band 8‐band
Krasnodarskiy kray 1021.6 930.8 758.0 151.9
Conclusions:Conclusions:MODIS NDVI time series classification was suitable for categorizingcrop seasonal performance. These results could potentially become
dditi l f i f ti f f ti i l i ld dan additional source of information for forecasting regional yield andproduction trends.
Limitations:Classifications within large, non‐uniform regions delivered lessdistinct separation between winter grains and other crops.Classifications derived from reduced (5‐band) temporal stacksClassifications derived from reduced (5 band) temporal stacksdelivered suitable results for well performing crops only. Delayedwinter crops and summer crops did not separate.
DisclaimerDisclaimer:The contents contained in this presentation do not necessarilyreflect the work and/or opinions of the USDA and its authorities.