REMOTE SENSING OF TROPICAL FOREST PHENOLOGY:

ISSUES AND CONTROVERSIES

Alfredo Huete1 and Scott Saleska2
1
Plant Functional Biology and Climate Change Cluster
University of Technology Sydney
P.O. Box 123, Broadway, NSW 2007
Tel. +61295144084, Fax +61295144079
alfredo.huete@uts.edu.au
2
Dept. Ecology & Evolutionary Biology, University of Arizona, Tucson, AZ USA,
saleska@email.arizona.edu

Abstract
Different spatial and spectral resolution satellite products collected at high
temporal frequencies from the AVHRR and Terra- MODIS sensors show
variable and inconsistent seasonal patterns over tropical forests with some
sensor products showing canopy drying in the dry season and negative forest
responses to seasonal drought, while other products show leaf flushing and
enhanced greening in the dry season with a positive response to seasonal
drought. Large inconsistencies have been reported in cross- satellite product
comparisons (that include MODIS and AVHRR) for tropical evergreen broadleaf
forests and inconsistencies are found among MODIS products and tropical field
observations for LAI-related products. In this study, we investigate various
remote sensing processing methods, from leaf to canopy scale, for deriving
spatial and temporal patterns of photosynthetic activity in tropical environments,
using the Brazilian Amazon and Monsoon Asia tropical forests as study areas.
We also consider the role and uncertainty played by atmospheric artifacts
associated with seasonal cloud cover and atmospheric aerosols, which render
the task of decoupling surface vegetation phenology from atmosphere
influences quite challenging.

1. Introduction
The vulnerability of tropical forest systems to climate change depends not just
on the physical climate system, but on the biological response of forests to
initial climate changes. While much attention has focused on tropical forest die-
back and savannization scenarios (Betts et al. 2004), other models imply forest
resilience. Differing modeled fates of the forest are due to model differences in
representation of forest function, not just differences in representations of
climate and current knowledge is insufficient to determine which representations
of vegetation function are most consistent with real forest ecosystems, but
continuing observations from satellites and from the network of eddy flux towers
provide tools that can rigorously test mechanisms of forest-climate interactions
on observable timescales.

1
Recent studies using MODIS remote sensing products have detected positive
‘greening’ vegetation responses to seasonal drying (Huete et al. 2006; Myneni
et al. 2007) and interannual drought (Saleska et al. 2007), which if accurate,
suggest that tropical vegetation, by apparently responding to increased dry-
period light availability, is more complex, and possibly more robust than most
ecosystem models suggest, with implications for long-term vulnerability to
climate change. Prior to these studies, models have assumed that tropical
forest canopies exhibited no seasonality, in part, because seasonal remote
sensing in the tropics is undermined by poor atmospheric conditions (Kobayashi
& Dye, 2005). Other models represented browning vegetation and decreased
activity in dry seasonal periods and interannual droughts, such as ENSO (Tian
et al. 1998; Botta et al. 2002). Similar to many modeled results, the long-term
AVHRR- NDVI time series show decreases in photosynthetic activity and
productivity with seasonal drought and interannual drought events such as
ENSO (Fig. 1) (Asner et al. 2000).

Figure 1. Seasonal precipitation and AVHRR NDVI (20 year averages) across an eco-
climatic gradient from south (Brasilia savanna) to Araguaia ecotone to the seasonally
dry rainforests to the north (Tapajos). All biomes are found to dry down during
seasonal drought periods (credit, Humberto Barbosa).

Various hypotheses for forest response to climatic variability include: (1) Light
limitation, in which tropical forest growth is primarily limited by availability of
photosynthetically-active radiation (PAR), with access to deep soil water via
deep tree roots minimizing the typical modeled water stresses; (2) Diffuse
radiation enhancement of Light-Use Efficiency (LUE), which is the amount of
photosynthesis that takes place per amount of PAR absorbed by the forest
canopy; (3) Leaf growth phenology.
However, issues remain in the interpretation and understanding of remote
sensing data. Different coarse resolution satellite products collected at high

2
temporal frequencies from the AVHRR and MODIS platforms show variable and
inconsistent seasonal and interannual patterns over tropical rainforests, with
some sensor products showing canopy drying in the dry season and negative
forest responses to drought, while other products show leaf flushing and
greening in the dry season and a positive response to drought (Fig. 2, Huete et
al. 2008). The MODIS NDVI product, on the other hand, show widespread
saturation over rainforest ecosystems (Huete et al. 2002).

Figure 2. Seasonal variations in satellite products from MODIS (GPP, FPAR, and EVI),
compared to eddy tower GPP (Pg) from SE Asia tropical forests (MaeKlong Station,
Thailand). The satellite products (all intended to relate to photosynthesis) show
divergent performance. FPAR (fraction absorbed PAR) shows no seasonality, MODIS
GPP shows a seasonality opposite to the tower, and only EVI shows a seasonality in
phase with the tower (Huete et al. 2008).

1.1 Vegetation Indices

The main scientific requirement of a VI measurement is that it combine the
chlorophyll-absorbing ‘red’ spectral region with the non-absorbing, leaf
reflectance signal in the ‘near-infrared’ (NIR) to depict vegetation “greenness” or
area-averaged canopy photosynthetic capacity. There are a variety of ways in
which the ‘red’ and ‘NIR’ bands may be combined to estimate vegetation
properties and greenness amounts, and this has resulted in a multitude of VI
equations, including band ratios, normalized differences, linear band
combinations, and ‘optimized’ band combinations. All of VIs have been
empirically related with various vegetation canopy properties, however, there
are significant differences in how they depict “greenness” and their suitability for
global, moderate resolution algorithms. The NDVI is computed as,

NDVI = [ρNIR – ρred] / [ρNIR + ρred] (1)

The MODIS enhanced vegetation index (EVI) product, on the other hand, has
minimal saturation problems through greater weighing of the near-infrared band.

3
The EVI gains its heritage primarily from the soil adjusted vegetation index
(SAVI, Huete 1988) and is an optimized combination of red and NIR bands,
designed to extract canopy greenness, independent of biases from differences
in soil background. The EVI also incorporates an ‘aerosol resistance’ term
developed in the atmosphere resistance vegetation index (ARVI) equations that
utilize the blue reflectance to stabilize aerosol influences and biased corrections
in red reflectances,

EVI = 2.5 [ρNIR – ρred] / [L + ρNIR + C1 ρred – C2 ρblue] (2)

where L is the canopy background adjustment factor, ρ is reflectance, and C1

and C2 are the aerosol resistance weights. The coefficients of the EVI equation
are L=1; C1=6 and C2 =7.5.
Since the role of the blue-band in the EVI does not provide additional
biophysical information on vegetation properties, but rather is aimed at reducing
noise and uncertainties associated with highly variable atmospheric aerosols, a
2-band adaptation of EVI should be compatible. A simple, 2-band EVI has been
formulated as a “backup algorithm” to the MODIS EVI product, and is used in
cases when the blue band yields problematic VI values, mainly in dense snow
and over certain clouds. It is derived by setting C1 = 2.4 and C2 = 0 (Jiang et al.
2008),

EVI2 = 2.5 [ρNIR - ρred] / [L + ρNIR + 2.4 ρred] (3)

where L=1. Although the EVI2 is computed without a blue band, it remains
functionally-equivalent to the EVI, although slightly more prone to aerosol noise,
which may become less significant with continuing advancements in
atmosphere correction. This also provides one method to generate a backward
compatibility of the EVI to the historical AVHRR record, and complement the
NDVI long-term record.
The primary difference between the NDVI and EVI is in the ‘L’ parameter in EVI,
which attempts to optimize the measure of ‘greenness’ through a simple, first-
order application of Beer-Lambert’s law to describe differential red and NIR
extinction through vegetation canopies, according to,

ρC = ρv + t2C ρs (4)

where canopy reflectance (ρC) is the sum of the vegetation layer reflectance (ρv)
and the two-way canopy transmitted (t2C) soil reflectance (ρs). The NDVI has a
very low optical depth penetration into canopies due to its heavier reliance on
the red channel, while the EVI will have a higher canopy penetration through
greater weighting on the NIR channel, hence allowing extended sensitivity over
higher LAI/ biomass areas where the NDVI saturates. EVI may provide a more
direct relationship with photosynthesis (GPP) in dense canopies by relying on

4
the more sensitive bio-structural NIR canopy reflectance which is less prone to
saturate. This is also supported through theoretical analyses concluding that
spectral indices that are more functional on the NIR best describe area-
averaged canopy photosynthetic capacities and gross primary productivity
(Sellers, 1987).
When ‘L’ is optimized correctly, there is no (or minimal) soil background
influences in the VI signal as the relative optical depths of the two bands are
more closely adjusted to only see the greenness of the canopy. As the NDVI
represents the special case of ‘L=0’, in contrast to L=1 in the EVI, the impact of
‘L’ on the seasonality of the ‘greenness’ signal can be shown by varying L from
0 to 1 (Fig. 3). As can be seen, the shift in ‘L’ has a significant impact on
observed seasonality/ phenology. In the Tapajos rainforest, the change in ‘L’
value from 0 (NDVI) to 1 (EVI) resulted in seasonal profiles completely out of
phase with one another, and emphasizes the importance of fusion of satellite VI
measurements with ground, in situ observations for more accurate inputs to
models.

Figure 3. (top) Sensitivity of vegetation index seasonal profiles to the ‘L’ parameter.
L=0 is the case of NDVI, while L=1 represents the EVI, EVI2; L=0.5 (dark blue) as in
SAVI; L=0.25 (light blue) and L=0.05 (orange); (bottom) simultaneous in situ tower
derived GPP monthly fluxes at the Tapajos Amazon rainforest site.

Large inconsistencies have been reported in cross- satellite product

comparisons (that include MODIS and AVHRR) for tropical evergreen broadleaf
forests (Garrigues et al. 2008). At local tower site scales, Doughty and Goulden
(2008) showed in situ LAI measurements differed markedly from seasonal
cycles of the MODIS LAI product (MOD15). Hutrya et al. (2007) showed MODIS
LAI and EVI products not in phase with each other. There is a great need to
better relate satellite products with canopy functional processes in tropical
forests in order to understand what changes in greenness signify.
Remote sensing methods directly detect photosynthetic pigments in vegetation,
and could be a powerful tool for investigating spatial and temporal patterns of

5
photosynthetic metabolism in challenging tropical environments, but uncertainty
remains about atmospheric artefacts from highly seasonal cloud cover and
aerosol loads in tropical atmospheres. Some argue that the observed remote
sensing responses are due to artifact. Kobayashi and Dye (2005) analyzed the
GIMMS AVHRR- NDVI time series over the Amazon and found the seasonal
signal to be primarily dominated by cloud and aerosol contamination.
We have previously argued (Huete et al. 2006; Saleska et al. 2007) that by
choosing satellite products that correlate with measurements on the ground
(e.g. satellite EVI with ecosystem scale photosynthetic fluxes measured from
towers), and by a combination of appropriate selection of high quality data from
relatively uncontaminated pixels and correction for residual contamination, we
arrive at observations that are robust to the problems cited above (for example,
because MODIS EVI uses atmospherically-corrected surface reflectances and
an aerosol resistance term, we believe it is minimally affected by residual
aerosols up to optical depths of over 1).

1.2. Interannual Drought

In 2005, large areas of the southwestern Amazon basin experienced one of the
most intense droughts of the last 100 years (Marengo et al. 2008). During the
2005 drought, satellite observations showed short-term (3 mos) increases in
MODIS EVI, suggesting increased forest photosynthetic capacity (Fig. 4,
Saleska et al. 2007), while Phillips et al. (2009), using field measurements,
showed a longer-term decline in the rate of biomass accumulation, slowing or
reversing previous trends, primarily due to an increase in tree mortality. It is not
currently understood whether these seemingly divergent responses (short term
increase in photosynthetic capacity during the peak months of drought,
encompassed by excess mortality integrated over several years) may in fact be
reconciled by accounting for differing time scales and effects of time lags.
The divergent responses, however, pose specific questions, including: (1) What
is the role of light vs. water limitation in controlling vegetation response to
drought; (2) What are the causes for variation in forest responses to the same
drought event (while most areas greened up, distinct “brown-down” regions
were also observed (Fig. 4); (3) How does a drought response interact with
nominal patterns of seasonal drought, given the varying nature of seasonal
drought responses (Fig. 2).
Samanta et al (2010) suggest that apparent tropical forest greening during the
Amazon drought of 2005 (as reported in Saleska et al. 2007) is in fact due to
atmospheric aerosol contamination of the surface reflectance and EVI, rather
than a true vegetation response. Some of the possible differences for
discrepancies between the Saleska et al. paper and the Samanta et al. papers
may be due to different MODIS Collections (the former study was conducted
with Coll. 4, while the more recent study with Coll. 5). Other causes of differing
results may be due to processing methods, statistical methods, and
interpretations. For example, the 2 studies may be based on different sets of
hypotheses, and hence conclusions.

6
 Figure 4. 2005 TRMM-based precipitation anomaly (left) and corresponding MODIS
EVU anomaly (right), with anomalies expressed as standard deviations. Contrary to
model predictions, short term drought caused an increase in EVI over 34% of the
drought affected area. Forests may be adapted to drought to take advantage of
increased sunlight (Saleska et al. 2007).

2. Methods
We analyzed the spatio-temporal tropical forest responses to seasonal and
interannual drought events with 9 years of MODIS enhanced vegetation index
(EVI) satellite data and 10 years of TRMM precipitation data (3B43 monthly
product at 25 km resolution). We used the 1 km and CMG versions of MODIS
EVI (MOD13A2 and MOD13C1) and used the quality assurance (QA) filters to
screen for and remove cloud and aerosol contaminated pixels, as in Saleska et
al. (2007). To be conservative, we aggregated the 16-­‐day C5 EVI product to a
quarterly anomaly time series according to the restrictive method of Samanta et
al. (2010). EVI and precipitation anomalies are defined relative to the
2000-­‐2008 mean (excluding 2005). The 2005 drought area is defined as those
forested areas within the Amazon basin where July-­‐Aug-­‐Sept precipitation
anomalies < -­‐1 standard deviation relative to the 1998-­‐2006 mean (excluding
2005).

3. Results and Discussion

We tested a wide variety of methods to process the MODIS EVI time series
data and found consistent patterns throughout our analysis, further confirming
the Amazon green-up hypothesis. All datasets, time periods, and methods of
analysis show that valid pixels in the drought region exhibited a statistically
significant anomalous increase in EVI, and MODIS EVI anomalies of the 2005
drought remain statistically significantly skewed towards greenness.

7
Figure 5. (A) Collection 5 EVI anomalies for July-Sept. of 2005 for the Amazon basin,
with QA filtering of 1km product (MOD13A2) by the method used in both Saleska et al
(2007) and Samanta et al (2010). Gray pixels contain no valid 2005 data and/or
insufficient valid data to calculate a mean and standard deviation across other years;
(B) same as in A, but aggregated according to the more restrictive method of Samanta
et al, which discards a majority of all valid observations in the aggregation. Gray pixels
(which cover most of the basin) have one month or more (out of 21 in the 7-­‐year time
series) without a valid observation; (C, D): EVI anomalies (as on left), but masked to
include only forested areas within the basin that were in the 2005 drought area (where
July-­‐Sept precipitation anomalies < -­‐1 standard deviation relative to the 1998-­‐2006
mean, excluding 2005). (insets to C,D) Corresponding histograms of EVI standardized
anomaly frequencies of valid pixels in the 2005 drought area (credit, Kamel Didan).

The method of Samanta et al. (2010) essentially eliminates much of the drought
region with stricter criteria for the time series pixels to be valid, but even so, a
disproportionate number of remaining pixels are anomalously green. We further
show that the full time series shows that the strongest positive EVI anomaly of
the entire satellite record corresponds to the strongest drought of the entire
record (2005). This observed time series stands in strong contrast to model
simulations (Fig. 6).

8
 Figure 6. (A) Model simulations, by the Hadley Center, of forest photosynthesis and
precipitation in central Amazonia in years relative to a hypothetical modeled drought
(Jones et al., 2001); (B) Satellite observations (from the third-­‐quarter in each year), of
the percent of valid pixels in the area defined by the 2005 drought with EVI anomalies
exceeding +1 standard deviations, and (lower panel) the percent of pixels in the area
defined by the 2005 drought in precipitation deficit (i.e. with TRMM precipitation
anomalies more than 1 standard deviation below the mean). Upper right panel from
Samanta et al. (2010) with the yellow point added from this analysis.

Separate tests were made to determine whether it was possible for aerosols to
have contaminated our results. We found that even in the worst-­‐case periods,
surface reflectances flagged as contaminated by high aerosols hardly
intersected the drought region at all, and played no role in the outcome of this
analyses. Thus, high aerosols were, for the most part, not in the drought area.
We further found high aerosol resistance in the EVI through the use of the blue
band.
In conclusion, the patterns of anomalous greening were found to be robust
across different versions and processings of the MODIS dataset, and across a
range of analysis methods. If drought had the expected negative effect on
canopy photosynthesis, it should have been especially observable during
extreme drought events, when anomalous interannual drought coincided with
the already seasonally low precipitation. The observations of intact forest
canopy “greenness” in the affected areas, however, are dominated by a
significant increase, not a decline.

9
4. Discussion
The extraction of meaningful forest canopy information from coarse resolution
satellite data is severely limited by clouds (primarily in the wet season) and by
aerosols (primarily in the dry season). Despite this limitation, there remains
sufficiently high quality pixels to derive meaningful biophysical information. In
the highly restrictive and strict quality criteria of Samanta et al. (2010), 33% of
the Amazon basin forests exhibited the highest quality pixels over a span of > 8
years. Thus, one third of the population of pixels were useful for interpretation of
vegetation dynamics and responses to seasonal and interannual drought
periods. Since more than two-thirds of the Amazon basin were unobservable
due to pixel contamination, Samanta et al (2010) classified such areas an non-
green, therefore satisfying their perceived hypothesis of showing that the
Amazon basin did not green up. However, the same data and results are
interpreted as ‘greening’ in the Amazon by Saleska et al. since a statistically
significant proportion of the pixels that were observable (25-35%) did green up,
when in fact it was hypothesized that Amazon forests should have browned-
down as the models show and predict. By this reasoning, the large proportion
of unobservable area does not become classified as green or non-green, since
the satellites could not accurately image such areas. Lastly, between the two
sets of studies, there is the added factor of mixing of Coll. 4 and Coll. 5 MODIS
data. One cannot readily separate the ‘greening’ differences from the 2 studies
from the myriad of other changes that occurred between the 2 Collections,
including significant changes to the surface reflectance product, the primary
input to the MODIS vegetation indices.
We believe that an important goal of remote sensing in the tropics is to devise
statistically meaningful schemes to retrieve information of ecosystem properties,
processes, and phenology. However, coarse resolution satellite imagery would
still suffer from the complexities in separating biophysical from biochemical
canopy factors as well as how to interpret dynamic signals from numerous
species with asynchronies in their seasonal patterns. As an example,
enhanced tree mortality associated with prolonged droughts may result in an
increase in undergrowth and herbaceous species that may appear as overall
enhanced greening in both moderate and coarse resolution data.

5. References
Asner, G.P., Townsend, A.R., and Braswell, B.H., 2000, Satellite observation of
El Niño effects on Amazon Forest phenology and productivity. Geophysical
Research Letters, 27: 981-984.
Betts, R.A, Cox, P.M, Collins, M., et al., 2004, The role of ecosystem-
atmosphere interactions in simulated Amazonian precipitation decrease and
forest dieback under global climate warming. Theoretical and Applied
Climatology, 78:157-175.
Botta, A., Ramankutty, N., and Foley, J.A., 2002, Long-term variations of

10
climate and carbon fluxes over the Amazon basin. Geophysical Research
Letters, 29:1319.
Doughty, C. E., and Goulden, M.L., 2008, Seasonal patterns of tropical forest
leaf area index and CO2 exchange, Journal of Geophysical Research, 113,
G00B06, doi:10.1029/2007JG000590.
Garrigues, S., Lacaze, R., Baret, F., 2008, Validation and intercomparison of
global Leaf Area Index products derived from remote sensing data. Journal of
Geophysical Research, 113:doi:10.1029/2007JG000635.
Huete, A., 1988, A soil-adjusted vegetation index (SAVI). Remote Sensing of
Environment, 25:295.
Huete, A.R., Didan, K., Shimabukuro, Y.E., et al., 2006, Amazon rainforests
green-up with sunlight in dry season. Geophysical Research Letters,
33:L06405.
Huete, A.R, Restrepo-Coupe, N., Ratana, P., et al., 2008, Multiple site tower
flux and remote sensing comparisons of tropical forest dynamics in Monsoon
Asia. Agricultural and Forest Meteorology, 148:748-760.
Hutyra, L.R.. Munger, J.W., Saleska, S.R., et al., 2007, Seasonal controls on
the exchange of carbon and water in an Amazonian rainforest. JGR-
Biogeoscience, 112, G03008, doi:10.1029/2006JG000365.
Jiang, Z., Huete, A.R., Didan, K., Miura, T., 2008, Development of a 2-band
Enhanced Vegetation Index without a blue band. Remote Sensing of
Environment, 112:3833-3845.
Kobayashi, H., and Dye, D.G., 2005, Atmospheric conditions for monitoring the
long-term vegetation dynamics in the Amazon using normalized difference
vegetation index. Remote Sensing of Environment, 97:519-525.
Marengo, J.A., Nobre, C., Tomasella, J., et al., 2008, The drought of Amazonia
in 2005. Journal of Climate, 21:495-516.
Myneni, R.B., Yang, W., Nemani, R.R., et al., 2007, Large seasonal swings in
leaf area of Amazon rainforests. Proceedings of the National Academy of
Science, doi:10.1073/pnas.0611338104.
Phillips, O.L, Aragão, L.E., Lewis, S.L., et al., 2009, Drought sensitivity of the
Amazon rainforest. Science, 323:1344-1347.
Saleska, S. R., Didan, K., Huete, A.R., and da Rocha, H.R., 2007, Amazon
forests green up during 2005 drought. Science, DOI: 10.1126/science.1146663.
Samanta, H., Ganguly, S.A., Hashimoto, Y. et al., 2010, Amazon forests did not
green-up during 2005 drought. Geophysical Research Letters, 37, L05401,
doi:10.1029/2009GL042154.
Sellers, P., 1987, Canopy reflectance, photosynthesis, and transpiration, II. The

11
role of biophysics in the linearity of their interdependence. Remote Sensing of
Environment, 21:1.
Tian, H., Melillo, J.M., Kicklighter, D.W., et al., 1998, Effect of interannual
climate variability on carbon storage in Amazonian Ecosystems. Nature, 396:
664-667.

12