New Measurements of Chlorophyll Fluorescence with Gome-2 and Comparisons with the Seasonal Cycle of Gpp from Flux Towers

Mapping of terrestrial chlorophyll fluorescence from space has shown potential for providing global measurements related to gross primary productivity (GPP). In particular, space-based fluorescence may provide information on the length of the carbon uptake period. Here, we show an example of the ability of global satellite fluorescence retrievals to track seasonal cycle of photosynthesis as estimated from tower gas exchange measurements. The satellite fluorescence retrievals are obtained using new observations near the 740 nm emission feature from the Global Ozone Monitoring Experiment 2 (GOME-2) instrument offering the highest temporal and spatial resolution of available global measurements. Because GOME-2 has a large ground footprint (~40km by 80km) as compared with that of the flux towers and the GOME-2 data require averaging to reduce random errors, we additionally compare with seasonal cycles of upscaled GPP estimated from a machine learning approach averaged over the same temporal and spatial domain as the satellite data surrounding the tower locations. We also examine the seasonality of absorbed photosynthetically-active radiation (APAR) estimated from satellite reflectance measurements. Finally, to assess whether global vegetation models may benefit from the satellite fluorescence retrievals through validation or additional constraints, we examine seasonal cycle of GPP at the tower site as produced from an ensemble of vegetation models. The models, particularly those driven by data, tend to overestimate the length of the photosynthetically-active period for this site shown. Satellite fluorescence measurements therefore show potential for improving the seasonal dependence of photosynthesis simulated by global models at similar spatial scales. 1. GOME-2 GLOBAL FLUORESCENCE RETRIEVALS GOME-2 is a grating spectrometer that measures backscattered sunlight in a nadir-viewing geometry at wavelengths 270-800 nm in four separate channels [1]. Here, we use data from the MetOp A platform launched on 19 October 2006 into a sun-synchronous orbit with equator crossing time near 09:30. We use radiances from channel 4 at a spectral resolution of ~0.5 nm with a signal-to-noise ratio of ~1000. The nominal ground pixel lengths at nadir are ~40km and 80km in the alongand across-track directions, respectively. The default swath width is 1920km. This enables global coverage of the Earth's surface within about 1.5 days. The actual coverage for fluorescence retrievals is reduced because data in heavily clouded overcast conditions are removed. We derive GOME-2 Solar Induced Fluorescence (SIF) primarily from the filling-in of solar Fraunhofer lines in the vicinity of the 740 nm far-red fluorescence emission peak as in [2] The basis for the retrieval relies on the fact that an additive signal such as fluorescence produces a filling-in of a solar line when an Earth backscattered spectrum is normalized with respect to the solar spectrum or a reference spectrum that does not 5th INTERNATIONAL WORKSHOP ON REMOTE SENSING OF VEGETATION FLUORESCENCE , 22-24 APRIL 2014, PARIS (FRANCE) 2 contain fluorescence. The retrieval uses a principal component analysis approach with a simplified radiative transfer model to disentangle the spectral signatures of three basic components: atmospheric absorption, surface reflectance, and fluorescence emission as illustrated in Fig. 1. GOME-2 SIF is given in radiance units (mW/m/nm/sr) and varies somewhat with viewing geometry. Figure 1. Schematic of simplified radiative transfer characterized by a Lambertian surface and no atmospheric scattering, where the two-way (sun to satellite) transmittance (τ2) is modeled according to the geometrical air-mass factor. Fluorescence emission is indicated with the dashed red line. The retrieval algorithm performed well when tested with simulated data and was applied to radiances from GOME-2 [2]. Fluorescence from GOME-2 compared well in terms of spatial and temporal variability with that from GOSAT [3] obtained with a more simple retrieval approach as shown in Figure 2. The GOME-2 fluorescence data set provides substantially improved sampling and precision (and therefore spatial resolution) as compared with GOSAT. 0.0 0.5 0.9 1.4 1.8 Fs (mW/m 2/sr/nm) 0.0 0.5 0.9 1.4 1.8 Fs (mW/m 2/sr/nm) Figure 2. SIF derived from GOSAT [3] (top) and GOME-2 (bottom) [2] for July 2009 at 758 nm. We have made some minor changes to the approach of [2] that bring the GOME-2 results into slightly better agreement with GOSAT and reduce small biases where no fluorescence is expected. We use a spectral fitting window between 715 and 758 nm and a single set of principal components (PCs), derived from cloudy data over ocean, to estimate the spectral structure of atmospheric absorption; absorption is dominated by water vapor over this spectral range. The oxygen-A band is not included in the fitting window. The principal components are derived from radiance spectra here as opposed to the logarithm of the radiance spectra as in [2]. We also attempted to correct for drift in the absolute instrument calibration by fitting a degradation factor to the GOME-2 solar spectra. 2. SEASONAL VARIATIONS IN GPP, MODISBASED APAR, AND GOME-2 SIF Here, we assess the ability of GOME-2 SIF to track the seasonality of photosynthesis as delineated by towerbased [4] and upscaled GPP estimates from the Max Planck Institute for Biogeochemistry (MPI-BGC) machine learning algorithm (Model Tree Ensembles, MTE) [5,6]. Figure 3 shows average seasonal cycles of GPP (normalized to unity at the maxima) from towers along with the upscaled GPP from MPI-BGC, GOME-2 SIF, and MODIS-based estimates of APAR [7], for the agricultural Nebraska site US-Ne1. All GOME-2 pixels with centers within 1 latitude and longitude of the flux tower site are averaged here, and similarly all MPI-BGC gridboxes within 1 latitude and longitude of the site are averaged. The GOME-2 SIF and MPI-BGC GPP data are averaged over the same years (2007-2011). Therefore, we expect better agreement of GOME-2 SIF with MPI-BGC GPP as compared with tower-based GPP due to the closer collocation in both time and space of GOME-2 SIF with MPI-BGC GPP. The tower GPP data indicate a relatively short growing season for US-Ne1 (Fig. 3, top) composed of continuous maize crops. This short duration of photosynthesis is similarly shown by both GOME-2 SIF and the upscaled MPI-BGC GPP estimates. The upscaled GPP and GOME-2 SIF display a slightly wider mean growing season as compared with the tower-based GPP. This may be expected as GOME-2 SIF and the upscaled GPP estimates are analyzed at similar and larger spatial scales as compared with the tower measurements. The larger footprints of GOME-2 SIF and MPI-BGC GPP likely contain contributions from different crops such as soybeans that have somewhat lengthier growing seasons and vegetation types other than croplands; this may explain the more gradual increases(decreases) in spring(autumn). A significantly earlier springtime rise and later autumn decline is seen in MODIS-based APAR as compared 5th INTERNATIONAL WORKSHOP ON REMOTE SENSING OF VEGETATION FLUORESCENCE , 22-24 APRIL 2014, PARIS (FRANCE) 3 with GOME-2 SIF and the GPP estimates for US-Ne1 and similarly for other agricultural sites (not shown). The later decline in autumn may be expected as APAR does not contain information about LUE and therefore may remain high in the presence of green vegetation that is not engaging in photosynthesis. However, the almost symmetrical early rise in spring is more difficult to explain as green vegetation in spring should be photosynthetically-active. One explanation is that the reflectances used to generate MODIS-based (MOD15) fPAR are sensitive to pigments in vegetation and soil other than chlorophyll. The MOD15 estimate of APAR applies to the whole vegetation canopy, both photosynthetic and nonphotosynthetic components, and therefore may not provide consistent relationships to photosynthetic processes at the ecosystem scale.