Sunlight mediated seasonality in canopy structure and photosynthetic activity of Amazonian rainforests

Resolving the debate surrounding the nature and controls of seasonal variation in the structure and metabolism of Amazonian rainforests is critical to understanding their response to climate change. In situ studies have observed higher photosynthetic and evapotranspiration rates, increased litterfall and leaf flushing during the Sunlight-rich dry season. Satellite data also indicated higher greenness level, a proven surrogate of photosynthetic carbon fixation, and leaf area during the dry season relative to the wet season. Some recent reports suggest that rainforests display no seasonal variations and the previous results were satellite measurement artefacts. Therefore, here we re-examine several years of data from three sensors on two satellites under a range of sun positions and satellite measurement geometries and document robust evidence for a seasonal cycle in structure and greenness of wet equatorial Amazonian rainforests. This seasonal cycle is concordant with independent observations of solar radiation. We attribute alternative conclusions to an incomplete study of the seasonal cycle, i.e. the dry season only, and to prognostications based on a biased radiative transfer model. Consequently, evidence of dry season greening in geometry corrected satellite data was ignored and the absence of evidence for seasonal variation in lidar data due to noisy and saturated signals was misinterpreted as evidence of the absence of changes during the dry season. Our results, grounded in the physics of radiative transfer, buttress previous reports of dry season increases in leaf flushing, litterfall, photosynthesis and evapotranspiration in well-hydrated Amazonian rainforests.


Introduction
Understanding the seasonal variation in functioning of rainforests and its controls are requisite for understanding how rainforests will respond to climate change. In situ studies report counter-intuitive seasonal variation in wet equatorial Amazonian rainforests -higher photosynthetic and evapotranspiration rates and increased litterfall and leaf flushing during the Sunlight-rich dry season (Saleska et al 2003, da Rocha et al 2004, Goulden et al 2004, Rice et al 2004, Hasler and Avissar 2007, Hutyra et al 2007, Negrón Juárez et al 2009, Costa et al 2010, Jones et al 2014. Water limitation during the dry season is alleviated in these forests through deep roots and hydraulic redistribution (Nepstad et al 1994, Oliveira et al 2005. Satellite data, which cover a large area and span a long time period, support findings of in situ studies-higher radiometric greenness level and green leaf area during the dry season compared to the wet season (Xiao et al 2005, Huete et al 2006, Myneni et al 2007, Samanta et al 2012, Hilker et al 2014, Jones et al 2014, Maeda et al 2014. This convergent view of seasonality, parsed from several studies, shows how sunlight interacts with adaptive mechanisms to result in higher rates of leaf flushing, litterfall, photosynthesis and evapotranspiration in tropical forests if water limitation is absent (Wright and Van Schaik 1994, Restrepo-Coupe et al 2013, Borchert et al 2015, Guan et al 2015. This community-consensual view was questioned in recent studies (Galvão et al 2011, Morton et al 2014). The studies claim that the dry season greening inferred from passive remote sensing data resulted from an artificial increase in forest canopy reflectance at near-infrared (NIR) wavelengths caused by variations in sun-satellite sensor geometry. Their analyses of satellite-borne lidar data suggested that these forests exhibited no seasonal variations in canopy structure or leaf area. Relying on model simulations to guide and imbue a physical meaning to the satellite data analysis, the studies conclude that Amazon rainforests maintain consistent structure and greenness during the dry season.
These contradictory results justify a re-examination of the same satellite data with the goal of assessing seasonality in wet equatorial Amazonian rainforests. In addition to data from NASA's Moderate Resolution Imaging Spectroradiometer (MODIS) on the Terra platform and the Geoscience Laser Altimeter System (GLAS) instrument onboard the Ice, Cloud and land Elevation Satellite (ICESat) used in (Morton et al 2014), we also include data from the MODIS instrument on Aqua and Multiangle Imaging Spectroradiometer (MISR) on the Terra satellite. The MISR sensor views the Earth's surface with nine cameras simultaneously, as opposed to the two MODIS sensors, which are capable of only one view each. This feature enables the rigorous use of the theory of radiative transfer in vegetation canopies-the fundamental theory that explains from first principles the mechanisms underlying the signals generated by the canopy and measured by a remote sensor (Knyazikhin et al 2005).
This study is focused on terra firme rainforests in central Amazonia that are relatively undisturbed by human activities (supplementary data and methods section 1, figure S1). The period June to May is treated as one seasonal cycle as per convention (Huete et al 2006, Morton et al 2014). It consists of a short dry season, June to October, and a long wet season thereafter (supplementary data and methods section 1). The following analysis of satellite borne sensor data addresses the question at the center of current debate -did previous studies (Xiao et

Data and methods
A detailed description of methods and data used is given in the supplementary information available at stacks.iop.org/ERL/10/064014/mmedia. A brief summary is provided here. The study region and the various data analysed in this study are detailed in the supplementary data and methods section 1−2. The sun-sensor geometry relevant to the discussion in this article is presented in the supplementary data and methods section 3. The theory of remote measurements and evaluation of NIR reflectance angular signatures (figure 3) and their interpretation is described in the supplementary data and methods section 4. A critical look at Morton et al 2014 analyses of MODIS and GLAS data is presented in the supplementary discussion. Abbreviations and symbols are listed in supplementary table S5.

Leaf area index seasonality
The seasonal cycle of green leaf area inferred from satellite data (figure 1(a)) exhibits rising values during the dry season (June to October), high values during the early part of the wet season (November to February) and decreasing values thereafter (March to May). This seasonal variation of about 20% is imposed on a base value of Leaf Area Index (LAI, one-sided green leaf area per unit ground area) of about 5.75, is greater than the uncertainty of the LAI product (0.66 LAI, Yang et al 2006) and is observed in nearly 70% of the rainforests in the study domain (figure S4(a)); the rest lacked valid data. Is this seasonal variation real or a misinterpretation of changes in satellite-sensor measurements caused by seasonal changes in sun position in the sky and the manner in which the sensor measures reflected radiation ('sun-sensor geometry')?
The answer requires an understanding of how this geometry changes during the seasonal cycle, which is described in the supplementary data and methods section 3.
The seasonal cycle of leaf area in figure 1(a) cannot be an artefact of seasonal changes in sun-sensor geometry because the algorithm with which leaf area is derived explicitly accounts for geometry changes, i.e. the algorithm is capable of differentiating between changes in measurements caused by leaf area changes and those caused by geometry changes (Knyazikhin et al 1999, Knyazikhin et al 1998. This is also evident from the fact that the seasonal cycle of leaf area does not track the seasonal course of either the Sun position in the sky ( figure 1(b)) or the MODIS sensor sampling (figures 1(c) and (d)). Instead, it tracks independently obtained observations of seasonal variation in sunlight ( figure 1(a)). This behavior is consistent with the idea that sunlight acts as a proximate cue for leaf production in moist tropical forests if water limitation is absent (Wright and Van Schaik 1994, Borchert et al 2015, Guan et al 2015. Thus, relatively high sunlight levels from absence of clouds during the dry season cause leaf area to increase, which in turn generates higher rates of photosynthesis (Saleska et

Evidence for seasonality after sun-sensor geometry correction
The Enhanced Vegetation Index (EVI) is a proven proxy for the potential photosynthetic carbon fixation by vegetation (Xiao et al 2005, Huete et al 2006, Brando et al 2010. It is calculated from satellite-sensor measurements of reflected solar radiation at three different wavelength bands. These measurements depend on sun-sensor geometry, but this dependency can be eliminated by expressing the measurements in a fixed geometry (Morton et al 2014, Lyapustin et al 2012. The EVI calculated from MODIS sensor measurements in a fixed geometry, i.e. nadir viewing direction and 45°solar zenith angle, shows a distinct wet season decrease (figure 2(a)) and dry season increase ( figure 2(b)). These changes are greater than a highly conservative estimate of the precision in 43% of the pixels during the wet season and 31% of the pixels in the dry season. Here, the precision is estimated as the spatial standard deviation of the EVI data in the study domain. Analogous to EVI, pixel level estimates of green leaf area show a strong decrease in the wet season and increase during the dry season. The wet season decrease (figure 2(a)) suggests net leaf abscission, i.e. more older leaves dropped than those newly flushed, and the dry season increase indicates net leaf flushing ( figure 2(b)), resulting in a sunlight mediated phenological behavior (Myneni et al 2007). The fact that both EVI and LAI show congruent changes during the seasonal cycle even though the Sun-sensor geometry effect is removed from measurements in different ways (Knyazikhin et al 1999, Knyazikhin et al 1998, Lyapustin et al 2012, Hilker et al 2014, Maeda et al 2014 is particularly noteworthy.

Evidence for seasonality from multiple sensors and geometries
Now we turn to satellite-sensor measurements of reflected solar radiation at the NIR wavelength band, which are at the heart of the controversy. These measurements are usually expressed as normalized quantities called reflectances (supplementary data and methods section 4.1−4.2). The geometric structure and radiation scattering properties of the rainforest canopy determine the magnitude and angular distribution of reflected radiation. The angular signatures of reflectance are therefore unique and rich sources of diagnostic information about rainforest canopies (Diner et al 1999). We first examine NIR angular signatures from the late dry season (October 15 to 30) and the middle part of the wet season (March 5 to 20). The Solar Zenith Angle (SZA) at the time when Terra (10:30 am) and Aqua (1:30 pm) satellites view the central Amazonian forests in March and October is between 20°and 30°. This variation minimally impacts the shape of angular signatures (supplementary data and methods section 4.4). MODIS and MISR sensors sample the rainforests very differently (figures S2(c)-(f); also see figure S1(c)). However, all the sensors record a distinct decrease in reflected NIR radiation in all view directions between October and March with no change in the overall shape of the angular signatures (figures 3(a) and (b)). Such a simple change in magnitude can only result from a change in canopy properties-this conclusion is based on the physics of how solar radiation interacts with foliage in vegetation canopies (supplementary data and methods section 4.3, figures S5(a) and (b)). The EVI, although evaluated from reflectances at NIR, red and blue wavelength bands, is tightly linked to NIR reflectance (Samanta et al 2012). Thus, the decrease in sun-sensor geometry corrected EVI (figure 2(a)) is in agreement with directly observed decreases in NIR angular signatures from October to March (figures 3(a) and (b)).
The wet season reduction in greenness is inconsistent with the hypothesis of invariant dry season greenness. Indeed the net loss of leaf area, without a corresponding net gain elsewhere during the seasonal cycle, will result in rainforests without leaves in a few years. If wet Amazonian forests somehow maintain consistent canopy structure and greenness during the dry season, then they must be either aseasonal or the entire seasonal cycle must be confined to the wet season, but this argument lacks empirical support. The question then arises whether variations in angular signatures of forest reflectance during the dry season support this inference?
Therefore, let us now consider NIR reflectances from early (25 June to 10 July) and the late dry season (15 October to 30 October) when both sun position in the sky and sensor sampling vary significantly (figures S2(a)−(d); also see figure S1(c)). MODIS and MISR measurements are made at significantly higher SZA in June (∼35°-40°) compared to October (∼20°-30°). The magnitude and shape of angular signatures are impacted when both canopy properties and SZA vary. However, a higher or equal reflectance at lower SZA relative to reflectance at higher SZA always indicates an increase in leaf area and foliage scattering properties according to the physics of radiation interaction in vegetation (supplementary data and methods section 4.4−4.5, figures S5(c)−(f)). This is observed clearly in MISR data (figure 3(d)) because this sensor views the Earth's surface with nine cameras simultaneously, as opposed to the two MODIS sensors (figure 3(c)), which are capable of only one view each (figure S3). Further, the juxtaposition of the two angular signatures in figure 3(d) is significantly different than that predicted by theory for the case of identical canopies (supplementary data and methods section 4.6). Thus, the NIR angular signatures in figure 3(d) indicate a change in vegetation structure (LAI) and greenness (EVI) during the dry season.

Conclusions
Satellite data indicate a distinct sunlight-mediated seasonality in leaf area and photosynthetic carbon fixation over unstressed rainforests in central Amazonia. This seasonal cycle is not an artefact of seasonal changes in sun position in the sky or how the satellitesensor measures the reflected radiation field. The spatially expansive remote sensing data agree with available in situ data. A better understanding of how the rainforests will respond to climate change depends on future ground campaigns as satellite data can complement, but not substitute, field data.