Seasonal coupling of canopy structure and function in African tropical forests and its environmental controls

,


INTRODUCTION
Tropical forests have the highest terrestrial biodiversity on the Earth (Dirzo and Raven 2003), contain around 25% of the carbon in the terrestrial biosphere (Bonan 2008), and intact tropical forests collectively possess the largest annual rate of carbon sequestration among all terrestrial ecosystems (Pan et al. 2011).Tropical forests are also strongly coupled to the climate system, both responding to climate variability (Nemani et al. 2003) and regulating the climate system through physical, chemical, and biological processes that affect planetary boundary layer dynamics, the hydrologic cycle, and atmospheric composition (Bonan 2008, Richardson et al. 2013).Nevertheless, many aspects of tropical forest dynamics remain poorly understood.The estimated size of the tropical forest carbon sink since 1990 is more uncertain than for any other forest biome (Pan et al. 2011), and models give a wide range of projections for the tropical forest response to climate change (Sitch et al. 2008).A particular gap in our understanding pertains to the seasonality, or phenology, of tropical forests (Saleska et al. 2009, van Schaik et al. 1993), which is exemplified by the lack of ground observation (Wright and van Schaik 1994) as well as the struggle to accurately simulate the seasonality of tropical forests by dynamic vegetation models (Kim et al. 2012).These uncertainties are likewise evident in the weak process level understanding of the seasonal variation in land-atmosphere exchange (Fisher et al. 2009).
The phenological development of tree species is characterized by distinct seasonal phases of bud burst, leaf flushing, flowering, senescence and dormancy (Calle et al. 2010).The timing of these events fulfills multiple objectives in a plant's life cycle, including matching resource demand with availability and sychronizing with pollinators (Yeang 2007a, Wright andvan Schaik 1994), while working within the constraints imposed by environmental extremes (e.g., cold temperatures, water deficit, light limitation).Phenological changes in vegetation are also associated with seasonal variation in land-atmosphere gas exchange, including photosynthesis and transpiration, due to the dependence of these fluxes on leaf area and its physiology.For tropical species that occupy a weakly seasonal environ-ment, the physical changes in canopy structure and physiological changes in canopy function are not necessarily synchronized in time and may have different top-down controls (Doughty andGoulden 2008, Huete et al. 2008).While there is a long history of studying rhythmic growth in individual tropical species (Halle ´et al. 1978, Reich 1995), there remains a gap in the understanding of how leaf display is related seasonally to leaf photosynthesis, and what the large-scale controls of this synchrony are across a range of climates and vegetation types in the tropics.Hereafter, we refer to the canopy leaf biomass as ''vegetation canopy structure'', and the canopy photosynthetic activities as ''vegetation canopy function''.
It is widely thought that environmental cues trigger changes in leaf biomass (Halle ´et al. 1978), though internal regulation of plants may play a role in highly aseasonal tree species (Reich 1995, Williams et al. 1997).In the tropics, the two most important environmental cues relate to radiation and water (van Schaik et al. 1993).In brief, where rainfall seasonality is pronounced, dry-season drought primarily constrains phenology (Reich and Borchert 1984); where water resources are ample, sunlight plays a more important role in triggering transitions in phenological phases (Wright and van Schaik 1994).However, the details of how these environmental constraints ultimately impact phenology are still controversial in tropical forests (van Schaik et al. 1993), and highly species-dependent (Williams et al. 1997).
For example, the impact of radiation has been hypothesized to be mediated either through daylength (Rivera and Borchert 2001, Rivera et al. 2002, Elliott et al. 2006, Williams et al. 2008), timing of sunrise or sunset (Borchert et al. 2005), or insolation thresholds (van Schaik et al. 1993, Wright and van Schaik 1994, Yeang 2007a, 2007b, Calle et al. 2009, 2010).The insolation threshold pertains to solar radiation intensity at the top of the atmosphere (i.e., shortwave downward radiation at the top of atmosphere, ''SWTOA'' hereafter).SWTOA has two peaks over the tropical equator on the equinoxes, and these two peaks in SWTOA coincide with semi-annual tropical forest flowering and leaf flushing found from field observations (Yeang 2007a, Wright and van Schaik 1994, Calle et al. 2009, 2010).Because SWTOA is related to the direct sunlight in the clear sky at midday, which can be perceived by plants, it is proposed that SWTOA is a strong tropical phenology trigger.Mechanistic evidence for this hypothesis is provided by the finding that strong insolation plays a direct role in the transcription of genes that influence floral/leaf development (Jackson 2009).However, maximum SWTOA control on tropical leaf phenology has not been verified at large scales, mostly because no suitable measurements have been available.It is worth noting that SWTOA is different from shortwave radiation at the top of canopy (SWCNP), which is diminished from SWTOA by clouds and aerosols in the atmosphere.
Other radiation-related hypotheses have been proposed and tested in the past.The daylength hypothesis fails to explain the seasonality of evergreen forests in the tropical equator since there is little variation in day-length at equator regions (Renner 2007).The hypothesis of timing of sunrise or sunset proposes that the seasonal shifts in these events as a result of the Earth's axial tilt and its elliptic orbit trigger the synchronous flowering/leaf-flushing at the equator (Borchert et al. 2005).The relevance for this hypothesis in the tropics has been questioned, because it would require plants to sense subtle cyclical shifts (30 minutes over the course of a year) that arise between solar time and an external reference chronometer time (Renner 2007, Yeang 2007a, 2007b).
The water-related hypotheses are also controversial because phenological patterns vary widely within the same species or among different species in tropical forests (Williams et al. 1997).Many tropical forests have well-established drought-adaptive mechanisms linking the seasonality of growth with the seasonality of water availability.These strategies include deploying deep roots to access soil water reserves, which potentially allows them to grow in the dry season (Nepstad et al. 1994, Dawson 1996, Borchert 1998, Meir et al. 2009, Markewitz et al. 2010), or by timing leaf flushing to precede the onset of the rainy season (Williams et al. 2008).
Remote sensing has provided unique opportunities to monitor large-scale temporally-continuous vegetation dynamics (Chambers et al. 2007).Vegetation Indices (VI) from optical remote sensing have been widely used in vegetation monitoring (Tucker et al. 2008), typically using some variation of a ratio or difference between reflectance in the red (R) and near-infrared (NIR), such as the Normalized Difference Vegetation Index: NDVI ¼ (NIR À R)/ (NIR þ R).In this study we used a related VI, the Enhanced Vegetation Index (EVI), which is in many ways similar to NDVI but includes a term that alleviates atmospheric effects of aerosols released in biomass burning (Huete et al. 2002).The difference in reflectance in R and NIR differentiates green vegetation from soil and non-photosynthetic vegetation.At the canopy level, EVI is linear to the fraction of absorbed photosynthetically active radiation (fAPAR), which is directly related to canopy photosynthetic capacity, but is not a reliable predictor of leaf area index or biomass, especially for vegetation with thick and closed canopy (Sellers 1985(Sellers , 1987(Sellers , 1992)).This relationship can also be confounded by soil reflectivity in sparse canopies (Sellers 1987).At landscape scales, EVI is a composite variable influenced by leaf biochemical properties, canopy structure and vegetation fractions (Glenn et al. 2008, Guan et al. 2012).For closed, dense canopies with little soil contribution to reflectance, such as those in tropical forests, EVI has limited sensitivity to extremely high leaf biomass (esp.those in the tropical evergreen forests) (Doughty and Goulden 2008), and its variation can be attributed to changes in canopy photosynthetic capacity for these forests (Huete et al. 2008).
Microwave remote sensing (both passive radiometry and active radar) has gradually gained attention for vegetation monitoring during the past two decades (Waring et al. 1995, Hardin and Jackson 2003, Frolking et al. 2005, 2006, 2012).Microwave instruments are generally much less sensitive to atmospheric aerosols and moisture than optical sensors (Ulaby et al. 1982), and insensitive to seasonal variations in incoming solar radiation (Frolking et al. 2011).The energy in microwave backscattering from terrestrial landscapes is determined by surface dielectric properties, modulated by microwave frequency (or wavelength), polarization and viewing angle.The dielectric property of a landscape is strongly dependent on its liquid water content (Jensen 2000).Different microwave frequencies have been used for different purposes.Large micro-wave wavelengths, such as P-band and L-band, have strong penetration capabilities through the whole crown layer of tropical forests, and the crown layer (mainly branches) and trunk are the major contribution of the observed backscatter.Thus they have been used to estimate total above ground biomass (Toan et al. 1992, Saatchi et al. 2011).As the frequency decreases, C-band and Ku-band capture more backscattering from leaf and twigs (leaf biomass) of the upper canopy of tropical forests (Toan et al. 1992, Waring et al. 1995).Past research has found that for sparse vegetation regions, such as grasslands and savannas, C-band and Ku-band backscattering observations require careful interpretation because soil moisture can modulate the signal (Jarlan et al. 2002).By contrast, in tropical evergreen forests, with closed canopies, C-band and Ku-band backscattering are more strongly related to leaf biomass dynamics, with limited contribution from soil moisture (Frolking et al. 2011).Another advantage of C-band and Kuband is the high revisit frequency (daily or subdaily) of the satellite platforms that carry these sensors.Nevertheless, despite these advantages, there are still relatively few studies applying these data for monitoring the dynamics of tropical forests (Frolking et al. 2011(Frolking et al. , 2012)).
In this paper, we aim to use multiple remote sensing datasets from both optical and microwave platforms to assess the environmental controls on the seasonality of vegetation function and structure in African tropical forests by testing multiple hypotheses that have been proposed from field research.Tropical Africa contains 18% of global tropical forests by area, but has been understudied compared to the Neotropical and Asian tropical forest.The specific science questions we will address in this paper are: (1) What are the spatial and temporal patterns of vegetation phenology in African tropical evergreen forests, tropical deciduous forests and woody savannas?(2) What are the environmental controls that best explain the spatio-temporal patterns of the vegetation phenology in African tropics?We answer these questions by assessing the correlations between satellite-measured vegetation indices with precipitation, SWCNP, SWTOA, day-length and sunrise (or sunset) time.

Data
We use two optical remote sensing datasets and two microwave scatterometer datasets for vegetation monitoring (Table 1, also see all the acronyms in Appendix A).Two optical remote sensing (ORS) datasets are NDVI from GIMMS (based on AVHRR) and EVI from MODIS (MOD13C1, Version 5), which have been widely used for regional and global vegetation applications (Huete et al. 2002, Tucker et al. 2008).Both AVHRR and MODIS sensors are polar-orbiting sun-synchronous satellites.EVI has improved performance over the traditionally-used NDVI, by addressing atmospheric effects as well as improving its sensitivity for high leaf biomass canopy (Huete et al. 2002).Two microwave remote sensing (MVRS) datasets are from QuikS-CAT and ERS-1/-2.QuikSCAT is a Ku-band (13.4 GHz, or 2.1 cm wavelength) active radar with two rotating pencil beam antennas operating in H and V polarizations at an incidence angles of 558 and 468.The Ku-band backscatter coefficient (called ''Ku-dB'' hereafter) product from QuikS-CAT that we use combines ascending and descending overpass together for a daily composite (Long 2001), and since H-band and V-band have little difference for our analysis, V-band results are shown in the paper.ERS-1/-2 are Cband (5.3 GHz, or 5.7 cm wavelength, called ''C-dB'' hereafter) active radar instruments, and only V-band data are available (http://www.scp.byu.edu/docs/ERS_user_notes.html).
Since both ORS and MVRS data contain random noise from various sources such as electronic noise and geometric positioning, an advanced smoothing scheme (Garcia 2010) was applied to aggregate all good-quality data into half-monthly level (the data filtering is done based on the data quality flags).This smoothing v www.esajournals.orgscheme is fully automated based on a penalized least squares method (Garcia 2010; see Appendix B: Fig. B1).We find that different smoothing approaches do not change the results or conclusions in this work.All the data were aggregated or interpolated to 0.1 degree (;10 km) and halfmonthly.We only show the result of EVI for ORS, and Ku-dB for MVRS because EVI and Ku-dB are representative of ORS and MVRS performances.Results of other data can be found in the Appendix B.

Study area
This study focuses on the regions dominated by tropical evergreen forests (58 N-78 S in latitude and 108 E-308 E in longitude), tropical deciduous forests and woody savannas (regions northern of 58 N and southern of 78 S).We use the annual LAI mean value of 1.0 (from SEVERI, see Table 1) as a threshold to delineate the regions of tropical forests with other ecosystems, which is generally consistent with land cover classifications from the GLC2000 Africa product (Mayaux et al. 2003, also see Fig. 1).

Methods
1. Power spectra and phase analysis.-Fouriertransforms were used to extract the power spectra at the per-pixel level.Each time series was normalized by its standard deviation before the Fourier transform for the purpose of intercomparison (Guan et al. 2011).For those regions with a pronounced single annual cycle (power spectra of single annual cycle .0.8), a sinusoid function, is fitted to the time series T(t) to estimate the optimal parameters " T, A and h simultaneously using the trust-region-reflective algorithm (Coleman andLi 1994, 1996), where " T refers to the mean value, A the magnitude of variation, and h the phase.
2. Correlation analysis.-Spearmancorrelation coefficients were calculated between the time series of vegetation indices in order to investigate their consistency.Spearman correlations coefficients were also calculated between the time series of vegetation indices and environmental drivers.
3. Hovmo ¨ller diagram.-Hovmo¨ller diagrams (longitudinal averages) are used to characterize the mean seasonal cycle along a latitude range.We focus on a typical tropical transect averaged over 1 degree in longitude (208 E-218 E) and spanning 30 degrees in latitude (108 N-208 S), with exclusion of pixels belonging to wetland.The pattern of Hovmo ¨ller diagrams shown here is only weakly dependent on the longitude because the Hovmo ¨ller diagrams centering on any of the longitudes from 198 E to 288 E show a similar seasonality (see Appendix B: Fig. B5).For better visual illustration, we show the Hovmo ¨ller diagrams of both the percentile value and the Hovmo ¨ller diagrams based on the data raw values.

Power spectra and phase analysis
From the power spectra and phase analysis, we find that deciduous forests and woodland in Africa almost all have pronounced single annual cycles, shown from both ORS and MVRS data (only single annual cycle of EVI and Ku-dB are shown in Fig. 2, power spectra of other datasets are provided in Appendix B: Fig. B2).The phase information for the single annual cycle regions from EVI and Ku-dB are consistent with each other, with gradients in south-to-north direction in both northern Africa and southern Africa.This power spectra/phase pattern of vegetation is consistent with that of precipitation (Fig. 2), indicating that water is the dominating controlling factor for vegetation seasonality in African tropical deciduous forests and woodlands.However, this does not rule out the possibility that some tropical dry forests may flush their leaves slightly before the rainy season starts because remote sensing can only capture the large-scale pattern rather than taxonomic heterogeneity in phenology.The spectra and phase pattern of SWCNP and SWTOA are very different from those of EVI and Ku-dB in tropical deciduous forests and woody savanna regions, indicating that SWCNP and SWTOA have little control on vegetation seasonality in these regions.
The situation in the tropical evergreen forest regions (38 N-78 S) is more complex.Both EVI and Ku-dB have weaker single annual cycles, and relatively strong double annual cycles in this region (Fig. 2).EVI and precipitation have more similar spatial patterns for the power spectra of double annual cycles, and both are different from Ku-dB, with the largest differences shown in the northwestern part of tropical evergreen forests (i.e., Gabon and Cameroon).

Correlation analysis
Correlation analysis shows that EVI and Ku-dB are strongly positively correlated in tropical deciduous forests and woodlands, but they are  uncorrelated or negatively correlated in tropical evergreen forests (Fig. 3A).Clear boundaries in Fig. 3A distinguish these two different regimes, which match well with the land-cover delineations of tropical evergreen and deciduous forests.EVI is highly correlated with precipitation almost over the whole tropical African region (Fig. 3B).
But Ku-dB shows significantly negative correlation with precipitation in evergreen forests (Fig. 3E).The seasonal coupling between EVI and Ku-dB in deciduous forest and woodland regions has been further illustrated in Fig. 4, where both EVI and Ku-dB have a high positive correlation with precipitation seasonality from latitude north of 58 v www.esajournals.orgN and south of 78 S.This coupling breaks down for tropical evergreen forest regions (between 58 N and 78 S), where Ku-dB and EVI show almost reversed correlations with precipitation.We also find that SWCNP is negatively correlated with both EVI and Ku-dB in almost all of tropical Africa, except for parts of the tropical evergreen forest.

Hovmo ¨ller diagram
The Hovmo ¨ller diagrams of mean annual cycle of different datasets are shown in Fig. 5 (two mean annual cycles are put together for a better visual illustration).We find that EVI generally follows the precipitation seasonality for all the tropical regions ranging from 108 N-208 S (the same for GIMMS-NDVI in Appendix B: Fig. B3).Ku-dB, EVI and precipitation are in phase in tropical deciduous forests and woody savanna regions (from 108 N-58 N and 108 S-208 S).But from the latitude of 38 N to 58 S (Fig. 5, second column for enlarged illustrations), there are pronounced phase shifts between Ku-dB and EVI.The regions of 58 N-38 N and 58 S-108 S exhibit a transitional state from in-phase to outof-phase between the two data sets.Two MVRS datasets (Ku-dB and C-dB) share the similar spatial pattern in the Hovmo ¨ller diagram (Appendix B: Fig. B3), with C-dB showing less temporal variability in the tropics due to its larger wavelength having less sensitivity to leaf biomass and leaf water content.
For tropical evergreen forest regions (38 N-58 S), SWTOA behaves as a precursor of Ku-dB, i.e., the two insolation maximum peaks correspond to the rising times of Ku-dB from the two bottoms in its seasonal evolution.The latitudinal shifts of Ku-dB and insolation are in the same direction for the two cycles over the course of a year.
We also find that SWCNP is out of phase with both EVI and Ku-dB almost across the whole tropical range 108 N-208 S. Seasonality patterns of day-length, sunrise, and sunset (Fig. 5) cannot explain the continuous equatorial transitions in either EVI or Ku-dB.

DISCUSSION
The results show that the seasonalities of ORS and MVRS data are strongly coupled in tropical deciduous forest and woody savannas, which are mainly controlled by precipitation.Fig. 6A and C shows two typical examples of this case (i.e., the time series of interested variables at 10.58 N and 15.58 S, respectively), which clearly demonstrate the precipitation controls on both EVI and Ku-dB trajectories.We find that the seasonality of tropical evergreen forest is weak but exists.ORS and MVRS data show very different seasonality patterns, though both of them have double annual cycles.ORS follows precipitation seasonality in general.Ku-dB data start to rise from minimum values synchronous with the maximum SWTOA (see Fig. 6B), and this correspondent pattern between Ku-dB and SWTOA persists in the tropical evergreen forest regions from 38 N to 58 S (Fig. 5).
SWCNP, day-length, sunrise or sunset variations cannot be attributed as controls on the seasonal evolution of either EVI or Ku-dB.In the tropical Amazon, both radiation and cloudiness have been proposed as candidates of controlling the vegetation seasonal evolution (Saleska et al. 2003, Myneni et al. 2007, Betts and Dias 2010, v www.esajournals.orgBradley et al. 2011), but this is not the case in tropical Africa.Our recent work has found that tropical evergreen forests in Africa have much less wet-season rainfall than their counterparts in the Amazon and Asian islands.This constrains the water storage which can be carried over from wet seasons to dry seasons, and limits the physiological functioning of African tropical forests during the dry season.Thus vegetation functions (e.g., photosynthetic activity or transpiration) follow precipitation rather than other environmental variables.
The different physical properties captured by ORS and MVRS data can explain their distinctive behaviors in tropical evergreen and deciduous forests.ORS represents landscape-integrated vegetation photosynthetic capacity depending on both vegetation fractional cover and leaf photochemical attributes (e.g., chlorophyll content), and leaf photochemical attributes is further determined by both leaf area and leaf-level photosynthetic capacity.For tropical deciduous forests and woody savannas, seasonality of leaf area/biomass and leaf-level photosynthetic activity usually co-vary, i.e., deciduous trees can only have photosynthesis and transpiration when leaves are present.Thus, ORS captures the multiplicative feature of leaf area and leaf-level photosynthetic activity (Guan et al. 2012).However, for tropical evergreen forests characterized with closed canopy, ORS has little sensitivity to leaf area due to the very high leaf biomass throughout the year.The seasonality in ORS is mostly attributed to variation in canopy-level leaf biochemical properties, which strongly tie to plant photosynthetic activity.This can explain why the seasonality of ORS follows precipitation in the tropical evergreen forests, since the soil moisture deficit limits the photosynthetic activities during dry seasons, given that African evergreen forests are still water-limited in general.It is also worth noting that all the ORS data show the similar pattern including an LAI product from SEVERI (Appendix B: Fig. B3).Currently most satellite LAI products are derived from VI or similar reflectance ratios based on visible and NIR spectra.How to interpret these derived LAI products in tropical evergreen forests requires caution because they may not be quantifying the actual leaf area (e.g., Doughty and Goulden 2008).
MVRS data (mainly valid for Ku-dB) represent leaf biomass and leaf water content for the top portion of the canopy, with contribution from soil moisture depending on soil fraction in a pixel.For tropical forests, soil moisture has little contribution for dB since little fraction of bare soil is in the scene (Frolking et al. 2011, Guan et al. 2012).Thus the dominating information of Ku-dB is from leaf biomass.For deciduous forests and woody savannas, leaf area/biomass co-varies with canopy-level photosynthetic activity; this explains why dB strongly correlates and also shares a similar seasonality with ORS in these regions.For tropical evergreen forests, dB, with more sensitivity to leaf biomass, characterizes the leaf phenological evolution to the largest extent.
This study is the first one to use remote sensing to test and confirm the field-based finding (Yeang 2007a, b) that insolation intensity controls tropical evergreen vegetation, i.e., the double annual cycles of leaf-flushing in tropical evergreen forest are synchronous with the annual bi-model insolation intensity.The SWTOA-dB feature is only found at 38 N-58 S because in this range the vegetation types are homogenously tropical evergreen forests (we exclude wetland pixels in our analysis).Beyond this range, tropical evergreen trees are mixed with more deciduous trees, where the microwave signal follows precipitation rather than SWTOA.Thus there is a transitional range beyond the homogenous tropical evergreen forest regions (38 N-58 S).
We interpret Ku-dB as the leaf biomass trajectory, which implies that leaf-flushing synchronizes with the insolation intensity peak.Since the SWTOA almost has no inter-annul variation, Ku-dB signals should also have little inter-annual variation at its two initiation points (corresponding to the two leaf-flushing times).We find there is little inter-annual variation in the Ku-dB initiation points given the noisy condition of dB in the tropics (results not shown here).We also observe that the double cycles of Ku-dB are asymmetric, with the cycle from April to September much stronger than the cycle from November to January.This also confirms the field-based finding by Yeang (2007b), who attributed the unequal solar radiation received at the ground to the cloud covers in these two periods.It is worth noting that there are relatively large uncertainties related to Ku-dB data in the tropical evergreen forests as shown in Appendix B: Fig. B1.We argue that the spatial pattern of Ku-dB after the smoothing approach reveals a continuous spatial structure rather than some spatially random patterns, which implies the existence of an underlying structure in the Ku-dB signals.

Conclusion
In this paper, the seasonality of vegetation functions (i.e., canopy photosynthesis capacity) and structure (i.e., canopy leaf biomass changes) in African tropical forests is investigated using multiple remote sensing datasets from optical and microwave sensors.The results reveal a strong seasonality coupling between vegetation function and structure for tropical deciduous forests and woody savannas, which is controlled by precipitation.Shortwave radiation at the top of canopy, day-length, sunrise or sunset variations have limited control on the seasonal evolution of tropical forests in Africa.For African tropical evergreen forests, seasonality decoupling of vegetation functions and structures are discovered from optical and microwave sensors.Canopy photosynthetic activity observed by ORS follows precipitation seasonality due to the deficit of soil moisture limiting plant biochemical activity during dry seasons.Canopy-top leaf biomass inferred from Ku-dB exhibits two leafflushings over a year, synchronous with the two peaks in insolation intensity, consistent with field observations.A pan-tropical analysis across the world with optical and microwave data is worth conducting to reveal a complete picture of tropical forest phenological patterns and their controls.We further suggest more field campaigns to collect large-scale tropical evergreen forest phenology information in the pan-tropics to verify our findings.

ACKNOWLEDGMENTS
We greatly appreciate all the data providers for their generosity of sharing datasets for scientific research.We are also grateful for the computational support provided by the PICSciE-OIT High Performance Computing Center at Princeton University.The Subject-matter Editor Dr. Epstein and anonymous reviewers provided constructive suggestions and comments to significantly improve the manuscript.Dr. Damien Garcia generously provided the smoothing algorithm with the calculation of confidence interval.K. Guan and E. F. Wood are supported by the NASA grant NNX08AL59G and NASA NESSF fellowship.K. K. Caylor is supported by the NSF grants EAR-847368 and BCS/ESE-1026334.QuikSCAT (2000QuikSCAT ( -2009)).All the Hovmo ¨ller diagrams are based on transects across the latitude from 108 S to 208 N, but the longitudes vary from 188 E to 318 E. Hovmo ¨ller diagrams from 198 E to 278 E show similar patterns for vegetation seasonality as we discussed in the paper for Ku-dB over tropical evergreen forests, and these regions are also homogenous tropical evergreen forest regions.

Fig. 2 .
Fig. 2. The power spectra of single annual cycle (first column) and double annual cycles (second column) for the normalized time series of MODIS EVI, Ku-dB, precipitation, SWCNP and SWTOA.For those regions with pronounced single annual cycle (power spectra of annual cycle .0.8), a sinusoid function T(t) ¼ " T þ Acos(2pt À h) is fitted to the time series to estimate the optimal parameters " T, A and h simultaneously.The phase information h of the corresponding datasets are shown in the third column.The black line indicates regions where mean LAI large than one.

Fig. 3 .
Fig. 3. Spearman Correlation coefficients of mean annual cycles between: (A) EVI and Ku-dB; (B) EVI and precipitation; (C) EVI and SWCNP; (D) precipitation and SWCNP; (E) Ku-dB and precipitation; (F)Ku-dB and SWCNP.Correlation coefficients greater than 0.24 or less than À0.24 satisfy the significance level of P , 0.1, which has been identified as the black lines in the colorbar.

Fig. 4 .
Fig. 4. Longitudinally-aggregated (208 E-308 E) Spearman Correlation coefficients of mean annual cycle for EVI and precipitation (blue line) and Ku-dB and precipitation (red line), across the latitude from 108 N to 208 S. The error bars show the standard deviations of Spearman Correlation within each longitude bin (28).

Fig. 5 .
Fig. 5. First column: Hovmo ¨ller diagrams of mean annual cycle with percentile (or rank) values across the latitude from 108 S to 208 N for: EVI, Ku-dB, rainfall, SWTOA, SWCNP, day-length and sunrise.All the Hovmo ¨ller diagrams are averaged for the pixels from 208 E to 218 E in longitude.Second column: enlarged Hovmo ¨ller diagrams of mean annual cycle with percentile (or rank) values for the latitude from 38 S to 58 N for the same set of variables.

Fig. 6 .
Fig. 6.Mean annual time series of the interested variables for: (A) 108 N; (B) 08; and (C) 158 S. The variables shown includes: EVI, Ku-dB (unit: dB), precipitation (unit: mm/day), and SWTOA at noon (unit: W/m 2 ).The mean annual time series are averaged for a square area with latitude width of 18 (centering at 108 N, 08 S, and 158 S, respectively), and with longitude width of 18 (all centering at 20.58 E).

Fig. B2 .
Fig. B2.The power spectra of single annual cycle (first column) and double annual cycles (second column) for the normalized time series of: GIMMS NDVI, C-dB, day-length, sunrise and sunset.For those regions with pronounced single annual cycle (power spectra of single annual cycle .0.8), a sinusoid function T(t) ¼ "T þ Acos(2pt À h) is fitted to the time series to estimate the optimal parameters " T, A and h simultaneously.The phase information h of the corresponding datasets are shown in the third column.

Fig. B3 .
Fig. B3.Hovmo ¨ller diagrams of mean annual cycle with raw values (first column) and percentile/rank values (second column) for: GIMMS NDVI, MODIS EVI, Ku-dB (dB), C-dB (dB), rainfall (mm/day), SWCNP (W/m 2 ), SWTOA at noon (W/m 2 ), day-length (hours), sunrise (hour of the day) and sunset (hour of the day).All the Hovmo ¨ller diagrams are based on transect across the latitude from 108 S to 208 N, averaging at the longitude from 208 E to 218 E.

Fig. B4 .
Fig. B4.Mean annual time series of the interested variables across the different latitudes from 68 N to 98 S. The variables shown includes: EVI, Ku-dB (dB), precipitation (mm/day), and daily mean SWTOA (W/m 2 ).The mean annual time series are averaged for a square area with latitude width of 18 (centering at the latitudes from 68 N to 98 S), and with longitude width of 18 (all centering at 20.58 E).

Fig. B5 .
Fig. B5.Hovmo ¨ller diagrams of mean annual cycle with percentile (or rank) for Ku-dB fromQuikSCAT (2000QuikSCAT ( - 2009)).All the Hovmo ¨ller diagrams are based on transects across the latitude from 108 S to 208 N, but the longitudes vary from 188 E to 318 E. Hovmo ¨ller diagrams from 198 E to 278 E show similar patterns for vegetation seasonality as we discussed in the paper for Ku-dB over tropical evergreen forests, and these regions are also homogenous tropical evergreen forest regions.

Table 1 .
Datasets used in this study: variable fields, resolution and source.