Miura

Continuity Efforts from Ground-up: 

Continuity Efforts from Ground-up T. Miura University of Hawaii, Honolulu H. Yoshioka Aichi Prefectural University, Japan A. Huete University of Arizona, Tucson J. Eidenshink USGS EROS B. Reed USGS Flagstaff Field Center K. Gallo NOAA NESDIS

Global Vegetation Index Time Series: 

Global Vegetation Index Time Series Monitoring ecosystem variability and responses to environmental changes, Spectral vegetation indices (VIs) providing long term observations of vegetation: Intra- and inter-annual changes in vegetation photosynthetic activities Phenological characterization Land cover classification / characterization LAI / fAPAR GPP / NPP Biomass (both green and woody) Tree cover

Multi-sensor Data Sources: 

Multi-sensor Data Sources

VI Long Term Data Record: 

VI Long Term Data Record

Issues in VI Continuity: 

Issues in VI Continuity Sensor/platform Bandpass Spatial resolution Orbital characteristics Sensor degradation Algorithms Atmospheric correction Temporal compositing Spatial aggregation VI formula

Dimensions of VI Continuity: 

Dimensions of VI Continuity

Definition of VI Continuity: 

Definition of VI Continuity VI is continuous if VI values computed from reflectance data produced by two sensors become the same for the same target under identical conditions. Target (The same canopy under identical conditions)

The “Vegetation Isoline” Concept for Continuity Studies: 

The “vegetation isoline” consists of the canopy reflectace points obtained by changing the optical properties of the canopy background materials with a fixed biophysical condition for constant external conditions. An analytical expression inter-relating two vegetation indices (VIs) from two different sensors , v1 and v2, can be derived by applying the vegetation isoline concepts (equations): The “Vegetation Isoline” Concept for Continuity Studies Yoshioka (2004) TGARS / Yoshioka, Miura, & Yamamoto (2005) SPIE

Isoline-based VI Translation:NOAA-14 AVHRR vs. Terra MODIS: 

Isoline-based VI Translation: NOAA-14 AVHRR vs. Terra MODIS Simulation Condition: SZA: 10 ~ 60 deg. AOT: 0.10 ~ 0.78 LAI: 0 ~ 4.0 Crown Cover: 50 ~ 100% Soil (rred): 0.1 ~ 0.25 6S+GeoSail Model (Huemmrich, 2000; Vermote et al., 1997) BOREAS TE-12 (Walter-Shea, UNL) AVHRR and MODIS NDVI Differences (DNDVI) Plotted against MODIS NDVI

VI Continuity Analysis Approach: 

VI Continuity Analysis Approach Theory Development (Multi-sensor VI Translation Equation) Model Simulation Studies Hyperspectral Image Simulation Studies “Real” Satellite Observation Studies Application: Phenological Metrics Direct Comparisons Error, Accuracy, Uncertainty Analyses Empirical investigation Cross-calibration or translation

Cross-sensor VI Relationships: 

Cross-sensor VI Relationships Simulation Condition: SZA: 10 deg. PAI: 0 ~ 8.0 Crown Cover: 0 ~ 100% Soil (rred): 0.1 ~ 0.25 GeoSail Model BOREAS TE-12 Tissue Optical Data Sensors: Terra/Aqua MODIS NOAA-14/16 AVHRR SPOT-4 VEGETATION NPP/NPOESS VIIRS Terra ASTER Boreal Forest

NDVI Relationships:EO-1 Hyperspectal Hyperion: 

NDVI Relationships: EO-1 Hyperspectal Hyperion The NDVI relationships among the sensors are neither linear nor unique and were found to exhibit complex patterns and dependencies on bandpasses. The “green peak” region at around 550 nm and the “red-NIR transitional” region from 680 nm to 780 nm were found to be the key factors in producing the nonlinear patterns. In particular, differences among the sensors in the extents to which their red and/or NIR bandpasses covered these features significantly influenced the trends and the degrees of nonlinearity in the relationships. Miura, Yoshioka, & Huete (2006) RSE Tropical Forest ~ Savanna, Brazil

Sensitivities of NDVI Relationships to Canopy Background Brightness: 

Sensitivities of NDVI Relationships to Canopy Background Brightness Spectral bandpasses NDVI-1: rred = 645 nm, rNIR = 860 nm NDVI-2: rred = 615 nm, rNIR = 860 nm (circle) rred = 642 nm, rNIR = 860 nm (square) rred = 675 nm, rNIR = 860 nm (triangle) Simulation conditions GeoSail + PROSPECT Nair view + 45 deg. solar zenith Uniform LAD LAI : 0.0 – 4.0 Soil (rred): 0.05 – 0.3 The values resulting from the same LAI values, but different soil brightness are connected by soil lines. Yoshioka, Miura, & Yamamoto (2006) SPIE

Sensitivities of NDVI Relationships to Atmospheric Correction Schemes: 

Sensitivities of NDVI Relationships to Atmospheric Correction Schemes 6S+SAIL2 AERONET, TOMS Sun/view geometry from satellites FIFE canopy parameters Konza Prairie, Kansas

Sensitivities of NDVI Relationships to Atmospheric Correction Schemes: 

Sensitivities of NDVI Relationships to Atmospheric Correction Schemes NOAA-14 AVHRR vs. SPOT-4 VEGETATION April 1998 – December 1998 “TOA” vs. Total Cor. Rayleigh/O3/H2O vs. Total Cor.

Sensitivities of NDVI Relationships to Sun/View Geometries: 

Sensitivities of NDVI Relationships to Sun/View Geometries NOAA-16 vs. NOAA-17 AVHRR/3 Konza Prairie, Kansas June – July, 2003 Data Description: Eidenshink (2006) PE&RS

The “h” Functions for Different Vegetation Types (Physiognomies): 

The “h” Functions for Different Vegetation Types (Physiognomies) Terra-MODIS vs. NOAA-14 AVHRR Terra-MODIS vs. VEGETATION

Polynomial Approximation to the Isoline-based Translation Equation: 

Polynomial Approximation to the Isoline-based Translation Equation Data sets used: Boreal Forest Tallgrass Prairie Tropical Forest Factors considered: LAI Canopy background brightness Aerosol loadings Polynomials: 1st – 4th orders SAVI & AOT Liner & non-linear

Polynomial Approximation to the Isoline-based Translation Equation: 

Polynomial Approximation to the Isoline-based Translation Equation Data sets used: Boreal Forest Tallgrass Prairie Tropical Forest Factors considered: LAI Canopy background brightness Aerosol loadings Polynomials: 1st – 4th orders SAVI & AOT Liner & non-linear

Isoline-based VI Translation:Cotton Experimental Data: 

Isoline-based VI Translation: Cotton Experimental Data An initial performance evaluation was conducted with the cotton data, in which two NDVI’s (red & NIR, green & NIR) were translated. Data Source: Huete et al. (1985) RSE Relationship between the red-based and green-based NDVI h values plotted against LAI.

Performance Evaluation: AVHRR, VEGETATION, MODIS: 

Performance Evaluation: AVHRR, VEGETATION, MODIS EOS Validation Core sites The conterminous United States NOAA-14 AVHRR (Daily) April 1, 1998 – March 31, 1999 SPOT-4 VEGETATION (10-days) Apr 1, 1998 – Mar 31, 1999 Jan 1, 2002 – Dec 31, 2002 Terra MODIS (Daily) Jan 1, 2002 – Dec 31, 2002 Translation equations Simple linear regression Nonlinear regression (2nd-order SAVI)

Performance Evaluation: AVHRR, VEGETATION, MODIS: 

Performance Evaluation: AVHRR, VEGETATION, MODIS VEGETATION to MODIS Translation Results AVHRR to VEGETATION Translation Results Bondville ARM/CART

Performance Evaluation: AVHRR, VEGETATION, MODIS: 

Performance Evaluation: AVHRR, VEGETATION, MODIS Isoline-based translation results: Reduction in variability (10 – 50%) Less prone to bias Best results obtained for agricultural areas, while the improvement limited for undisturbed ecosystems

Phenological Metrics:Terra-MODIS vs. NOAA-17 AVHRR: 

Phenological Metrics: Terra-MODIS vs. NOAA-17 AVHRR Comparison of the 2005 Start-Of-Season Metrics (SOS) over the conterminous United States MODIS-derived SOS AVHRR-derived SOS *The pixel-by-pixel analysis indicated the MODIS-derived SOS estimates are ~14 days later than the AVHRR-derived results. Reed (2006) USGS Flagstaff Field Center

Backward Compatibility of the Enhance Vegetation Index (EVI): 

Backward Compatibility of the Enhance Vegetation Index (EVI) Recently, Huete et al. (2006) proposed an EVI without a blue band, “EVI2”. The “vegetation isoline” technique can also be applied to inter-relating two different VI formulas, i.e., NDVI to EVI. Results of the NDVI-to-EVI translation via a non-linear regression approach. The data used here were simulated using the SAIL2 model (Braswell et al. 1996) constrained with in situ measured parameter values for a tropical forest in Hawaii (Suzuki et al, 2006). Every parameter was varied within a range observed in a field randomly to generate the data set.