Simulations of tropical rainforest albedo : is canopy wetness important ?

Accurate information on surface albedo is essential for climate modelling, especially for regions such as Amazonia, where the response of the regional atmospheric circulation to the changes on surface albedo is strong. Previous studies have indicated that models are still unable to correctly reproduce details of the seasonal variation of surface albedo. Therefore, it was investigated the role of canopy wetness on the simulated albedo of a tropical rainforest by modifying the IBIS canopy radiation transfer code to incorporate the effects of canopy wetness on the vegetation reflectance. In this study, simulations were run using three versions of the land surface/ecosystem model IBIS: the standard version, the same version recalibrated to fit the data of albedo on tropical rainforests and a modified version that incorporates the effects of canopy wetness on surface albedo, for three sites in the Amazon forest at hourly and monthly scales. The results demonstrated that, at the hourly time scale, the incorporation of canopy wetness on the calculations of radiative transfer substantially improves the simulations results, whereas at the monthly scale these changes do not substantially modify the simulated albedo.

is very significant especially in the dry season of that region.In this context, a good representation of albedo by climate models is essential to correctly address the problem of changing the tropical deforestation climate.
However, Berbet and Costa (2003) also verified that even a complex, state-of-the-art, land surface scheme coupled to a climate model is unable to correctly reproduce details of the seasonal variability of the albedo of tropical rainforests, although it reproduces well the annual mean and some aspects of the seasonal variability.This indicates that there is still much to be learned -and incorporated into models -about the sources of variation of the albedo at both hourly and monthly time scales.
At the monthly time scale, Culf et al. (1995) analyzed the surface albedo at forest sites in Amazonia.The authors reported that the albedo seasonality at these sites is not related to changes on the solar elevation angle nor to cloudiness, but suggested a relationship with soil moisture.Although soil moisture affects ground albedo, changes in the forest albedo are most likely related to soil moisture-correlated variables: smaller soil exposure, darker leaves (associated with the leaf water potential) and higher canopy wetness (Berbet and Costa 2003).Canopy wetness, in particular, is a strong candidate for changing canopy albedo because the reflectance of liquid water varies depending on the wavelength, between 0-5%, much lower than the 12-13% usually measured above tropical rainforest canopies.Hence the presence of liquid water on the canopy increases the absorption of solar radiation, reducing the canopy overall albedo.
Here, we investigated the role of canopy wetness in the simulated albedo of a tropical rainforest.In this study we have run simulations using three versions of the land surface/ecosystem model IBIS: the standard version with the original calibration used by Delire and Foley (1999), the same version recalibrated to fit the data of albedo on tropical rainforests, and a modified version that incorporates the effects of canopy wetness on calculated surface albedo.The next section describes the sites and instrumentation that measured the albedo data used here to validate the model simulations.Sections 3 and 4 describe the IBIS model, the modifications to incorporate canopy wetness and the experiment design, whereas Section 5 presents and discusses the simulation results.

SITES, INSTRUMENTATION AND DATA
Field data used in this paper were measured at three sites in the Amazon during the ABRACOS (Anglo Brazilian Amazonian Climate Observation Study) and LBA projects (Large-Scale Biosphere-Atmosphere Experiment in Amazonia).Ducke and Cuieiras (K34) Reserves are reserves of forest protected by INPA (Instituto Nacional de Pesquisas da Amazônia) located about 25 km northeast and 70 km north of Manaus, respectively.These sites are surrounded by undisturbed forests for at least 5 km.Jaru Reserve is an IBAMA (Instituto Brasileiro do Meio Ambiente e dos Recursos Naturais Renováveis) forest reserve and is located about 80 km north of Ji-Paraná (Fig. 1).
At the Cuieiras Reserve a piranometer (Kipp & Zonen, Delft, Netherlands), connected on a datalogger model (21X, Campbell Scientific) measured incident and reflected solar radiations at each minute, storing the averages at every 20 minutes.For the remaining sites, incident and reflected solar radiations were measured using two solarimeters (Kipp and Zonen, Delft, the Netherlands).These instruments are part of an automatic weather station (Didcot Instruments, Abingdon, UK) connected to a datalogger (CR10, Campbell Scientific, Shepshed, UK), and hourly-averaged data were recorded.
We used incident and reflected solar radiations and precipitation data collected from June 1999 to September 2000 at Cuieiras Reserve, from January to December 1995 at Ducke Reserve, and from January to December 1993 at Jaru Reserve.Data used in this study are available on-line through www.cptec.inpe.br/abracos/available.htmland http://lba.cptec.inpe.br/beija-flor.

DESCRIPTION OF THE IBIS MODEL
To simulate the diurnal albedo of an Amazon tropical forest we used the Integrated Biosphere Simulator-IBIS (Foley et al. 1996).It includes representations of land surface processes, like energy, water and momentum exchange among the soil-vegetation-atmosphere system, canopy physiology, vegetation phenology, vegetation dynamics and terrestrial carbon balance.Originally, IBIS was globally calibrated (more details in Delire and Foley 1999, Kucharik et al. 2000), and since then it has been used in several studies of the biosphere-atmosphere interactions in Amazonia (Costa and Foley 2000, Botta and Foley 2002, Foley et al. 2002, Berbet and Costa 2003).
One of the processes simulated by IBIS and specially important in this study is the exchange of solar radiation among the soil-vegetation-atmosphere system.Solar radiation transfer is calculated following the twostream approximation, with separate calculations for direct and diffuse radiations in both visible and nearinfrared bands.The canopy radiative transfer code of IBIS is standard in the literature (Norman and Jarvis 1975, Sellers 1985, Sellers et al. 1986, Pollard and Thompson 1995, Bonan 1996, Oleson et al. 2004).In this study we modified the IBIS canopy radiation transfer code to incorporate the effects of canopy wetness on the vegetation reflectance.Although IBIS also calculates the canopy wetness, in this work the parameters ω (scattering coefficient, eq. 3 in Sellers 1985; eq.11 in Sellers et al. 1986;eq. 3.5 in Oleson et al. 2004), β (upscatter parameter for diffuse radiation, eq. 3 in Sellers 1985; eq.11 in Sellers et al. 1986;eq. 3.6 in Oleson et al. 2004) and β o (upscatter parameter for direct radiation, eq. 4 in Sellers 1985; eq. 12 in Sellers et al. 1986;eq. 3.7 in Oleson et al. 2004) are modified to include the radiative effects of canopy wetness, according to Equations 1 to 4: where f wet is the total wet (water and snow) fraction of the canopy ( f wet = f water + f snow ), f water is the fraction of the canopy wet by liquid water, and f snow is the fraction of the canopy wet by snow.The superscripts dry, water and snow denote dry, wet by water and wet by snow canopies, respectively.ν is the ratio of the scattering coefficients of the canopy surfaces wet by water and dry canopy surfaces individually applied to leaves and stems.

EXPERIMENT DESIGN
In this study we conducted three simulations of the diurnal surface albedo for each of the three Amazon forest sites using the off-line IBIS version, as follows: 1) DF99: for reference to previous studies, this simulation uses the set of optical parameters used by Delire and Foley (1999); 2) DC i , the dry-canopy (control) simulations: use the original code without modifications to incorporate the effects of wetness on canopy reflectance, and a set of leaf optical parameters to better fit the simulated results to the experimental data of the studied sites, minimizing the RMSE between the observed and simulated albedos.The subscript i may be equal to M for the Manaus-nearby sites (Ducke and Cueiras Reserves), or J for Jaru Reserve.
3) WC i , the wet-canopy simulations: similar to DC i , but including the modifications described in Equations (1) through ( 4).
The terms dry-canopy (DC) and wet-canopy (WC), when referring to versions of the IBIS code, denote only the status of the canopy during the radiative transfer calculations.It should be noted that both versions simulate the interception of water by the canopy, and the evaporation and dripping of the canopy-stored water.
To calibrate the DC simulations, we initially performed a sensitivity analysis of the simulated albedo to several optical parameters of the canopy.The most sensitive parameters are the upper and lower canopy leaf orientation (χ leaf-up and χ leaf-lo , respectively), and upper and lower canopy visible and near-infrared (NIR) leaf reflectance (α  I shows, for the DC simulations, the parameter combinations that provide the best adjustment of the diurnal albedo without considering the effect of canopy wetness.Next, this set of parameters is added to the WC version of the code, and the parameters f wetmax (maximum fraction of water cover on two-sided leaf), τ drip (decay time for intercepted liquid drip off) and ν (ratio of the scattering coefficients of the canopy surfaces wet by water and dry canopy surfaces individually applied to leaves and stems) are calibrated.

RESULTS AND DISCUSSION
The three IBIS versions are run for each of the three sites.Figures 2 to 4 show the diurnal profile of the surface albedo for Cuieiras, Ducke and Jaru Reserves, respectively, for selected days of the year.To facilitate the interpretation, the selected days represent either no-rain days -charts on the left side (a, b, c) of the figuresor days with a single daytime rainfall event -right side (d, e, f) of the figures.Figures 2 and 3 show results for the Manaus (M) set of parameters, while Figure 4 shows results for the Jaru ( J ) set of parameters, according to Table I.
The DF99 simulations overestimate the surface albedo, particularly when the zenith angle is high (Figs.(DC M and DC J ) fit better to the observed data in days without precipitation occurrence (left side of Figs. 2 to 4), but the DC version of the model does not represent the observed albedo drop during precipitation events (right side of Figs. 2 to 4).However, WC simulations show that the modified version of IBIS is able to reproduce the considerable decrease in surface albedo during precipitation hours (right side of Figs. 2 to 4).This reduction in the surface albedo is consistent with an increase of the absorption of the incident solar radiation by the liquid water deposited on the leaves, increasing solar radiation absortance and reducing solar radiation reflectance.
Table II shows the statistics of observed and simulated surface albedos for the precipitation hours and for the entire time series at three sites in the Amazon rain- Statistics of observed and simulated surface albedos for the entire time series and for the precipitation hours at three sites in the Amazon rainforest (Cuieiras, Ducke and Jaru reserves), for the simulations based on the DF99 calibration, and for the new calibration using the dry-canopy (DC) and wet-canopy (WC) versions of the model.X is the average albedo, ε is the mean relative error, and RMSE is the root mean square error.forest (Cuieiras, Ducke and Jaru reserves), for the simulations based on the DF99 calibration, and for the new calibration using the dry-canopy (DC) and wet-canopy (WC) versions of the model.Mean relative error (ε) and root mean square error (RMSE) are defined according to Equations 5 and 6:

Entire time series
where X s and X o are simulated and observed albedos, and n is the number of data points.Given our interest here in the effects of canopy wetness on the simulated rainforest albedo, we analyzed the data during the "precipitation hours" apart from the entire data (Table II).
For the entire time series, for the two sites near Manaus, the new calibration (DC M ) represents a considerable improvement over the DF99 calibration, lowering the mean relative error by an order of magnitude, and the RMSE by nearly half.The inclusion of the effects of canopy wetness on the radiative transfer code further reduces ε and RMSE, which is consistent with the results shown in Figures 2 to 4. The mean errors involved in the simulations are within the acceptable range for the studied sites, with RMSE being typically smaller than 0.08 (Stroeve et al. 2005).Looking for a single parameterization for Amazonia, we also tested the Manaus parameters at the Jaru site.However, it turns out that the average of the observed albedo at the Jaru site is much higher than at the Manaus sites, and the parameters used for Manaus are not suitable to Jaru.We then selected a new set of parameters for the Jaru site (DC J ).
Results for the Jaru site, using a site-specific calibration, showed an improvement similar to the one obtained at the Manaus sites.
Although the WC simulation shows an improvement in the simulation statistics for the entire time series, its true effect is best seen in the analysis of data during the precipitation hours.Initially, we could verify that the DF99 and DC simulation results during the precipitation hours are much worse than the equivalent result for the entire time series, which by itself indicates that a precipitation-related process is missing in the model.In addition, the WC simulations for the precipitation hours improve considerably the results when compared to the control (DC), with ε dropping by an order of magnitude and RMSE decreasing by about one-fifth.These results demonstrate that, at the hourly time scale, the incorporation of canopy wetness on the radiative transfer calculations substantially improves the simulation results, in particular when the canopy is wet, but also brings an improvement to the simulation of the entire period.We also compared the model results at the monthly time scale for the three sites studied (Fig. 5).In all cases, the WC albedo is smaller than the DC albedo, but the changes introduced did not substantially modify the simulated albedo.As seen in Figures 2 to 4, because of evaporation and dripping, the effect of canopy wetness on the surface albedo was restricted to the duration of a rainfall event plus one or two hours.
Although Amazonia is one of the rainiest climates on Earth, the frequency of rainfall events (7.1% at Ducke, 14.2% at Cuieiras and 17.0% at Jaru) is relatively low for a more significative effect of canopy wetness on albedo seasonality.
An exception is observed at the Ducke Reserve (Fig. 5b), where a more pronounced drop in the WCsimulated albedo was observed in April.Even in this extreme case, the simulated decrease in monthly albedo accounts for less than half of the observed change, which led us to conclude that canopy wetness alone is not sufficient to represent correctly the seasonal variability of the albedo of a tropical rainforest.

CONCLUSIONS
The goal of this work is to study the effect of canopy wetness on the simulated albedo of a tropical rainforest.Simulations were run using three versions of the IBIS model: the standard version, the same version recalibrated to fit the data of albedo on tropical rainforests, and a modified version that incorporates the effects of canopy wetness on surface albedo, for three sites in the Amazon forest in both hourly and monthly time scales.
The incorporation of canopy wetness on the radiative transfer calculations improves the simulation results at the hourly time scale, reproducing the observed decrease in surface albedo during the precipitation hours, when the canopy is wet.Although the canopy wetness has an important effect, this effect is restricted to when the canopy is actually wet, which is a relatively short period of time at monthly or longer time scales.Therefore, the changes introduced are not sufficient to substantially improve the representation of albedo seasonality.
While these results exclude the role of canopy wetness as a main source of seasonal variability of the albedo on tropical rainforests, this study narrows the choice of sources of albedo seasonal variation.Following the discussion of Culf et al. (1995) and Berbet and Costa (2003) on the subject, we recommend that future studies investigate the role of photoinhibition and leaf water potential on the seasonality of the albedo on tropical rainforest.The clear definition of these roles, and their incorporation into climate models will eventually allow us to do much more detailed studies of the climatic effects of tropical deforestation.
Leaf VIS-up , α Leaf VIS-lo , α Leaf NIR-up and α Leaf NIR-lo , respectively).The model is run several times with different combinations of the parameters above.Table

Fig. 2 -
Fig. 2 -Diurnal variation of the simulated and observed surface albedo in the Cuieiras Reserve for the selected days.

Fig. 3 -
Fig. 3 -Diurnal variation of the simulated and observed surface albedo in the Ducke Reserve for the selected days.

Fig. 4 -
Fig. 4 -Diurnal variation of the simulated and observed surface albedo in the Jaru Reserve for the selected days.

Fig. 5 -
Fig. 5 -Monthly profile of the observed and simulated surface albedos, and monthly precipitation and frequency of rainfall events at three sites in the Amazon rainforest: (a) Cuieiras Reserve, from June 1999 to September 2000, (b) Ducke Reserve, from January to December 1995, and (c) Jaru Reserve, from January to December 1993.