Vegetation Indices

1. Vegetation Indices MODIS Vegetation Indices Products Accuracy Analysis EVI (Enhanced Vegetation Index) EVI = 2.5 [ρNIR-ρred] / [L+ρNIR+C1ρred-C2ρblue], where L=1 is the canopy background adjustment factor, C1=6.0 and C2=7.5 are the aerosol resistance weights. NDVI (Normalized Difference Vegetation Index) NDVI = [ρNIR-ρred] / [ρNIR+ρred], where ρNIR/red is the measured reflectance in the red/NIR channel Eric F. Vermote1,2, Nazmi Z. Saleous2,3, and Svetlana Y. Kotchenova1 Department of Geography, University of Maryland, USA; 2NASA/GSFC Code 614.5; 3SAIC E-mail: eric@ltdri.org Biophysical Processes State of Vegetation net primary production photosynthesis LAI, fAPAR chlorophyll content relative abundance transpiration percent green cover Analysis of the Performance Radiative Transfer Validation To evaluate the performance of the MODIS Collection 5 algorithms, we analyzed 1 year of Terra data (2003) at 150 AERONET sites (more than 4000 cases). We developed an evaluation approach that allowed us to analyze a one-year long time series in a timely manner and provided us with a quantitative measure of the surface reflectance code improvement. The approach consists in processing subsets of Level 1B data over AERONET sites using an algorithm equivalent to that of the standard surface reflectance and comparing the results to a reference data set. The reference data set is created by atmospherically correcting the TOA reflectance derived from Level 1B subsets using the vector 6S and AERONET measurements (aerosol optical thickness, particle distribution, and water vapor content). For each case in our study, we compute the difference between the reflectance values obtained by the standard code and the reference data set. If the difference is less than the theoretical uncertainty of (0.005+5%), the observation is considered ‘good’. The percentage of ‘good’ observations for each AERONET site is displayed on a map such as the one shown in Fig. 2a or 3a. These maps are available at http://mod09val.ltdri.org/cgi-bin/mod09_c005_public_allsites_onecollection.cgi. Fig. 2a. Comparison of MODIS NDVI and the reference data set for all available AERONET data for 2003. The circles are centered on AERONET sites. The circle colors indicate the percentage of comparisons that fall within the theoretical MODIS one sigma error bar (green > 80%, 65% < yellow < 80%; 55% < magenta < 65%, red < 55%). The circle radii are proportional to the number of observations used in the data comparisons. Globally, 97.1% of the 2003 comparisons fell within the theoretical MODIS one sigma error bar (error bars = +/-(0.02+2%)). • 6S (Second Simulation of a Satellite Signal in the Solar Spectrum) is a basic radiative transfer (RT) code used in the atmospheric correction of MODIS data. The vector version of the 6S code, 6SV1.0B, was released in May 2005, and can be downloaded from http://6S.ltdri.org. • Recent updates(http://rtcodes.ltdri.org): • Future publication of the paper “Validation of a vector version of the 6S radiative transfer code for atmospheric correction of satellite data. Part I: Path radiance,” by S.Y. Kotchenova, E.F. Vermote, R. Matarrese, and F. Klemm, Applied Optics, in press (45, September 10th, 2006); • Preparation of the paper “Validation of a vector version of the 6S radiative transfer code for atmospheric correction of satellite data. Part II: Homogeneous Lambertian and anisotropic surfaces,” by S.Y. Kotchenova and E.F. Vermote, to be submitted to Applied Optics; • Participation of 6S in a joint vector/scalar RT code comparison study; • Release of the updated manual of 6S. Accuracy of Surface Reflectances Accuracy of Atmospheric Correction Accuracy of Vegetation Indices Accuracy of the vector 6S RT code Fig. 2b. Summary of the results for the Alta Floresta site. Each bar corresponds to a date and time where coincident MODIS and AERONET data are available. The size of a bar indicates the percentage of ‘good’ MODIS observations for the given date and time. The graph displays the percentage of ‘good’ NDVI observations for all combined time periods. Fig. 3a. Comparison of MODIS EVI and the reference data set for all available AERONET data for 2003. (See the Fig. 2a captions for further explanation. Clicking on the location of a particular site will provide more detailed results for this site.) Globally, 93.64% of the 2003 comparisons fell within the theoretical MODIS one sigma error bar (error bars = +/ (0.02+2%)). In relation to the previous 6S accuracy issues Reproduced conditions and results of the previous “scalar 6S vs. SHARM” comparison study: Description of SHARM SHARM is a 1-D scalar RT code designed to perform simultaneous computations of monochromatic radiance/ fluxes in the shortwave region for a large set of initial geometric conditions and multiple wavelengths. Conditions - standard continental aerosol (70% of dust, 29% of water- soluble and 1% of soot particles) -  = 0.2 (clear atmosphere) and  =0.8 (hazy) - λ = 750 nm- AZ = {0°; 90°; 180°}, VZA = {0°-79°} - SZA = {0.0°; 10.0°; 23.07°; 45.0°; 58.67°; 75.0°} - 2 types of ground surface: Lambertian with  = 0.25 (Fig. 1a), and the RPV grass model** (Fig. 1b). Source of error In the previous scalar version of 6S the influence of anisotropic surface was incorporated into the RT body of the code by using approximate empirical formulas. Solution To check this source of error, we removed the approximate formulas from 6SV1.0B and incorporated the surface influence directly (Fig. 1c). There is now very good agreement between the 6S and SHARM simulations. The maximum relative error does not exceed 0.32%. * A. Lyapustin, Radiative transfer code SHARM-3D for radiance simulations over a non-Lambertian non-homogeneous surface: intercomparison study, Applied Optics, 41, 5607-5615, 2002. ** H. Rahman, B. Pinty, and M.M. Verstraete, Coupled Surface-Atmosphere Reflectance (CSAR) Model. 2. Semiempirical surface model usable with NOAA advanced very high resolution radiometer data, Journal of Geophysical Research, 98 (D11), 20791-20801, 1993. Results a Fig. 2c. More detailed results for the Alta Floresta (2003197 14:30) site obtained by clicking on the bar corresponding to site id 32 in Figure 2b: A scatter plot of the retrieved NDVI versus the reference data set along with the linear fit results. The blue and green lines indicate the limits of the theoretical uncertainties. In addition to this plot, the Web site displays a table summarizing the AERONET measurements and geometrical conditions, and shows a browse image of the site before and after atmospheric correction. Fig. 3b. The percentage of ‘good’ EVI observations for the Alta Floresta site for all combined time periods. (See the Fig. 2b captions for further explanation.) b Fig. 3c. A scatter plot of the retrieved EVI vs. the reference data set along with the linear fit results. The plot appears on the screen by clicking on the bar corresponding to site id 24 in Figure 3b. (See the Fig. 2c captions for further explanation.) Theoretical error budget Table 1. Overall theoretical accuracy of the atmospheric correction method considering the error source on calibration, ancillary data and aerosol inversion for 3 aerosol optical thick-nesses (0.05: clear, 0.3: avg., 0.5: hazy). The selected sites are Savanna (Skukuza), Forest (Belterra), and Arid (Sevilleta). The uncertainties are considered independent and summed in quadratic. E.F. Vermote and N.Z. Saleous, 2006, Operational atmospheric correction of MODIS visible to middle infrared land surface data in the case of an infinite Lambertian target, Book chapter in “Earth Science Satellite Remote Sensing”, Springer, in press. c Similar maps are also available for all MODIS surface reflectance products (bands 1-7). Conclusions Fig. 1. SHARM vs. 6SV1.0B in scalar mode for the conditions specified on the left. (1) The direct incorporation of surface BRDF into the vector 6S has improved the accuracy of the code (vs. SHARM) by an order of magnitude compared to the previously used approximation. (2) In addition to its utility in refining the uncertainty estimates, the developed performance evaluation approach allows us to test new versions of the surface reflectance code using long time series of data. (3) The error budget needs to be reanalyzed using a better defined climatology of aerosols.