photosynthetic seasonality of global tropical supplementary … · “photosynthetic seasonality of...
TRANSCRIPT
1
Supplementary Information for
“Photosynthetic seasonality of global tropical forests constrained by hydroclimate”
Kaiyu Guan*, Ming Pan, Haibin Li, Adam Wolf, Jin Wu, David Medvigy, Kelly K.
Caylor, Justin Sheffield, Eric F. Wood, Yadvinder Malhi, Miaoling Liang, John S.
Kimball, Scott Saleska, Joe Berry, Joanna Joiner, and Alexei I. Lyapustin
*Correspondence to: Kaiyu Guan
Tel: +1-609-647-1368
Email: [email protected]
1. Datasets:
Enhanced Vegetation Index (EVI): EVI is an index of landscape-integrated
vegetation greenness [Huete et al., 2006] and photosynthetic capacity [Sellers et
al., 1992], which is related to photosynthetic potentials under ideal environmental
conditions, and thus EVI reflects an inherent vegetation photosynthetic property.
EVI has been found to correlate with independent ground-based measures of
photosynthesis from eddy flux towers across multiple biomes [Rahman, 2005;
Huete et al., 2006, 2008; Sims et al., 2006], though the direct linkage between
EVI and photosynthesis rate (or gross primary production, GPP) in tropical
evergreen forests is more complicated and is an ongoing research topic, as leaves
in tropical rainforests are present throughout the year. Here we use the EVI as a
proxy for canopy photosynthetic capacity in tropical forest regions. We use the
MODIS MOD13C1 Collection 5 EVI product [Didan and Huete, 2006]. This
product contains cloud-free spatial composites of the gridded 16-day 1-km
MOD13A2, and cloud-free global coverage is achieved by replacing clouds with
the historical MODIS time series climatology for that specific time step. This 16-
day composite EVI has 0.05 degree resolution, extending from 2000 to present.
Data from 2002 to 2012 are used in this work. EVI has improved performance
over the traditionally used Normalized Difference of Vegetation Index (NDVI),
by addressing atmospheric contamination effects and having improved sensitivity
Photosynthetic seasonality of global tropical forests constrained by hydroclimate
SUPPLEMENTARY INFORMATIONDOI: 10.1038/NGEO2382
NATURE GEOSCIENCE | www.nature.com/naturegeoscience 1
© 2015 Macmillan Publishers Limited. All rights reserved
2
to higher canopy leaf biomass levels [Huete et al., 2002]. The MODIS Collection
5 EVI has also been refined in its correction and filtering of atmosphere aerosol
and cloud contamination [Didan and Huete, 2006], and its quality has been
largely improved as compared with the Collection 4 EVI product [Samanta et al.,
2012]. Only “good-quality” EVI retrievals identified from the first two bits of the
MOD13C1 product Quality flag were used in this study for each pixel. It is worth
noting that the standard MODIS EVI product used here is not specifically
corrected for sun-sensor geometry (or Bi-directional reflectance distribution
function, BRDF), which can relatively exaggerate (when the sensor is in the
direction of solar backscatter ) or minimize (when the sensor is in the direction of
solar forward-scatter) surface reflectances [Jupp and Strahler, 1991]. The
standard 16-day EVI product composites individual observations that are taken
from a mix of back- and forward-scatter directions, and thus in principal could
lead to seasonal patterns that are artifacts if composites are made from
systematically biased observations [Morton et al., 2014]. We tested whether such
artifacts were present, and whether they were sufficient to influence our results
by comparing the mean seasonal cycle of the standard MODIS EVI product with
another EVI version from the Multi-Angle Implementation of Atmospheric
Correction algorithm (MAIAC) [Lyapustin et al., 2012]. The MAIAC EVI
incorporates both rigorous BRDF correction and stricter cloud, atmosphere and
aerosol corrections [Lyapustin et al., 2012]. We found that MAIAC EVI
normalized to the fixed sun-view geometry using derived BRDF model has
smaller absolute changes in seasonal variation than the standard MODIS EVI,
indicating that the artifact was present; but both corrected and uncorrected data
records have very consistent relative seasonal patterns (Figure S1, also see details
in section 2 below). These results demonstrate that, though present at low levels,
the seasonal artifact is not the cause of the signal of dry season green-up in the
central Amazon (Figure S1, S2). Additional details are provided in section 2
below. We conclude that the seasonal variation in the standard MODIS EVI
product is dominated by a true vegetation signal and is thus appropriate for
understanding the co-variation of photosynthetic capacity with climate.
© 2015 Macmillan Publishers Limited. All rights reserved
3
Solar-induced Fluorescence (SIF): SIF is an electromagnetic emission in the 650-
800nm range originating from plant photosynthetic machinery, and it is
theoretically linearly correlated with the electron transport rate of photosynthetic
activity [Zhang et al., 2014]. Theoretical and experimental studies [van der Tol et
al., 2009; Zarco-Tejada et al., 2013] demonstrate that under normal temperature
range and normal light conditions (e.g. most satellite overpass times), SIF is
linearly correlated with GPP and co-varies with GPP under environmental
stresses. At canopy and ecosystem scales, SIF is found to be better correlated
with and more closely tracks the seasonality of GPP observations determined
from tower eddy covariance flux measurements than other available GPP
products and reflectance-based vegetation indices [Guanter et al., 2014; Joiner et
al., 2014]. The SIF data used here are retrieved near the λ=740nm far-red
fluorescence emission peak from the Global Ozone Monitoring Experiment-2
(GOME-2) instrument onboard Eumetsat’s MetOp-A satellite. GOME-2 is in a
sun-synchronous orbit with overpass at about 09:30AM local time, near the
window of peak daily photosynthesis at these latitudes. The SIF algorithm
[Joiner et al., 2013] disentangles three spectral components near the peak of the
far-red chlorophyll fluorescence emission feature: atmospheric absorption (due to
water vapor), surface reflectance, and fluorescence radiance. The derived
GOME-2 SIF record (version 25, level 3) extends from 2007-present, and the
data from 2007 to 2012 has been used in this study. The SIF data has been cloud
filtered, and the data with the effective cloud fraction (fc)>0.4 have been
eliminated, where fc was computed from the black-sky 16 day gridded filled land
surface albedo product from Aqua MODIS (MOD43B3) at 656 nm [Schaaf et al.,
2002]. Using more or less stringent criteria on cloud filtering within a moderate
range did not substantially alter the derived spatial and temporal patterns of SIF
[Joiner et al., 2013]. All SIF data with SZA > 70° were also eliminated. The
current version of SIF has not incorporated any sun-sensor geometry correction,
but we find that its multi-year mean seasonal cycle is highly consistent with that
of the standard MODIS EVI and the BRDF-corrected MAIAC EVI (Figure S1),
which provides confidence of using these data in our analysis. The SIF data have
© 2015 Macmillan Publishers Limited. All rights reserved
4
been gridded to 0.5° spatial resolution and monthly time resolution, with
estimated errors of 0.1-0.2 mW/m2/nm/sr [Joiner et al., 2013].
Precipitation: The precipitation dataset is from TRMM Multi-satellite
Precipitation Analysis (TMPA) 3b42V6 [Huffman et al., 2007], which has 0.25
degree resolution and 3-hourly time step, covering period from 1998 to near
present. Data from 2002 to 2012 are used in this analysis. The TMPA is merged
from microwave and infrared satellite data, and adjusted to match local gauge
data at the monthly scale.
Potential evapotranspiration (PET): PET was calculated using the Penman-
Monteith equation forced by various climate records and remote sensing data
[Vinukollu et al., 2011] at 0.5 degree resolution from 2002 to 2012; we also
calculated PET using a revised Priestley-Taylor approach [Fisher et al., 2009;
Vinukollu et al., 2011] with the same forcing records, and found that two PET
algorithms have less than 5% difference in the mean annual range of PET, and
thus using either of them does not affect our conclusion.
Terrestrial Water Storage (TWS) Anomaly: The Gravity Recovery and Climate
Experiment (GRACE) monthly TWS Anomaly data used in this study were
obtained from the NASA data archive at the Jet Propulsion Laboratory (JPL)
[Swenson and Wahr, 2006]. We used the recommended ensemble mean
[Sakumura et al., 2014] of three independent versions of the data record from
three difference research units: GFZ (GeoforschungsZentrum Potsdam), CSR
(Center for Space Research at University of Texas, Austin), and JPL in our study.
The GRACE data are available in a global grid with 1 degree spatial resolution
and have been processed using a consistent 300 km wide Gaussian filter. GRACE
monthly TWS data from 2002 to 2012 are used in this analysis. The GRACE data
were downloaded from ftp://podaac-
ftp.jpl.nasa.gov/allData/tellus/L3/land_mass/RL05/netcdf/. The data details can
be found in http://grace.jpl.nasa.gov/data/gracemonthlymassgridsland/.
Land Cover: the Global Land Cover 2000 database (Global Land Cover 2000
database, European Commission, Joint Research Centre, 2003
(http://bioval.jrc.ec.europa.eu/products/glc2000/glc2000.php) was used.
© 2015 Macmillan Publishers Limited. All rights reserved
5
GLC2000 uses 14 months of pre-processed daily global data acquired by the
VEGETATION instrument on board the SPOT 4 satellite. The data was acquired
from http://bioval.jrc.ec.europa.eu/products/glc2000/glc2000.php. In this study,
we focus only on intact tropical evergreen forests and try to exclude all other
human impacted area and land cover types.
2. Comparison between MODIS MOD13C EVI and MAIAC EVI over Amazonia:
Figure S1 and S2
A recent study [Morton et al., 2014] suggested that the effects of sun-sensor geometry on
reflectance of vegetation introduced seasonal artifacts in Amazon forests. They claimed
that this artifact was sufficiently large to eliminate the previously reported dry-season
“green-up” in the EVI signal that had been attributed to vegetation dynamics [Huete et al.,
2006; Myneni et al., 2007]. To investigate whether sun-sensor geometry has a significant
impact on the mean seasonality pattern of EVI reported here, we compared the mean
seasonal cycles of two EVI data products for Amazon evergreen forest regions from 2002
to 2012: (1) the standard MODIS (MOD13C1, Collection 5) EVI product used in this
study, which is processed without BRDF correction; and (2) a new EVI dataset from the
MAIAC algorithm which also includes BRDF adjusted to a fixed sun angle and nadir
view as described below.
MAIAC is an advanced algorithm that uses time series analysis and a combination of
pixel- and image-based processing to improve accuracy of cloud detection, aerosol
retrievals, and atmospheric correction based on MODIS calibrated and geolocated (L1B)
measurements [Lyapustin et al., 2011a, 2011b, 2012]. MAIAC provides a suite of
gridded 1km resolution atmospheric and surface products including bidirectional
reflectance factors (BRF, also called "surface reflectance"), albedo, and Ross-Thick Li-
Sparse (RTLS) [Lucht et al., 2000] BRDF model parameters in 7 spectral land bands. The
data were downloaded from ftp://ladsweb.nascom.nasa.gov/MAIAC/. We conducted the
regional EVI comparison for Amazon tropical evergreen forests because MAIAC data
were only available for the Amazon and U.S. region at the time of this study.
© 2015 Macmillan Publishers Limited. All rights reserved
6
We calculated both MAIAC EVI and BRDF-corrected EVI (called “EVIn” below). The
BRDF correction normalizes BRF from a given view geometry to the fixed geometry of
nadir view and 45 sun zenith angle using the retrieved BRDF model, and then
normalized EVIn is computed from these values.
We compared the mean seasonal cycle based on bi-monthly data among MODIS EVI,
MAIAC EVI and MAIAC EVIn for Amazon tropical evergreen forest pixels (Figure S1).
We found that the three EVI versions are highly correlated in their mean seasonal cycle.
80% of the pixels show statistically significant positive correlations (p-value<0.1) of
mean seasonality between MODIS EVI and MAIAC EVIn, and 65% of the pixels have
Spearman correlation coefficients above 0.5. Similar results were found for the
correlations of mean seasonality between MODIS EVI and MAIAC EVI (78% of all the
pixels are statistically significantly positive correlations, with 59% having correlation
coefficient of more than 0.5). The strong correlation of mean seasonality between
MAIAC EVI and MAIAC EVIn (98% of all the pixels have statistically significantly
positive correlations, and 95% have correlation coefficient of more than 0.5) demonstrate
that the added BRDF correction brings little change to the mean seasonal cycle of EVI.
Thus the difference between MODIS EVI and MAIAC EVI/EVIn can presumably be
attributed primarily to the improved mitigation of cloud, aerosol and atmospheric effects
by the MAIAC algorithm.
We also analyzed changes in the EVI mean seasonality during the dry-season (June to
October) for the Amazon tropical evergreen forest regions in Figure S2, following the
similar analysis of Morton et al. (2014). Our results show that the great majority of pixels
show positive EVI change from June to October from the different data sources,
including MODIS EVI (89%), MAIAC EVI (90%), and MAIAC EVIn (82%). The
MOD13C1 and MAIAC records show very similar ranges of dry-season EVI change,
with an average increase of 8.0±6.4% of the maximum EVI range at per-grid level from
June to October from MOD13C1 product, and 9.1±7.1% of the maximal EVI range from
the MAIAC product; the MAIAC EVIn shows a smaller dry-season EVI increase
(4.8±6.0%), but the dry season signal increase is still well above noise (1%~2% for
© 2015 Macmillan Publishers Limited. All rights reserved
7
MAIAC EVI) [see also Hilker et al., 2012]. We thus conclude that the Amazon dry-
season green-up is a robust signal from the different MODIS EVI data sources, and that
the BRDF correction can decrease the absolute range of seasonality, but does not change
the relative seasonal variation. This analysis adds confidence to our analysis and findings
based on the seasonality of EVI.
3. Relationship between SIF and EVI: Figure S3 and S4
We compared the multi-year mean seasonality of SIF and EVI (both from the standard
MODIS EVI product and the MAIAC EVIn) at the monthly scale for tropical evergreen
forests in the Amazon, by calculating their Spearman correlation coefficient at the
gridded resolution (Figure S3). We find that SIF seasonality is highly correspondent with
that of EVI across the Amazon, with 96.3% of all valid grid cells having positive
correlation (67.4% of all the valid grids are statistically significant at p<0.1), and with
92.3% having positive correlation with the MAIAC EVIn (55.3% are statistically
significant).
We also compared the ΔEVI(wet-dry) and ΔSIF(wet-dry) for all pan-tropical evergreen forest
regions (Figure S4, only using the standard MODIS EVI-derived ΔEVI(wet-dry)). We find
that these two derived variables are highly correlated, with Spearman correlation
coefficient of 0.86, and the fitted Lowess curve has a very small y-axis intercept (0.02).
The above results demonstrate that SIF and EVI have a consistent seasonal pattern across
the broad area spanning the entire pan-tropical evergreen forest domain.
4. Validation of ΔEVI(wet-dry) and ΔSIF(wet-dry) with flux tower observations in
Amazon: Figure S5
We applied the same approach of calculating wet- and dry-season differences normalized
by their annual mean values to GPP observed at the eight flux tower sites located across
the Amazon basin. These sites were chosen based on data availability and relative land-
cover homogeneity (i.e. 90% of the land cover type in the 0.5 degree grid should be the
same type that the flux tower site has). All the flux tower information can be found in
© 2015 Macmillan Publishers Limited. All rights reserved
8
Restrepo-Coupe et al. [2013]. The same large-scale precipitation and PET records at 0.5
degree resolution used in our study were also used at the flux-tower sites to derive wet
and dry season, and the derived dry season length is consistent with the site-level
measurements and analysis [Restrepo-Coupe et al., 2013]. ΔEVI(wet-dry) and ΔSIF(wet-dry)
were extracted at the finest resolution of the products (MODIS EVI: 0.05 degree; MAIAC
EVIn: 1km; SIF: 0.5 degree) at each corresponding flux tower location.
Our results (Figure S5) show that GPP at the flux towers shows a consistent pattern in
relation to those of EVI and SIF. Almost all the data show a significantly positive
difference for the wet-season value minus dry-season value at the four sites (i.e. PDG,
JAV, FNS and RJA), where their MAP values are well below the determined rainfall
threshold (2000 mm/year). The other four sites (i.e. K67, K83, K34 and CAX), with
MAP close to or above the determined threshold, have wet-dry season differences that are
either negative or insignificant (close to zero). These flux tower GPP estimates thus
support our findings which are based on the continental scale analysis.
The authors would like to thank all the PIs and the staff of each tower site for their
technical, logistical and extensive fieldwork.
5. t-test distribution for ΔEVI(wet-dry) and ΔSIF(wet-dry) patterns: Figure S6
We employed the t-test to determine the statistical significance of ΔEVI(wet-dry) and
ΔSIF(wet-dry) patterns for each grid, with the null hypothesis that ΔEVI(wet-dry) (or ΔSIF(wet-
dry)) is equal to zero (paired t-test, n=11 years per grid for EVI, and n=6 for SIF). Figure
S6 shows the spatial pattern for the t-test distribution, with blue indicating significantly
negative values, and red indicating significantly positive values. These spatial patterns
largely mirror those in Figure 1 d-i. For ΔEVI(wet-dry), the Amazon has 26.1% of tropical
evergreen forest grids as significantly positive (p<0.05), while 33.7% are significantly
negative; SE Asia has 16.3% (11.8%) significantly positive (negative); Africa has 75.4%
(0.5%) significantly positive (negative); the remaining areas are insignificant. For
ΔSIF(wet-dry), the Amazon has 25.0% of all the valid grids as significantly positive, while
© 2015 Macmillan Publishers Limited. All rights reserved
9
20.0% are significantly negative; SE Asia has 14.7% (1.3%) significantly positive
(negative); Africa has 78.4% (0.1%) significantly positive (negative).
6. Hydrological characteristics in the pan-tropical forest regions
6.1 Mean annual precipitation and dry season length: Figure S7
6.2 Spatial pattern of “maximum release of water storage”, “dry season total deficit” and
their difference: Figure S8
6.3 Impacts of dry-season length on “maximum release of water storage”, “dry season
total deficit” in tropical rainforests: Figure S9
6.4 Different rainfall regimes in supply-demand space: Figure S10
The scatterplot of “Maximum release of water storage” (supply) and “dry season total
deficit” (demand) shows the supply-demand relationship for dry season plant water use.
MAP of each grid are color shaded in this scatterplot (Figure S10) showing how total
annual rainfall progresses in supply-demand space.
6.5 Dry-season supply-demand relationships for different continents: Figure S11
7. Basin-scale quantification of water budget: Figure S12 and S13
A basin-scale analysis involving a complete quantification of the water budget (only for
Amazon and Congo basins due to data availability, Pan et al., 2012) further corroborates
our findings that 39.8% of the total dry-season water supply in the Amazon basin is from
the previous wet-season water storage, compared with only 14.4% in the Congo basin
(Figure S13). The seasonal carry-over of subsurface water storage plays a much more
important role in tropical forests of Amazonia than Africa. These continental differences
in the water budget mostly arise from longer dry seasons in Africa, as well as more
excess precipitation inputs and thus relatively larger subsurface water storage during wet
seasons in Amazonia (Figure S7).
We used a multi-source optimized basin-scale water budget record [Pan et al., 2012]. A
systematic method is used to optimally combine estimates of the terrestrial water budget
from different data sources and to enforce the water balance constraint using data
assimilation techniques. Specifically, a conventional analysis of the errors and biases in
© 2015 Macmillan Publishers Limited. All rights reserved
10
different data sources is conducted based on existing validation/error studies and other
information such as sensor network density, model physics, and calibration procedures.
Then, the data merging process combines different estimates so that biases and errors
from different data sources can be compensated to the greatest extent and the merged
estimates have the best possible confidence. Finally, water balance errors are resolved
using the constrained Kalman filter technique. Ten global datasets have been included
including: in-situ observations, remote sensing retrievals, land surface model simulations,
and global reanalyses (Table 1 in [Pan et al., 2012]). The procedure is applied to 32
globally distributed major basins for the 1984–2006 record. We use Amazon and Congo
basin data in this study to represent the water budget in Amazonian and African tropical
forest regions (Figure S13).
We find that the Amazonian tropical forest region has much wetter and longer wet
seasons than those in Africa, though both regions have similarly low monthly
precipitation during the dry seasons at 99.6 mm/month and 95.1 mm/month, respectively
(Figure S13). We find that the surplus of precipitation during the long wet season in
Amazonia (~9 months at the basin scale) leads to a significant storage of water, primarily
as soil moisture and ground water [Crowley and Mitrovica, 2006; Pokhrel et al., 2013].
This reserve is subsequently used by tropical forests during the short dry seasons of ~3
months duration, with a depletion rate of approximately 65.5 mm/month, through the
mechanism of deep plant roots and facilitated hydraulic redistribution [Nepstad et al.,
1994; Meir et al., 2009; Betts and Silva Dias, 2010]. In contrast, African tropical forests
do not gain substantial water storage during their relatively shorter wet seasons of ~5
months, which limits water consumption during dry seasons to a depletion rate of only
15.7mm/month. When taking into account carry-over from water storage, Amazonian
tropical forest regions have approximately 165.3mm/month total water supply during the
dry seasons, which is 49% larger than tropical forests in Africa (111.1mm/month). Thus
water surplus across the wet and dry seasons generally satisfies the plant water need in
Amazonian tropical forest regions but not in Africa, leading to contrasting sensitivity of
photosynthetic seasonality to seasonal drought. The basin-scale analysis is thus consistent
with what we discovered at the grid level.
© 2015 Macmillan Publishers Limited. All rights reserved
11
8. The relationship between groundwater table depth and rooting depths in tropical
forest regions: Figure S14 and S15
A global dataset for groundwater table depth [Fan et al., 2013] is used here to assess
whether characteristic rooting depth of tropical forest is sufficient to enable access to
groundwater. The spatial pattern of groundwater table depths for global tropical regions
are shown in Figure S14, which has a native spatial resolution of 30 arc-sec (~1km), and
was aggregated to 0.25 degree here. Elevations above 500 meters are filtered out because
of biased deeper water table depths in mountain regions. The elevation data is from the
GLDAS at 0.25 degree resolution (http://ldas.gsfc.nasa.gov/gldas/GLDASelev.php). As
the coarse resolution of recharge input (1 degree) to the model of Fan et al. [2013] could
not provide reliable details particularly for the Asian Islands (i.e. many small patchy
islands), the data provider suggested focusing the analysis only on continental Amazon
and Africa.
Figure S15 shows that groundwater table depth decreases with increasing MAP for the
tropical forest regions. The result confirms that in very humid tropical regions, deep roots
are not necessary because the groundwater table is shallow (~ 2m) from which
groundwater could be easily accessed via shallow roots. The water table depth starts to
increase sharply with MAP decreasing below 2500 mm/year, indicating that deeper
rooting systems are necessary for accessing ground water especially in drier regions with
a deeper water table, e.g. southeast Amazon. It has been reported that tropical forests in
eastern and southern Amazon have deep roots of 10-15 meters [Nepstad et al., 1994,
2004; Davidson et al., 2011]. We find that the groundwater table depths in tropical forest
regions are around 10±3 meters at the MAP of 2000 mm/year, which allows deep-rooted
forests in these regions to tap the groundwater resources.
References:
Betts, A. K., and M. A. F. Silva Dias (2010), Progress in understanding land-surface-
atmosphere coupling from LBA research, J. Adv. Model. Earth Syst., 2, 6,
doi:10.3894/JAMES.2010.2.6.
© 2015 Macmillan Publishers Limited. All rights reserved
12
Crowley, J., and J. Mitrovica (2006), Land water storage within the Congo Basin inferred
from GRACE satellite gravity data, Geophys. Res. Lett., 33, 2–5,
doi:10.1029/2006GL027070.
Davidson, E., P. A. Lefebvre, P. M. Brando, D. M. Ray, S. E. Trumbore, L. A. Solorzano,
J. N. Ferreira, M. M. C. Bustamante, and D. C. Nepstad (2011), Carbon Inputs and
Water Uptake in Deep Soils of an Eastern Amazon Forest, For. Sci., 57(1), 51–58.
Didan, K., and A. Huete (2006), MODIS Vegetation Index Product Series Collection 5
Change Summary
(http://landweb.nascom.nasa.gov/QA_WWW/forPage/MOD13_VI_C5_Changes_Do
cument_06_28_06.pdf).
Fan, Y., H. Li, and G. Miguez-Macho (2013), Global patterns of groundwater table
depth., Science, 339(6122), 940–3, doi:10.1126/science.1229881.
Fisher, J. B. et al. (2009), The land-atmosphere water flux in the tropics, Glob. Chang.
Biol., 15(11), 2694–2714, doi:10.1111/j.1365-2486.2008.01813.x.
Guanter, L. et al. (2014), Global and time-resolved monitoring of crop photosynthesis
with chlorophyll fluorescence, Proc. Natl. Acad. Sci. U. S. A., 111(14), E1327–33,
doi:10.1073/pnas.1320008111.
Hilker, T., A. I. Lyapustin, C. J. Tucker, P. J. Sellers, F. G. Hall, and Y. Wang (2012),
Remote sensing of tropical ecosystems: Atmospheric correction and cloud masking
matter, Remote Sens. Environ., 127, 370–384, doi:10.1016/j.rse.2012.08.035.
Huete, A., K. Didan, T. Miura, E. P. Rodriguez, X. Gao, and L. G. Ferreira (2002),
Overview of the radiometric and biophysical performance of the MODIS vegetation
indices, Remote Sens. Environ., 83, 195–213.
Huete, A. R., K. Didan, Y. E. Shimabukuro, P. Ratana, S. R. Saleska, L. R. Hutyra, W.
Yang, R. R. Nemani, and R. Myneni (2006), Amazon rainforests green-up with
sunlight in dry season, Geophys. Res. Lett., 33(6), L06405,
doi:10.1029/2005GL025583.
Huete, A. R., N. Restrepo-Coupe, P. Ratana, K. Didan, S. R. Saleska, K. Ichii, S.
Panuthai, and M. Gamo (2008), Multiple site tower flux and remote sensing
comparisons of tropical forest dynamics in Monsoon Asia, Agric. For. Meteorol.,
148(5), 748–760, doi:10.1016/j.agrformet.2008.01.012.
Huffman, G. J., D. T. Bolvin, E. J. Nelkin, D. B. Wolff, R. F. Adler, G. Gu, Y. Hong, K.
P. Bowman, and E. F. Stocker (2007), The TRMM Multisatellite Precipitation
Analysis (TMPA): Quasi-Global, Multiyear, Combined-Sensor Precipitation
Estimates at Fine Scales, J. Hydrometeorol., 8(1), 38–55, doi:10.1175/JHM560.1.
© 2015 Macmillan Publishers Limited. All rights reserved
13
Joiner, J., L. Guanter, R. Lindstrot, M. Voigt, a. P. Vasilkov, E. M. Middleton, K. F.
Huemmrich, Y. Yoshida, and C. Frankenberg (2013), Global monitoring of
terrestrial chlorophyll fluorescence from moderate spectral resolution near-infrared
satellite measurements: methodology, simulations, and application to GOME-2,
Atmos. Meas. Tech., 6, 2803–2823, doi:10.5194/amtd-6-3883-2013.
Joiner, J. et al. (2014), The seasonal cycle of satellite chlorophyll fluorescence
observations and its relationship to vegetation phenology and ecosystem atmosphere
carbon exchange, Remote Sens. Environ., 152, 375–391,
doi:10.1016/j.rse.2014.06.022.
Jupp, D. L. B., and A. H. Strahler (1991), A hotspot model for leaf canopies, Remote
Sens. Environ., 38(3), 193–210, doi:10.1016/0034-4257(91)90089-O.
Lucht, W., C. B. Schaaf, and A. H. Strahler (2000), An algorithm for the retrieval of
albedo from space using semiempirical BRDF models, IEEE Geosci. Remote.
Remote, 38(2), 977–998, doi:10.1109/36.841980.
Lyapustin, a., Y. Wang, I. Laszlo, R. Kahn, S. Korkin, L. Remer, R. Levy, and J. S. Reid
(2011a), Multiangle implementation of atmospheric correction (MAIAC): 2. Aerosol
algorithm, J. Geophys. Res., 116(D3), D03211, doi:10.1029/2010JD014986.
Lyapustin, A., J. Martonchik, Y. Wang, I. Laszlo, and S. Korkin (2011b), Multiangle
implementation of atmospheric correction (MAIAC): 1. Radiative transfer basis and
look-up tables, J. Geophys. Res., 116(D3), D03210, doi:10.1029/2010JD014985.
Lyapustin, A. I., Y. Wang, I. Laszlo, T. Hilker, F. G.Hall, P. J. Sellers, C. J. Tucker, and
S. V. Korkin (2012), Multi-angle implementation of atmospheric correction for
MODIS (MAIAC): 3. Atmospheric correction, Remote Sens. Environ., 127, 385–
393, doi:10.1016/j.rse.2012.09.002.
Meir, P. et al. (2009), The Effects of Drought on Amazonian Rain Forests, Amaz. Glob.
Chang. (eds M. Keller, M. Bustamante, J. Gash P. Silva Dias), Am. Geophys. Union,
Washington, D. C., 429–449.
Morton, D. C., J. Nagol, C. C. Carabajal, J. Rosette, M. Palace, B. D. Cook, E. F.
Vermote, D. J. Harding, and P. R. J. North (2014), Amazon forests maintain
consistent canopy structure and greenness during the dry season., Nature, 506(7487),
221–4, doi:10.1038/nature13006.
Myneni, R. B. et al. (2007), Large seasonal swings in leaf area of Amazon rainforests.,
Proc. Natl. Acad. Sci. U. S. A., 104(12), 4820–3, doi:10.1073/pnas.0611338104.
Nepstad, D., C. de Carvalho, E. Davidson, P. H. Jipp, P. A. Lefebvre, G. H. Negreiros, E.
D. da Silva, T. A. Stone, S. E. Trumbore, and S. Vieira (1994), The role of deep
© 2015 Macmillan Publishers Limited. All rights reserved
14
roots in the hydrological and carbon cycles of Amazonian forests and pastures,
Nature, 3, 666–669.
Nepstad, D., P. Lefebvre, U. Lopes da Silva, J. Tomasella, P. Schlesinger, L. Solorzano,
P. Moutinho, D. Ray, and J. Guerreira Benito (2004), Amazon drought and its
implications for forest flammability and tree growth: a basin-wide analysis, Glob.
Chang. Biol., 10(5), 704–717, doi:10.1111/j.1529-8817.2003.00772.x.
Pan, M., A. K. Sahoo, T. J. Troy, R. K. Vinukollu, J. Sheffield, and E. F. Wood (2012),
Multisource Estimation of Long-Term Terrestrial Water Budget for Major Global
River Basins, J. Clim., 25(9), 3191–3206, doi:10.1175/JCLI-D-11-00300.1.
Pokhrel, Y. N., Y. Fan, G. Miguez-macho, S. Han, P. Sciences, N. Brunswick, and N. P.
Group (2013), The Role of Groundwater in the Amazon Water Cycle : 3 . Influence
on Terrestrial Water Storage and Comparison with GRACE, J. Geophys. Res.
Atmos., 118(8), 3233–3244.
Rahman, a. F. (2005), Potential of MODIS EVI and surface temperature for directly
estimating per-pixel ecosystem C fluxes, Geophys. Res. Lett., 32(19), L19404,
doi:10.1029/2005GL024127.
Restrepo-Coupe, N. et al. (2013), What drives the seasonality of photosynthesis across
the Amazon basin? A cross-site analysis of eddy flux tower measurements from the
Brasil flux network, Agric. For. Meteorol., 182-183, 128–144,
doi:10.1016/j.agrformet.2013.04.031.
Sakumura, C., S. Bettadpur, and S. Bruinsma (2014), Ensemble prediction and
intercomparison analysis of GRACE time-variable gravity field models, Geophys.
Res. Lett., 1389–1397, doi:10.1002/2013GL058632.1.
Samanta, A., S. Ganguly, E. Vermote, R. R. Nemani, and R. B. Myneni (2012),
Interpretation of variations in MODIS-measured greenness levels of Amazon forests
during 2000 to 2009, Environ. Res. Lett., 7(2), 024018, doi:10.1088/1748-
9326/7/2/024018.
Schaaf, C. B. et al. (2002), First operational BRDF , albedo nadir reflectance products
from MODIS, Remote Sens. Environ., 83, 135–148.
Sellers, P. J., J. A. Berry, G. J. Collatz, C. B. Field, and E. G. Hall (1992), Canopy
Reflectance , Photosynthesis , and Transpiration . III . A Reanalysis Using Improved
Leaf Models and a New Canopy Integration Scheme ., Remote Sens. Environ., 42,
187–216.
Sims, D. a. et al. (2006), On the use of MODIS EVI to assess gross primary productivity
of North American ecosystems, J. Geophys. Res., 111(G4), G04015,
doi:10.1029/2006JG000162.
© 2015 Macmillan Publishers Limited. All rights reserved
15
Swenson, S., and J. Wahr (2006), Post-processing removal of correlated errors in
GRACE data, Geophys. Res. Lett., 33(8), L08402, doi:10.1029/2005GL025285.
Van der Tol, C., W. Verhoef, and a. Rosema (2009), A model for chlorophyll
fluorescence and photosynthesis at leaf scale, Agric. For. Meteorol., 149(1), 96–105,
doi:10.1016/j.agrformet.2008.07.007.
Vinukollu, R. K., E. F. Wood, C. R. Ferguson, and J. B. Fisher (2011), Global estimates
of evapotranspiration for climate studies using multi-sensor remote sensing data:
Evaluation of three process-based approaches, Remote Sens. Environ., 115(3), 801–
823, doi:10.1016/j.rse.2010.11.006.
Zarco-Tejada, P. J., a. Morales, L. Testi, and F. J. Villalobos (2013), Spatio-temporal
patterns of chlorophyll fluorescence and physiological and structural indices
acquired from hyperspectral imagery as compared with carbon fluxes measured with
eddy covariance, Remote Sens. Environ., 133, 102–115,
doi:10.1016/j.rse.2013.02.003.
Zhang, Y., L. Guanter, J. a Berry, J. Joiner, C. van der Tol, A. Huete, A. Gitelson, M.
Voigt, and P. Köhler (2014), Estimation of vegetation photosynthetic capacity from
space-based measurements of chlorophyll fluorescence for terrestrial biosphere
models., Glob. Chang. Biol., doi:10.1111/gcb.12664.
© 2015 Macmillan Publishers Limited. All rights reserved
16
Figure S1 | Spearman correlations of the mean EVI seasonal cycles among three different
EVI data sources: MOD13C1 (MODIS EVI), MAIAC without BRDF correction
(MAIAC EVI), and MAIAC with BRDF correction (MAIAC EVIn). The left column
shows the spatial patterns of correlation coefficients, and the right column shows the
corresponding histograms of the correlation coefficients. The red regions in the
histograms indicate the proportion of grids with statistically significant correlation at the
p-value of 0.1, while the black regions are statistically insignificant at the same p-value
level.
© 2015 Macmillan Publishers Limited. All rights reserved
17
Figure S2 | EVI changes from October to June normalized by the maximum EVI range at
per grid level for Amazon tropical evergreen forest regions, based on the mean EVI
seasonality of three data sources: MODIS MOD13C1 (a, b), MAIAC without BRDF
correction (c, d), and MAIAC with BRDF correction (e, f). The left column shows the
spatial patterns of normalized EVI changes, and the right column shows the
corresponding histograms of the EVI changes, with the error bars showing the mean and
standard deviation of the whole distribution.
© 2015 Macmillan Publishers Limited. All rights reserved
18
Figure S3 | Spearman ccorrelations of the mean seasonal cycles between SIF and two
EVI data sources: MOD13C1 (MODIS EVI) and MAIAC with BRDF correction
(MAIAC EVIn). The left column shows the spatial patterns of correlations, and the right
column shows the corresponding histograms of the correlations. The red regions in the
histograms indicate the proportion of grids with statistically significant correlation at the
p-value of 0.1, while the black regions are statistically insignificant at the same p-value
level.
© 2015 Macmillan Publishers Limited. All rights reserved
19
Figure S4 | Density scatterplot between multiple-year mean ΔEVI(wet-dry) and ΔSIF(wet-dry)
for the global tropical forests areas studied here. The red dashed line is the robust Lowess
fit for the data with a span of 2.
© 2015 Macmillan Publishers Limited. All rights reserved
20
Figure S5 | Inter-comparison of satellite-based photosynthetic properties and flux-tower-
based GPP estimates for wet-/dry-season difference at eight flux tower sites in the
Amazon. The wet- and dry-season differences are normalized by their annual mean
values. All flux tower information can be found in [Restrepo-Coupe et al., 2013].
© 2015 Macmillan Publishers Limited. All rights reserved
21
Figure S6 | Statistical significance of ΔEVI(wet-dry) and ΔSIF(wet-dry) pattern, quantified as
percentiles of t-test distribution under the null hypothesis that ΔEVI(wet-dry) (or ΔSIF(wet-
dry)) is equal to zero (paired t-test, n=11 years per grid for EVI, and n=6 for SIF). Red
color indicates that the grid has significantly positive ΔEVI(wet-dry) (or ΔSIF(wet-dry)), and
blue color indicates significantly negative ΔEVI(wet-dry) (or ΔSIF(wet-dry)).
© 2015 Macmillan Publishers Limited. All rights reserved
22
Figure S7 | Dry season length and mean annual precipitation (MAP) for the study areas.
MAP values beyond 3000mm/year are not differentiated.
© 2015 Macmillan Publishers Limited. All rights reserved
23
Figure S8 | Spatial patterns of maximum release from water storage (first column), dry
season total water deficit (second column) and their difference (third column) for tropical
forest regions in Amazonia, Insular Southeast Asia and Africa.
© 2015 Macmillan Publishers Limited. All rights reserved
24
Figure S9 | The impacts of MAP and dry season length on the dry season water deficit
(demand) and maximum withdraw from storage (supply) over the tropical forest regions.
© 2015 Macmillan Publishers Limited. All rights reserved
25
Figure S10 | Scatterplots between maximum release from water storage and dry season
total water deficit for all tropical evergreen forest regions, color shaded with mean annual
precipitation, with different panels showing the correspondence between various ranges
in MAP and the 1:1 line in the scatterplot. The 1:1 line indicates whether maximum
release from water storage can offset dry season total deficit and thus satisfy plant water
needs during the dry season. Circle, triangle and square symbols represent respective
grids from Amazonia, Insular SE Asia and Africa tropical rainforests. In each panel, all
grids within specific MAP ranges are black-edged, and the red dashed line is the linear
regression of these points; (a) 1250-1550mm/year; (b) 1550-1850mm/year; (c) 1850-
2150mm/year; (d) 2150-2450mm/year; (e) 2450-2750mm/year; (f) 2750-3050mm/year.
© 2015 Macmillan Publishers Limited. All rights reserved
26
Figure S11 | Scatterplots between the maximum release from water storage and dry
season total water deficit for the different tropical evergreen forest regions; color shading
refers to mean annual precipitation. The black-edged circles refer to grids with MAP
range between 1850 to 2150 mm/year; a-Amazonia; b-Insular Southeast Asia; c-Africa.
© 2015 Macmillan Publishers Limited. All rights reserved
27
Figure S12 | Geographic locations of Amazon and African Congo basins used for basin-
scale water budget analysis (revised from Pan et al., 2012).
© 2015 Macmillan Publishers Limited. All rights reserved
28
Figure S13 | Left column: mean monthly water budget for Amazonia and Congo basins.
The water budget terms (P-precipitation, ET and Q-runoff) are from a global long-term
record of the terrestrial water budget by merging a number of global datasets using data
assimilation techniques [Pan et al., 2012]. The PET data is derived based on the Penman-
Monteith equation forced with various remote sensing products [Vinukollu et al., 2011].
Right column: water budgets during wet and dry seasons in Amazonia and Congo basins.
Wet/dry seasons are defined as the months where PET is smaller/larger than P. This
water budget result shows that 39.8% of the total dry-season water supply in the Amazon
basin is from the previous wet-season water storage, compared with only 14.4% in the
Congo basin.
© 2015 Macmillan Publishers Limited. All rights reserved
29
Figure S14 | Simulated groundwater table depths from Fan et al. (2013) in tropical forest
regions: Amazon (A), Asian Islands (B) and Africa (C). Regions with elevations higher
than 500 meters were excluded from the analysis.
Figure S15 | Median (50% quantile) of the simulated water table depths (excluding
regions with elevation higher than 500 meters and SE Asian Islands due to large data
uncertainties) as a function of MAP for the global tropical forest regions, with the error
bar representing the 75% and 25% quantile values.
© 2015 Macmillan Publishers Limited. All rights reserved