Remote Sensing Phenology. Bradley Reed Principal Scientist USGS National Center for Earth Resources Observation and Science Sioux Falls, SD

Transcription

1 Remote Sensing Phenology Bradley Reed Principal Scientist USGS National Center for Earth Resources Observation and Science Sioux Falls, SD

2 Remote Sensing Phenology Potential to provide wall-to-wall phenology estimates Potential to provide information on medium-term trends in phenology Potential to provide improved phenology estimates to global models (climate, ecosystem, etc.)

3 Remote Sensing Phenology Background High spatial resolution sensors Landsat 1972-present, 16-day repeat cycle, 3-25-m resolution SPOT 1986-present, 1m Pan, 2m multispectral IKONOS 1999-present, high resolution; 1m Pan, 4m multispectral, local coverage

4 Sensors for Phenological Studies Hi-resolution sensors history Large Area Crop Inventory Experiment (LACIE) AgRISTARS temporal profiles crop calendars Limited by temporal resolution of sensors

5 Remote Sensing Phenology Background High temporal resolution sensors AVHRR 1981-present; (8-km) global coverage 1989-present; (1-km) conterminous US SPOT Vegetation 1998-present; 1-km resolution Envisat MERIS 22; 3m resolution MODIS 2-present; 25m, 5m, 1-km resolution

6 Vegetation Indices for R.S. Phenology Normalized Difference Vegetation Index (Deering and others 1976) NDVI = (NIR-Red)/(NIR+Red) Soil Adjusted Vegetation Index (Huete 1988) SAVI = (1+L)(ρ nir -ρ red )/(ρ nir +ρ red +L) ρ reflectances L adjustment factor for red and NIR extinction through canopy

7 NDVI Vegetation Indices Time-tested Saturates at high values Coupled to red band reflectance, photosynthetic capacity (fpar, fractional green cover) SAVI (EVI) Coupled to infrared band reflectance; structural canopy parameters (LAI, biomass) More stable, higher dynamic range at high end, but less dynamic range at low end

8 Time-series VI measurements

9 Atmospheric and Sensor noise Cloud contamination throughout composite period sub-pixel clouds Illumination angle and viewing geometry Atmospheric aerosols Water vapor, haze, other contaminants Sometimes unreliable calibration All of the above usually reduce NDVI values

10 Example contaminated pixel Time periods affected by Atmospheric contamination

11 Noise reduction methods Maximum value compositing reduces noise, but still affected by persistent clouds/haze BISE (Viovy and others 1992) FASIR (Sellers and others 1994) Weighted least-squares regression (Swets and others 1998) Other temporal smoothers polynomial; FFT; compound mean/median

12 Example smoothed NDVI pixel Critical to retain temporal nuances

13 Identifying start of season (SOS) Key to seasonal characterization other seasonal metrics depend on SOS looking for a trend shift toward high values Methods for identifying SOS Thresholds Inflection points Curve derivation

14 SOS Threshold Method Pre-defined threshold (Lloyd 199) e.g., NDVI =.99 Half-maximum (White and others 1997) mid-point between minimum and maximum NDVI 1% amplitude (Jönsson and Eklundh 22) NDVI Time Pixel value Half-max Threshold 1%

15 SOS Inflection Point Method Inflection point Badhwar (1984).6 Pixel value Inflection pt. Time derivative transition Moulin and others (1997) NDVI.4.2 Time Maximum curvature Zhang and others (21)

16 SOS Curve Derived Method Delayed Moving Average Reed and others (1994).6 Pixel value DMA Largest Increase Time of Largest Increase NDVI.4.2 Kaduk and Heimann (1996) Time

17 SOS Comparisons.6 Pixel value DMA Largest Increase Inflection Point Threshold Half maximum NDVI.4.2 Time

18 Average SOS deciduous forest Average SOS evergreen forest Average SOS mixed forest Day of year Year Inf l. Pt. Gr. Inc. DMA Day of year Year Inf l. Pt. Gr. Inc. DMA Day of year Year Inf l. Pt. Gr. Inc. DMA Average SOS shrubland Average SOS grasslands Average SOS pasture Day of year Inf l. Pt. Gr. Inc. DMA Day of year Inf l. Pt. Gr. Inc. DMA Day of year Infl. Pt. Gr. Inc. DMA Year Year Year Average SOS row crops Average SOS small grains Average SOS All land cover types Day of year Infl. Pt. Gr. Inc. DMA Day of year Infl. Pt. Gr. Inc. DMA Day of year Infl. Pt. Gr. Inc. DMA Year Year Year

19 What is SOS Measuring? DMA first sustained flush of greenness? Half-max primary leaf expansion? Greatest Increase early season growth peak (perceived spring)? Inflection pt. environmental conditions preceding first flush? what biophysical phenomena should be represented? Application specific.

20

21 GPP avpg NDVI GPP avpg NDVI GPP avpg NDVI GPP avpg NDVI GPP avpg NDVI Woodward, OK Mandan, ND Days offset n = 13 x = 2.23 std = = Satellite SOS Satellite SOS vs. GPP estimates (USDA-Agriflux towers)

22 Additional metrics can be derived from the annual VI cycle Seasonal integrated NDVI

23 Annual summaries of the metrics can be created to assess interannual trends

24 Regions with significant trends in annual greenness: Is the seasonally integrated greenness from increasing or decreasing? 7 6 Seasonal Greenness Seasonal Integrated Greenness

25 Regions with Significant Trends in Annual Greenness: Is the slope (b) of best-fit line significantly different from? Seasonal Greenness b = 2.12 standard error (s) =.46 t-test: t = b/s = 4.6 t-distribution at.5 level of significance and df = 11 = 2.21, 4.6 > 2.21, therefore is significant

26 Earlier SOS Later SOS SOST

27 Earlier EOS Later EOS EOST

28 Trends in Duration of growing season Shorter Duration Longer Duration

29 Decreasing Greenness Increasing Greenness Trends in Total NDVI

30 Analysis of Trends Driving Forces Fire recovery Land use change Land use practice Biological succession Short and long-term climate change Yellowstone National Park

31 Issues in Satellite Phenology non-vegetation related environmental conditions Snow Soil moisture Plant litter Atmospheric perturbations

32 Issues in Satellite Phenology unusual pixels NDVI Shrubland NDVI Double-crop Year 1 Year 2 Year 3 Year 41 Poorly defined seasons Year 1 Year 2 Year 3 Year 4 Bimodal growing seasons 11 Evergreen Evergreen NDVI.4.2 NDVI Year 1 Year 2 Year 3 Year Year 1 Year 2 Year 3 Year 4 11 Evergreen systems with differing seasonality

33 Issues in Satellite Phenology- heterogeneity 1km 32m resolution Satellite sensors integrate the constituents of each pixel, therefore we may not be estimating the phenology of any single plant type, but rather the sum of the pixel constituents 1km 1km 16m resolution 8m resolution

34 Issues in Satellite Phenology- field verification Traditional field measures are plant/plant type specific Pixel heterogeneity makes scaling up from plant specific data difficult Remote sensing optimized approach needs to be considered

35 Conclusions Remote Sensing can generate consistent, objective estimates of phenology start, end, peak, duration of growing season Variety of approaches are available which may be measuring fundamentally different phenomena stages of plant growth, environmental conditions preceding growing season, etc.

36 Conclusions Improvements/understanding of estimates are needed What factors are influencing VI signals and resulting phenology estimates? Field validation is difficult, but critical Users beware Carefully consider which approach is most appropriate for particular applications