Ecosystem Modeling and Dynamic Effects of Deforestation on Trace Gas

Forest Ecology and Management 152 62001) 97±117

Ecosystem modeling and dynamic effects of deforestation on trace gas ¯uxes in Amazon tropical forests Christopher Pottera,*, Eric Davidsonb, Daniel Nepstadb, Claudio Reis de Carvalhoc aNASA Ames Research Center, Ecosystem Science and Technology Branch, Moffett Field, CA 94035, USA bWoods Hole Research Center, Woods Hole, MA, USA cEMBRAPA, Amazonia Ocidental, Belem, Para, Brazil Received 21 March 2000; accepted 8 August 2000

Abstract

To improve predictive capabilities for water, carbon, and nitrogen gas ¯uxes in the Amazon region, we evaluated the performance of the NASA±CASA simulation model for tropical ecosystem biogeochemistry against independent ¯ux measurements from two Amazon forest sites located in the Brazilian states of RondoÃnia and ParaÂ. Re®nements of this ecosystem model include stand water balance equations, moisture holding and retention capacity for Amazon soils, and addition of a dynamic deforestation sequence to include land use change as a factor in simulations of tropical ecosystem ¯uxes. Results suggest that model predictions for evapotranspiration and soil water content are consistent with the overall range and seasonal changes in measured values at the two forest sites selected as test cases. The predicted carbon balance from the model implies that relatively undisturbed Amazon forest ecosystems may be large net sinks for atmospheric carbon, with À2 annual net ecosystem production values on the order of 200 g C m per year. Measured ¯uxes of soil N2O for the two Amazon forests closely match our model prediction for the ParaÂ forest, but not for the RondoÃnia site, suggesting that process algorithms controlling nitrogen trace gas ¯uxes, particularly in relatively sandy tropical soils will require further study. In terms of net ecosystem carbon ¯uxes during deforestation and for 2 years afterward, the model predicts that these sites switch from being a net sink for carbon to a substantial source following the large loss of biomass from simulated burning. During crop regrowth simulation in the ®rst year after deforestation, the net source of carbon to the atmosphere is nearly 1.6 kg C mÀ2 per year, a ¯ux magnitude roughly equivalent to 10 years of undisturbed CO2 sink ¯uxes in the Amazon forest. Compared to the primary forest that was cut and burned, predicted changes in soil nitrogen cycling lead to a doubling in annual emissions of

N2O gas during the ®rst year following deforestation, with lower emissions thereafter. Implications for scaling up these model predictions to the Amazon forest region are discussed with reference to necessary improvements in land cover, land use, and soils classi®cations for the area. Published by Elsevier Science B.V.

Keywords: Carbon; Nitrogen; Trace gas ¯ux; Amazon forest; Deforestation

1. Introduction

Deforestation and forest ®res caused by humans can * Corresponding author. Tel.: 1-650-604-6164; fax: 1-650-604-4680. add large amounts of carbon dioxide 6CO2) and nitro- E-mail address: [email protected] 6C. Potter). gen oxides to the airborne concentrations of these

0378-1127/01/$ ± see front matter. Published by Elsevier Science B.V. PII: S 0378-1127600)00593-4 98 C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117 gases. Estimated anthropogenic contributions to the Amazon 6e.g. Keller et al., 1986; Livingston et al., global carbon cycle commonly include a major net 1988; Steudler et al., 1996; Verchot et al., 1999), the ¯ux to the atmosphere 6ca. 1.6 Gt C per year) resulting largest remaining area of relatively undisturbed tro- from land use change in the tropics 6Schimel et al., pical rain forest in the world. These measurement data 1996; Potter, 1999). Quantifying these ¯uxes is critical sets for year-round trace gas ¯uxes are dif®cult and to understand the overall response of the carbon and expensive to collect. Studies at only a few sites in the nitrogen cycles to human perturbation worldwide. Amazon have collected biogenic gas measurements Land use change is also basic to understand the rate together with climate data and ecosystem attributes of increase in radiative forcing of the climate system such as biomass amount and nutrient content, and soil 6i.e. the `greenhouse' gas effect), particularly since biochemistry. Hence, construction of an accurate

CO2 accounts for most of the anthropogenically driven regional emission budget for any biogenic trace gas increase in forcings. from a relatively small data set of measurements Soils of the humid tropical zone are recognized as carries a high degree of uncertainty, especially in a major natural sources of carbon and nitrogen trace gas region as vast and diverse in terms of land cover and emissions to the atmosphere 6Keller et al., 1983, 1986, soil types as Amazonia. The Large Scale Biosphere- 1993; Livingston et al., 1988; Luizao et al., 1989; Atmosphere Experiment in Amazonia 6LBA, 1996) is Keller and Reiners, 1994; Riley and Vitousek, 1995; an international research effort designed to augment Potter et al., 1996a,b, 1998). For instance, nitrous the region-wide data set of trace gas measurements in oxide 6N2O) is an important greenhouse gas and a the years to come. catalyst of stratospheric ozone depletion 6Ramanathan Use of a process-oriented simulation model that et al., 1985; Cicerone, 1987). Levels of N2O are includes major gas ¯ux controllers can facilitate inter- increasing in the atmosphere at a rate of 0.2±0.3% polation of gas ¯ux estimates along regional gradients per year 6Prinn et al., 1990; Khalil and Rasmussen, of interest, and assist in generation of region-wide

1992). Associated with N2O in natural emission bud- model estimates of seasonal ¯ux budgets for water and gets is nitric oxide 6NO), which plays an important trace gases, which may be veri®ed using a series of role in photochemistry and ozone production in the single site measurements. For example, a pre-LBA troposphere 6Logan et al., 1981; Thompson, 1992). regional application study with the NASA Ames NO is also a precursor to nitric acid, a major compo- ecosystem model version of CASA 6Carnegie- nent of acid deposition 6Calvert and Stockwell, 1983). Ames-Stanford Approach) over the Brazilian territory

Atmospheric concentrations of both N2O and NO may 6Potter et al., 1998) has been used to identify the be related to trends in land use, namely conversion of potential importance of eco-climatic gradients of sea- tropical forests and woodlands to pasture and other sonal moisture availability, soil texture, and land use agricultural management. as controls on ecosystem production and associated

Biotic and abiotic factors interact to create condi- soil trace gas emissions 6CO2,N2O, and NO). tions favorable for N trace gas production and emis- In this paper, we report on results from more sions in tropical forest soils. Rapid mineralization of N detailed studies using the daily version of the from decaying organic matter provides abundant sub- NASA±CASA ecosystem model for simulation of strate for biological production of trace gas. At the controls on water, C and N cycling, and trace gas cellular level, the bacterial processes of nitri®cation emissions in applications at two Amazon forest sites, and denitri®cation are generally cited as the principal located in the Brazilian states of RondoÃnia and ParaÂ. trace gas sources in the soil 6Hutchinson and David- The daily version of NASA±CASA can be compared son, 1993). The relative availability of electron donors to relatively rapid changes in measured climate, soil 6commonly soluble carbon compounds from organic conditions, and gas ¯uxes, in a manner not possible matter decomposition) and electron acceptors 6O2 and with a monthly model version 6Potter et al., 1998). The N oxides) determine gaseous end products in soils Amazon sites selected differ in terms of seasonality of 6Davidson, 1991). rainfall, length of the annual dry period, and soil Soil trace gas emissions have been measured pre- properties, which makes the comparison of ecosystem viously at a small number of locations in the Brazilian biogeochemistry and trace gas ¯ux useful within the C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117 99 context of regional assessments for Amazonia 6LBA, example, the fraction of net primary production

1996). Selected forest stand measurements and annual 6NPP), de®ned as net ®xation of CO2 by vegetation, gas ¯ux estimates for soil CO2,N2O, NO, and CH4 is computed on the basis of light-use ef®ciency 6Mon- have been collected at the two sites 6Neill et al., 1995, teith, 1972); new production of plant biomass is 1997; Davidson et al., 2000; Verchot et al., 1999, estimated as a product of intercepted photosyntheti- 2000), which provide a starting point for evaluating cally active radiation 6IPAR) and a light utilization ecosystem model predictions. Although these are two ef®ciency term that is modi®ed by temperature and of the most extensively studied forest sites in the soil moisture. For this simulation study, daily air Amazon region to date, no tower-based gas ¯uxes surface temperature, irradiance, and precipitation were made at these locations. Therefore, available soil together regulate the modeled NPP results, using measurements are based on small chamber-based monthly images the Normalized Difference Vegeta- ¯uxes only, and reported mainly as annual ¯ux tion Index 6NDVI) from the Advanced Very High estimates for comparison to model predictions of Resolution Radiometer 6AVHRR) satellite sensor to the same. estimate changes in leaf cover properties at the land The main objectives of this modeling study were to surface 6Potter et al., 1999). 61) compare predictions from the NASA±CASA simu- For the soil C and N component 6Fig. 1), our design lation model to independent ¯ux measurements avail- remains comparable to a somewhat simpli®ed version able from the Amazon forest sites, in an effort to of the CENTURY ecosystem model 6Parton et al., improve both types of data sets, 62) re®ne the concepts 1992), which simulates carbon and nitrogen cycling and algorithms upon which this generalized ``leaky with a set of compartmental difference equations. In pipe'' scheme for tropical soil trace gas emissions of the CASA model, as with many other ecosystem nitrogen can be built, including effects of deforesta- models, C and N cycling are coupled, with NPP tion on ecosystem biogeochemical cycling, and 63) use considered a useful model driver for N transformation inter-site comparisons to improve the potential for rates. The effect of temperature on litter and soil C and accurate model extrapolation of interannual water, N mineralization ¯uxes was de®ned as an exponential carbon, and trace gas ¯ux predictions over the entire response using a Q10 6the multiplicative increase in Amazon forest region, with integration of satellite soil biological activity for a 108C rise in temperature), data to characterize land surface properties 6Potter with a value of 1.5 for surface litter decomposition and et al., 1998). a value of 2.0 for soil decomposition. In Potter et al. 61996a,b), we outlined a generalized

simulation approach for emissions of NO and N2O 2. Model description from soils with a simpli®ed application of the con- ceptual ``leaky-pipe'' model proposed by Firestone The NASA±CASA model is a representation of and Davidson 61989). The primary controlling factors major ecosystem carbon and nitrogen transforma- used in this leaky pipe scheme are gross rates of tions. It includes interactions of trace gas ¯ux controls: N mineralization and an index of water ®lled pore nutrient substrate availability, soil moisture, tempera- space 6WFPS; de®ned as the ratio of volumetric soil ture, texture and microbial activity. The model is water content to total porosity of the soil; Papendick designed to simulate daily and seasonal patterns in and Campbell, 1980). In our basic ``leaky pipe'' carbon ®xation, nutrient allocation, litterfall, and soil component for N trace gas emission, processes of nitrogen mineralization, and CO2 exchange, in addi- ammoni®cation and nitri®cation are lumped into tion to N2O and NO production, and CH4 consump- combined mineralization ¯uxes from litter, microbial tion. Potter et al. 61997) provides a complete and soil organic matter pools to a common mineral description of the previous model design used in N pool 6Fig. 2). this study. The proportion of N trace gas 6NO:N2O:N2) emitted For application in this study, several model compo- as a by-product of total N mineralization is repre- nents of daily NASA±CASA version described by sented as overlapping functions of WFPS. This ratio of

Potter et al., 1996a,b, 1997 remain unchanged. For NO:N2O:N2 represents the size of holes in the N ¯ux 100 C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117

Fig. 1. Litter and soil C and N transformations in the NASA±CASA model which lead to substrates for trace gas production. Structure follows the CENTURY model of Parton et al. 61987, 1988). Carbon pools are outlined in black and labeled with C-to-N ratios, C ¯uxes in solid arrows,

CO2 production in stippled arrows; nitrogen pools in gray, N ¯uxes in gray arrows. Levels of litter, microbe 6MIC) and soil organic 6SLOW and OLD) pools are shown. Structural 6S) and metabolic 6M) pools are shown for leaf and root litter.

``pipe''. Both NO and N2O may be produced at for tropical forest soils. Our initial model setting for intermediate levels 620±60%) of WFPS. NO emission NT of both gases was conservative at 2% 6Potter et al., declines as the soil becomes largely water-®lled, 1996a,b), although estimates range up to 5% in ferti- possibly due to diffusion limitations from sites of lized agricultural soils 6Eichner, 1990). production at the intra-aggregate pore space 6Schuster Consumption of CH4 in relatively well-drained soil and Conrad, 1993). At higher moisture levels 660± pro®les is simulated independently from the nitrogen 90%) where reducing conditions develop, relatively cycling components in NASA±CASA using a mod- more N2O is produced in an exponential response. i®ed version of Fick's ®rst law based on computations Under very wet soil conditions 6>90% WFPS), only for diffusivity in aggregated media 6Potter et al.,

N2 is produced. Although the model still requires 1996a,b), together with the daily soil water balance extensive veri®cation, these patterns are consistent model described in the following section. These CH4 with observed ¯ux measurements and current under- consumption algorithms are based on the assumption standing of process-level regulation of nitri®cation that, in relatively well-drained soils at ambient atmo- and denitri®cation end products 6Linn and Doran, spheric concentrations, gas-phase transport of 1984; Robertson, 1989; Matson and Vitousek, 1990; methane becomes rate limiting 6Striegl, 1993), and Williams et al., 1992). Potter et al. 61993, 1996a, temperature response is damped. As soil moisture 1997) provides equations and de®nitions of all vari- increases, air-®lled porosity decreases, resulting in ables shown in Figs. 1 and 2. restricted methane diffusion into the soil and lower

The potential loss of either N2O or NO from soils as rates of consumption. The NASA±CASA methane a percentage of total mineralized nitrogen 6abbre- uptake algorithms are therefore applicable for estima- viated as NT) is a key variable in the model design. tion of oxidative methane consumption in well- Few estimates of the NT term have been published drained topographic sites in the Amazon study areas. C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117 101

Fig. 2. Schematic representation of NASA±CASA soil water, decomposition, and nitrogen trace gas model components. Gross mineralization of nitrogen from litter and soil N pool is a function of rainfall 6R), temperature 6T), soil texture 6TEX) and substrate quality 6q). The scalar for relative production of N trace gases is a function of a %WFPS probability index 6Iw).

3. Re®nements for Amazon ecosystem simulations derived according to the methods described by Wood- ward 61987) and Monteith and Unsworth 61990). In order to more accurately represent climate con- R s rc e T Àe=r trols and soil processes for Amazon ecosystem C and PET net p s a a 61) s g r r =r N cycles, several modi®cations are introduced in this a s a study for the soil hydrology and nutrient cycling where Rnet is the daily net 6shortwave) radiation ¯ux model described by Potter et al. 61997). These changes to the canopy 6W mÀ2), s is the rate of change of satura- include re®nement of water balance equations, moist- tion vapor pressure with temperature 6mbar 8CÀ1), r is À3 ure holding and retention capacity for Amazon soils, the density of air 6kg m ), cp is the speci®c heat of air À1 À1 and addition of a dynamic deforestation sequence to 6J g 8C ), [es TaÀe] is the difference in water include land use change as a factor in simulations of vapor pressure 6mbar) between ambient air 6e) and air tropical biogeochemistry and hydrology. at saturation [es Ta], g is the psychrometric constant À1 À1 6mbar 8C ), ra is the canopy boundary layer resis- À1 À1 3.1. Water balance equations tance 6s m ), and rs is the stomatal resistance to water vapor 6s mÀ1). Model settings for maximum To estimate plant and soil water balance in the conductance in the canopy conform to optimized previous versions of the model, we used a formulation values for Amazon forest sites, as reported by Wright of the empirical Priestly and Taylor 61972) evapotran- et al. 61996a). spiration equation developed by Campbell 61977) and Estimated evapotranspiration ¯ux is calculated by Bonan 61989). For this study, a more physiologically comparing daily PET to the multi-layer model esti- based daily potential evapotranspiration 6PET) is estimate for daily soil moisture content 6Fig. 2). The soil mated using a Penman±Monteith algorithm 6Eq. 61)), pro®le is treated as a series of four layers: ponded 102 C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117 surface water 6for ¯ooded areas only), surface organic water retention curves are designed to re¯ect the matter, topsoil 60.2 m), and subsoil to rooting depth hybrid character of Amazon oxisols, which may act 61.0±10.0 m). These layers can differ in soil texture, like sands in terms of water movement at low tensions, moisture holding capacity, and carbon±nitrogen while holding water like clays at higher tensions. We dynamics. Water balance in the soil is modeled as used the estimated Brooks and Corey 61964) para- the difference between precipitation 6PPT) or volu- meters from these moisture retention curves to modify metric percolation inputs, and PET and drainage our model parameters for a relative drying rate 6RDR) outputs for each layer. All moisture inputs and outputs variable, described by Potter et al. 61993). The result- are assumed to progress from the surface layer down- ing family of Tomasella and Hodnett 61998) logistic ward. A modi®cation for this study was that at least drying curves 6Fig. 3) for an Amazon soils RDR scalar 30% of daily PET demands are met by water extrac- is based on a transformation of the relationship tion from rooting zones located below the surface and between soil water potential and volumetric moisture topsoil layer 6Nepstad et al., 1994). Inputs from rain- content 6Eq. 62)), in the form reported for example by fall may recharge the soil layers to ®eld capacity. Saxton et al. 61986). Excess water percolates through to lower layers and 1 a may eventually leave the system as runoff. RDR 62) 1 aYb 3.2. Soil moisture retention where a and b are soil texture-dependent empirical coef®cients derived from Brooks and Corey 61964) Moisture retention curves for Amazon soils have parameters and Y is the predicted volumetric moisture been derived by Tomasella and Hodnett 61998). These content 6m mÀ1).

Fig. 3. Relative drying rate 6RDR) scalar for Amazon soils, as a function of soil texture and predicted volumetric moisture content 6Tomasella and Hodnett. 1998). Texture classes are de®ned by Zobler 61986) based on clay 6cl) content ranging from 10 6coarse) to 70% 6®ne). C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117 103

3.3. Deforestation effects 4. Field site data for model testing

In order to estimate dynamic effects of forest con- Initialized state and driver variables used in model version in the Amazon by cutting and burning, several evaluation were designed to reproduce as closely as major changes in the model's ¯ow equations and possible ecosystem conditions during 1994 at two storage pools were programmed to occur at the time different rain forest sites located in the Brazilian states of simulated deforestation. Aboveground pools of of RondoÃnia and ParaÂ 6Fig. 4). The spatial resolution biomass carbon and nitrogen, which have been accu- for these simulations was assumed to be on the order mulated either from net primary production 6C) or soil of 1 m2. With respect to soil properties 6Table 1), the mineral uptake 6N) into leaf and wood tissues over the RondoÃnia forest is based on the Fazenda 6ranch) Nova preceding initialization period of several decades Vida site described in detail by Neill et al. 61995, 6Potter and Klooster, 1999), are immediately diverted 1997). The eastern ParaÂ forest is based on the Fazenda by tree cutting into two ¯ow pathways. One loss Vitoria site, located near the city of Paragominas pathway is burning, whereby C and N gases are 6Nepstad et al., 1994), with respect to canopy struc- emitted from the forest stand to the surrounding atmo- ture, leaf chemistry, rooting depth, and soil properties. spheric pool. Burning losses occur in proportion to It is assumed that vegetation at both sites is dominated tissue combustion factors 6mean value of 0.45 for total by moist primary forest species. biomass loss) measured previously in Amazon defor- The two forest sites differ notably in soil type and estation studies 6Kauffman et al., 1995; Carvalho et al., texture. RondoÃnia forest is located on a red-yellow 1998; Guild et al., 1998). The other main pathway for podzolic latosol with a sandy loam texture 6Neill et al., cut forest biomass involves transfer of residual C and 1997). The ParaÂ forest is located on a Kaolinitic N, i.e. the amounts not combusted during burning and yellow latosol 6oxisol) with a clay texture 6Davidson remaining at the soil surface, into actively decompos- and Trumbore, 1995). This difference in sand:clay ing litter pools of structural leaf and wood materials. content of the two sites has a direct effect on the model As decomposition occurs over the following days and setting for soil water balance and potential decom- months, a portion of this unburned residual biomass position rates. will be lost into the atmosphere more slowly 6com- Net primary production 6NPP) of the forests at these pared to direct burning losses) as daily C and N trace study sites has not been measured directly, and thus gas emissions from soil organic matter mineralization. was estimated from NASA±CASA model algorithms Immediately following a deforestation event, land 6Potter et al., 1998) with inputs of monthly AVHRR- cover settings in the model are changed to represent NDVI and daily climate drivers. To generate the ®nal properties of converted agricultural systems, either as simulation results, daily air temperature, surface solar shifting cultivation use, or directly to a pasture cover radiation, rainfall, and predictions of evapotranspira- type, or a temporal sequence of the two. Major model tion and stored soil moisture for 1994 were allowed to changes related to ecosystem structure include frac- control daily predictions of NPP ¯ux at the Amazon tional reduction in absorbed photosynthetic radiation sites. While surface radiation ¯ux was the major 6FPAR) and leaf area index 6LAI) after deforestation. daily controller of NPP at these sites, average daily In place of NDVI inputs, altered canopy cover attri- temperature and diurnal temperature range were butes for the simulated regrowth functions of the site used to drive CASA's NPP algorithm for temperature vegetation are based on typical values reported in the regulation, whereas daily rainfall dynamics were literature for the respective agricultural ecosystems in aggregated to a 30-day running index of moisture Amazonia 6Fageria et al., 1991). With reductions in controls 6Potter et al., 1993). Seasonal patterns of LAI automatically come higher simulated soil surface litterfall, plus initial states for forest ¯oor litter pools temperatures and lower rainfall interception by leaf and soil C pools at the sites were also adopted from surfaces. Furthermore, mineralization rates of C and N methods used in our previous regional modeling from litter and soil pools are programmed to increase studies 6Potter et al., 1998). as a result of changes in the soil microbial environ- Daily climate data sets for time series model inputs ment that occur with disturbance. were constructed for the two sites using records from 104 C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117

Fig. 4. Locations of the forest study sites in the Brazilian Legal Amazon 6map source: Instituto de Pesquisa Ambiental da AmazoÃnia).

1994, compiled at the nearby weather stations. For the tion that rainfall timing can vary markedly from year RondoÃnia site simulations, hourly climate records to year and from site to site across the Amazon Basin. from the Reserva Jaru forest, Ji-Parana station At the ParaÂ location, rainfall in 1994 continued into 6108050S618550W) were obtained from the ABRA- the typically dry months of June and July. There are COS 6Anglo-Brazilian Climate Observation Study) noteworthy seasonal differences in rainfall between archive 6Nobre et al., 1996). Missing data were ®lled the two locations. Rainfall in 1994 was more abundant in with hourly climate records from the nearby late in the year 6September±October) at the RondoÃnia Fazenda Nossa Senhora station 6108450S628220W). site, in contrast to high rainfall amounts during For the ParaÂ site simulations, complete daily data were February±April at the ParaÂ site 6Fig. 5). compiled from hourly climate records at the Fazenda Owing to a lack of reliable radiation measurements Vitoria station near Paragominas 6Nepstad et al., 1994; for the ParaÂ site during the 1994 wet season, we used Jipp et al., 1998). the relationship between diurnal temperature range

The year 1994 was selected for the simulations 6DTR) and net irradiance 6Rnet) ¯ux to reconstruct this principally because this is the single year for which model driver. Atmospheric transmissivity, more com- hourly weather station data was available for both monly called daily cloud cover, is inferable from the forest locations in RondoÃnia and ParaÂ. Final data sets difference between the daily maximum and minimum for daily climate drivers were generated from preci- air temperature using the method of Bristow and pitation totals over a 24 h period, and by averaging net Campbell 61984). The more transmissive the atmo- radiation ¯ux measurements over daylight hours. sphere, the greater the DTR. We calibrated this Rnet± Compared to long-term mean annual rainfall totals DTRfunction using ABRACOS measurements from 2 of 1945 mm at the RondoÃnia forest location and Maraba 6r 0:93; Rnet 29:5 DTR 8:1), to deter- 1730 mm at the ParaÂ forest location, 1994 was drier mine daily radiation ¯uxes at the Paragominas site for than usual in RondoÃnia and somewhat wetter in ParaÂ, the 1994 wet season. We did not adjust the reported with 1571 and 1963 mm rainfall at the two respective dry season Rnet at the Paragominas site, since it was locations. For convenience, we de®ne the typical reported in a consistent manner for all 4 years, albeit Amazon forest `wet season' as October±May and with lower ¯ux values than generally predicted from the `dry season' as June±September, with the realiza- DTRalone. It is conceivable that biomass burning C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117 105

Table 1 Estimated NASA±CASA parameter settings and sources for initialization of Amazon forest simulationsa

Model parameter RondoÃnia ParaÂ

Geographic Latitude and longitude 108300S628300W28590S478310W Elevation 6m) 120b 30c Climate drivers 6daily) Ji-Parana Stationb Fazenda Vitoriac Air temperature 6min/max) 68C) Ji-Parana Stationb Fazenda Vitoriac Precipitation 6cm) Ji-Parana Stationb Fazenda Vitoriac Relative humidity 6%) Ji-Parana Stationb Fazenda Vitoriac Surface radiation 6W mÀ2) Ji-Parana Stationb Fazenda Vitoriac,d Wind speed 6m sÀ1) Ji-Parana Stationb Fazenda Vitoriac Vegetation Net primary production 6g C mÀ2 per year) 1247e [700]e 1264e [700]e Leaf nitrogen content 6C:N ratio) 85d [80]66d [80] Leaf lignin content 6%) 18 [23]18d [23]d One-sided maximum LAI 6m2 mÀ2) 4.6b [2] 5.4c [2.9]c Litter C allocation 6%leaf:root:stem) 45:25:30 [60:50:0] 45:25:30c [60:50:0] Rooting depth into mineral soil 6m) 12 [5]12c [8] Mineral soils Bulk density 6g cmÀ3) 1.28f [1.5]b 0.96c [1.24] Texture 6%sand:silt:clay) 75:4:21f [85:6:9]b 9:12:79c [9:12:79] Minimum water content 6% bed volume) 17.0g [21] 13.0c [25] Field capacity 6% bed volume) 32.0g [25] 37.0b [35] Total porosity 6% bed volume) 48.0b [47] 56.0g [53] Total organic carbon 6to 20±30 cm) 6kg C mÀ2) 3.23f [3.98]f 4.1c [4.8]c Thickness of humus 6cm) 3 [1] 3 [1]

a In the absence of published estimates, best guesses are shown in italics, based on other reported values in the same parameter category. Modi®ed settings for nearby pasture sites are also shown in square brackets. b Wright et al. 61996b). c Nepstad et al. 61994); Jipp et al. 61998). d Guild et al. 61998); Davidson 6unpublished data) and Markewitz 6unpublished data). e Potter et al. 61998). f Neill et al. 61995, 1997). g Tomasella and Hodnett 61998). during the dry season at a site like Paragominas overall patterns in radiation ¯uxes during the wet substantially reduces atmospheric transmissivity dur- season are consistent with seasonal patterns in DTR ing what would otherwise be a nearly `clear sky' solar recorded at weather stations across the Amazon, radiation ¯ux 6Malhi et al., 1999). This is supported by including Ji-Parana, Manaus and Maraba stations from measurements of spectral irradiance and aerosol prop- ABRACOS. erties during the 1995 Smoke, Clouds, and Radiation- Brazil 6SCAR-B) experiment reported by Eck et al. 61998), which imply that PAR¯uxes at the land 5. Evaluation of modeling results surface are reduced between 20 and 45% by biomass burning sources of aerosols in the Amazon Basin. 5.1. Evapotranspiration and WFPS Net radiation ¯ux was slightly less seasonal at the RondoÃnia site, compared to the discernible decrease in Accurate representation of the water balance in radiation ¯ux towards the end of the wet season and tropical soils is crucial to modeling the environmental early in the dry season at the ParaÂ site 6Fig. 5). These controls on soil organic matter decomposition, nutrient 106 C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117

Fig. 5. Daily rainfall 6a) ParaÂ; 6b) RondoÃnia and solar irradiance ¯ux; 6c) ParaÂ; 6d) RondoÃnia for two Amazon forest sites during 1994. mineralization, and trace gas emission ¯uxes. For of the soil pro®le, plant rooting depth, and predicted example, optimum rates of soil respiration ¯ux of evapotranspiration ¯uxes. Hence, we evaluated esti-

CO2 to the atmosphere are generally measured at near mated evapotranspiration 6EET) ®rst as a diagnostic ®eld capacity for water content of the soil 6Linn and variable and the main out¯ow of stored rainfall in the Doran, 1984). This pattern is explained mechanisti- soil pro®le depth to which plant roots have access. cally as a limitation of oxygen diffusion rates required Mean daily rates of EET are predicted by the model for optimal aerobic microbial respiration 6Rh) at high to be 0.43 and 0.47 cm per day 6wet and dry season, soil water content, and limited availability of soluble respectively) at the RondoÃnia location, 0.39 and organic-C substrates in water ®lms at low soil water 0.29 cm per day 6wet and dry season, respectively) content. Moisture content can also in¯uence rates of at the ParaÂ forest location 6Fig. 6a and b). In most

CO2 diffusion in tropical soil pro®les 6Davidson and cases, we ®nd close agreement 6<10% difference) Trumbore, 1995). between our predicted EET ¯uxes and mean measured

As the primary control on CO2 emissions from Rh, EET ¯uxes on a seasonal basis for the study sites, and associated N gas emissions from tropical soils, the comparing EET at the RondoÃnia forest reported at WFPS index of moisture content is predicted in our 0.38 cm per day for the dry season 6Hodnett et al., model for the topsoil layer of the rooting depth pro®le 1996), and EET at the ParaÂ forest reported at 0.38 and 6Fig. 2). Estimation of daily soil water content as 0.45 cm per day for the wet and dry seasons, respec- %WFPS in the NASA±CASA model depends on tively 6Jipp et al., 1998). In terms of environmental precipitation inputs 6corrected for vegetation canopy controls, predicted daily rates of EET from our model interception), moisture holding and retention capacity 6Fig. 6) are in¯uenced more by variability in surface C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117 107

Fig. 6. Predicted evapotranspiration 6a) ParaÂ; 6b) RondoÃnia, and WFPS; 6c) ParaÂ; 6d) RondoÃnia for two Amazon forest sites. solar irradiance, and to a lesser degree by relative %WFPS reported Hodnett et al. 61996) for forests humidity or daily rainfall amounts to recharge soil at Ji-Parana, RondoÃnia, assuming the reported total moisture pools. Williams et al. 61998) reported this soil porosity value of 48% volumetric water content at same general pattern of model sensitivity for EET in these sites. The seasonal pattern in both predicted and Amazon forests. In contrast, higher measured rates of measured data sets for WFPS is a peak near 70% from EET by Jipp et al. 61998) during the dry season at the mid-February into May, followed by low values near ParaÂ forest site are dif®cult to explain considering that 35% WFPS during the typical three month 6June± measured net radiation ¯uxes were about 18% lower in August) dry period, with a return to the peak annual the entire dry season than in the wet season during value by early March. Predicted WFPS in the topsoil 1994 at that site. layer at the ParaÂ forest site ranges between 36 and Predicted differences in %WFPS between the two 82%, which closely tracks the overall range and forest sites 6Fig. 6c and d) are attributable mainly to seasonal changes in measured monthly values for length of the dry season, and to soil texture settings, %WFPS reported by Verchot et al. 61999) for the and hence moisture holding and retention capacity in primary forest at Paragominas. In both predicted the model. The RondoÃnia forest on a relatively coarse- and measured results, the period of low soil moisture textured soil shows the greater seasonal and day-to- 6WFPS < 55%) lasts nearly 6 months 6e.g. July± day variability in predicted WFPS in the topsoil layer, December) in forests of eastern ParaÂ. Our model with a value range between 35 and 73% WFPS. These replicates a sustained level of soil moisture > 65% predictions are consistent with overall range and WFPS estimated for the wet March±May period in seasonal changes in measured values monthly for 1995 and 1996 by Verchot et al. 61999). 108 C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117

5.2. Primary production, heterotrophic respiration, ParaÂ forest site was predicted to be a net sink of about and soil methane uptake 190 g C mÀ2 per year. Although these annual NEP ¯uxes are of similar magnitude, the seasonal responses Plant carbon cycling in tropical forests is tightly are quite different at the two sites. At the RondoÃnia coupled to water ¯uxes, litter fall inputs to soil organic site, predicted NPP is highest shortly after the onset of matter for decomposition, and soil nutrient minerali- the wet season 6November) and declines gradually zation. The accumulated difference between net car- throughout the rest of the yearly cycle 6Fig. 7). At the bon ®xation in NPP 6sink ¯uxes), and its subsequent more seasonal forest site in ParaÂ, predicted NPP release back to the atmosphere by Rh 6source ¯uxes), declines rapidly at the end of the wet season 6May) determine annual net ecosystem production 6NEP) and does not reach a maximum level until the follow- for a site. Recent eddy correlation tower studies in ing March±April period in the yearly cycle. These small undisturbed areas of Amazon forest imply that predicted seasonal carbon ¯ux results are consistent these ecosystems may be large net sinks for atmo- with Amazon tower study results 6Malhi et al., 1998) spheric carbon, with annual NEP values in the range and other modeling analyses 6Williams et al., 1998), of 100±600 g C mÀ2 per year 6Grace et al., 1995, which suggest that limited water availability in forests Malhi et al., 1998). of the eastern Amazon slightly outweighs any com- In a manner consistent with these previous tower pensatory effect of periodically decreased cloudiness studies, our ecosystem model predicts that annual during the dry season. However, the annual net carbon NEP in 1994 at the RondoÃnia forest site was equivalent sink predicted for both RondoÃnia and ParaÂ forest sites to a net sink ¯ux of 276 g C mÀ2 per year, while the is about 50% smaller than that estimated by Malhi et al.

Fig. 7. Predicted net primary production 6a) ParaÂ; 6b) RondoÃnia, and net ecosystem production; 6c) ParaÂ; 6d) RondoÃnia for two Amazon forest sites. C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117 109

61999) for an Amazon forest site near Manaus at 590 g estimates are somewhat higher than annual Rh ¯uxes CmÀ2 per year 6see further explanation below). estimated for a tropical forest near Manaus, Brazil at Predicted soil Rh ¯uxes are closely coupled to daily 970 g C mÀ2 per year 6Malhi et al., 1999), who also À2 ¯uctuations in soil water content in the sandy RondoÃ- estimate total soil CO2 emissions at 1650 g C m per nia forest soils, compared to Rh ¯uxes in the heavy year. This estimate of Rh at Manaus could be biased to clay soils of eastern ParaÂ. Relatively more frequent the low side however, because it was apparently wetting-drying cycles simulated at the RondoÃnia site derived using total soil CO2 emissions and plant 6Fig. 6d) appear to drive stronger moisture controls on respiration estimates from an entirely different site soil organic matter decomposition, nutrient minerali- in RondoÃnia. Underestimation for Rh ¯uxes of soil zation, and turnover of microbial biomass. Optimal CO2 can result in overestimation of the annual net conditions for soil organic matter decomposition at the carbon sink for the forest. RondoÃnia site occur during or close to rainfall events, The NASA±CASA model suggests that moisture creating conditions for net source ¯uxes of CO2 from conditions for microbial decomposition of organic the forest during the wettest months of the year, matter in heavy clay soils are wet enough during whereas the heavy clay soils at the ParaÂ forest site the high rainfall months at the ParaÂ forest site to retain moisture longer after the rains have ended, and sustain high Rh ¯uxes, and dry out to near optimal can sustain Rh ¯uxes of CO2 well into the dry season, soil moisture and temperature conditions for Rh ¯uxes to roughly balance declining NPP ¯uxes. Average of CO2 during the ®rst half of dry season. For the predicted rates of Rh as component ¯uxes of NEP sandy soils at the RondoÃnia forest site, the model at the RondoÃnia forest site are 3.9 and 2.8 g C mÀ2 per suggests that soil moisture conditions for microbial day for the wet and dry season, respectively. At the decomposition are favorable during the high rainfall

ParaÂ forest site, the range for Rh ¯uxes of CO2 during months, but dry out to sub-optimal conditions for Rh À2 the later 1994 wet season months is 3.2±4.1 g C m ¯uxes of CO2 early in the dry season. Predicted per day, and during the driest months is 3.1±3.3 g percentage of total yearly litterfall varies little over CmÀ2 per day, a response pattern that is in close the seasons at the two forest sites, although small agreement with measured seasonal trends in total soil increases in leaf litter during the late dry season at the

CO2 emission at the same site 6Davidson et al., 2000). ParaÂ forest site can increase Rh ¯uxes slightly com- Our ecosystem model does not, however, account for pared to wet season ¯uxes. root respiration contributions to total soil respiration at As another component of soil C cycling, predicted either site, which would be included in ®eld measure- seasonal patterns of soil CH4 uptake are in¯uenced ments of total soil CO2 emissions. notably by soil moisture changes and daily rainfall Total annual Rh ¯uxes are predicted by the model patterns at both sites 6Fig. 8). Mean predicted rates of À2 À2 at 1298 and 1245 g C m per year for the RondoÃnia CH4 uptake are 0.52 and 0.56 mg m per day 6wet and ParaÂ forest sites, respectively. These model and dry seasons, respectively) at the RondoÃnia forest

Fig. 8. Predicted soil CH4 uptake 6a) ParaÂ; 6b) RondoÃnia for two Amazon forest sites. 110 C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117 site, and 0.60 and 0.61 mg mÀ2 per day at the ParaÂ site mainly a function of soil texture settings, which lead to 6wet and dry seasons, respectively). The model pre- prediction of greater retention 6or higher microbial dicts that high rainfall events periodically depress CH4 ef®ciency of carbon substrate use) of organic matter in uptake capacity of the soil at both forest sites, although the heavy clay soils of the eastern ParaÂ site. Predicted estimated CH4 uptake rates recover rapidly with dry- nitrogen storage in the NASA±CASA pools for SLOW ing of the uppermost soil layers. and OLD soil N is likewise higher at the ParaÂ location Comparisons of model results to measured rates of 60.36 kg N mÀ2) than at the RondoÃnia location methane consumption in these soils are valid for the 60.21 kg N mÀ2). Both our predicted pools of C and dry season when potential methane emission ¯uxes, N closely match measured pool sizes of 3:2 Æ 0:3kg which are not included in this application of the CmÀ2 and 0:27 Æ 0:02 kg N mÀ2 at RondoÃnia forest model, can be excluded as confounding factors. Dry sites 6Neill et al., 1997). Measured carbon pool sizes at season rates of CH4 uptake were measured at the eastern ParaÂ forest sites were reported for the upper 0.98 mg mÀ2 per day for primary forest soils near 10 cm soil 6Trumbore et al., 1995), which when Paragominas 6Verchot et al., 2000). Field measure- doubled, would again give a closely matching estimate ments at the ParaÂ site indicate that periodic methane of 5.2 kg C mÀ2 to the model prediction. emissions from forest soils are positively correlated with soil ¯uxes of CO2, and that formation of anae- 5.4. Nitrogen mineralization rates and trace robic soil microsites may cause a switch from a gas emissions consistent soil CH4 sink to an intermittent CH4 source 6Verchot et al., 2000). Our forest model was not Using the vegetation parameter settings in Table 1, used in this manner to include potential soil CH4 net nitrogen mineralization rates are predicted to be production, although future tests of are now planned somewhat higher on average at the ParaÂ location to do so using the methane emission model version 655 mg N mÀ2 per day) than at the RondoÃnia forest 6Potter, 1997). location 626 mg N mÀ2 per day). Annual net miner- À2 Measurements of dry season rates of CH4 uptake at alization is predicted at 19.9 g N m per year for the forest sites in RondoÃnia were reported to be on the ParaÂ forest and 9.3 g N mÀ2 per year for the RondoÃnia order of 1.2 mg mÀ2 per day 6Steudler et al., 1996). forest. This difference is attributable almost entirely to The model settings for the sandy loam soil type of the lower leaf litter N content assumed for RondoÃnia these RondoÃnia forest sites result in a slightly higher forest sites, based on measurements reported by Guild overall soil water holding capacity than for the heavy et al. 61998). If the same leaf C:N ratio of 66 is used for clay soils of eastern ParaÂ sites, but also result in the RondoÃnia site simulations as for the ParaÂ forest prediction of slightly lower water content 6computed site, annual net mineralization is predicted at nearly relative to saturation capacity) when moisture levels 18 g N mÀ2 per year, making the two sites nearly approach the wilting point setting than predicted for equivalent in terms of N cycling rates. We note that a clay soils of the eastern ParaÂ sites. These settings measured range of 79±94 was reported for forest should have the effect of increasing diffusivity and leaf C:N in RondoÃnia by Guild et al. 61998), and up potential uptake of methane at the RondoÃnia forest to 72 C:N for forest leaf litter in eastern ParaÂ sites sites, although under the default model settings not 6Markewitz, unpublished data). This potential site quite to the levels consistent with ®eld measurements variability implies that appropriate measurements of maximum of CH4 uptake rates. and spatial analyses of litter C:N as a model input parameter are needed to reduce uncertainties in nitro- 5.3. Soil C and N turnover and storage gen cycling predictions. At both sites, predicted daily rates for emission of Predicted carbon storage to 20 cm soil depth in the both N trace gases correspond closely to predicted model's pool for SLOW carbon 6mean residence time rates of gross N mineralization 6r2 > 0:8 for non- of ca. 10 years at the two Amazon forest locations) is linear regression using n 365 days). Daily N2O higher at the ParaÂ location 65.8 kg C mÀ2) than at the ¯uxes are weakly correlated with daily WFPS RondoÃnia location 63.9 kg C mÀ2). This difference is 6r2 > 0:2) at both sites, whereas predicted NO ¯uxes C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117 111

Fig. 9. Emission ratio of nitrous-to-nitric oxide 6N2O:NO) versus percent water-®lled pore space 6%WFPS) from three different sources: measured ¯uxes at Paragominas 6Verchot et al., 1999); measured ¯uxes from a collection of soils in Costa Rica 6Keller and Reiners, 1994); and NASA±CASA model settings used in this simulation study. are not signi®cantly correlated 6r2 < 0:1) with daily Verchot et al. 61999) suggests that for WFPS between WFPS, mainly because of the broad range of WFPS 40 and 70%, the model settings somewhat under- levels over which relatively high NO emissions can be estimate the N2O:NO ratio measured at these eastern predicted by the model 6Fig. 2). ParaÂ sites 6Fig. 9). This would explain some, if not

Annual emission ¯uxes of N2O and NO are pre- most, of the discrepancy between model predictions dicted at 0.24 and 0.26 g N mÀ2 per year, respectively, and measurements of N trace gas ¯uxes for these for the ParaÂ forest site, and 0.11 and 0.17 g N mÀ2 per Amazon forest sites. For further comparison with year for the RondoÃnia forest site using the vegetation available measurements from tropical ®eld sites, parameter settings in Table 1. Consistent with predic- Fig. 9 includes the N2O:NO ratio versus %WFPS tion of higher litter fall N, wetter soil conditions, and relationship based on a synthesis of ¯ux measurements therefore higher net N mineralization rates, annual N from soils in Costa Rica 6Keller and Reiners, 1994), gas ¯uxes from the model are higher overall at the ParaÂ which suggests yet another model setting that would site. Measured ¯uxes of soil N2O are 0.25 and 0.24 g even more markedly underestimate the N2O:NO ratio NmÀ2 per year for RondoÃnia and ParaÂ forest sites, measured at eastern ParaÂ sites by Verchot et al. 61999). respectively 6Verchot et al., 1999), which closely Additional measurements and model re®nements are matches our model prediction for the ParaÂ forest, apparently needed to resolve these differing response but not for the RondoÃnia site, which predicts an annual functions shown in Fig. 9. À2 soil N2O ¯ux of only 0.14 g N m per year, even if the higher leaf C:N ratio of 66 is used. The mean predicted daily emission ratio at the RondoÃnia forest 6. Simulation of deforestation effects site of 0.62 for nitrous-to-nitric oxide 6N2O:NO) is substantially lower than the N2O:NO daily emission A deforestation event was simulated for the ratio of 1.1 at the ParaÂ forest site, because of the wetter RondoÃnia forest site 6initialized to steady state con- predicted soil conditions at the ParaÂ site. ditions) to examine the model response in terms of A comparison between our model settings 6Fig. 2) biomass losses through burning, decomposition of and a measured N2O:NO ratio versus %WFPS by residual unburned biomass, and overall soil carbon 112 C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117 and nitrogen changes, assuming the forest site would setting for N content of tropical forest leaves, which be subsistence farmed for about 2 years following must be re®ned for other sites in future simulations. slash and burn of the forest trees, without the outside The simulated deforestation sequence we generated input of chemical fertilizers. The choice here of a predicts that surface soil C pools decline rapidly to cultivation land use following forest clearing was not about one-third of their pre-cut forest level, following intended to diminish the importance of understanding a small input of unburned forest leaves after cutting effects of pasture development in the RondoÃnia area. 6Fig. 10). Meanwhile, sub-surface soil C pools grow We note that pasture grasses may in fact have different initially by about 10% from the transfer and storage of production and nutrient use ef®ciencies than some the decomposing unburned forest biomass, including crop plants, which will require more information residual wood from charred tree trunks. This pattern of from LBA and related studies to incorporate into carbon accumulation appears to have leveled off after our modeling analysis. about 2 years of simulated cultivation and should For deforestation events studied in RondoÃnia, decline in years of land use to follow. Guild et al. 61998) reported mean losses from forest In terms of net ecosystem carbon ¯uxes, the site slash and burn of 8800 g C mÀ2 and 118 g N mÀ2. switches from being a net sink for carbon to a sub- These ¯uxes are comparable to our model prediction stantial source following the large loss of biomass of 9339 g C mÀ2 and 39 g N mÀ2. The difference in from simulated burning. The 2-year old cultivated site nitrogen losses derives from the model's default is predicted to be an annual net source of ecosystem

Fig. 10. Model predictions for a simulated deforestation event 6indicated by the black arrow) at the RondoÃnia forest site, assuming the site would be subsistence farmed for 2 years following slash and burn of the forest trees. C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117 113 carbon to the atmosphere of about 1.58 kg C mÀ2 per incorporated into the modeling of terrestrial ecosys- year. Over this period, mineral N levels in the soil are tem processes over a vast area like Amazonia, which predicted to increase by a factor of about ®ve for the commonly experiences large agricultural ®res during 2-year simulation following deforestation. Compared driest months, a model driver generated from the to the primary forest that was cut and burned, this AVHRR-NDVI has the potential to provide extensive causes a doubling in annual emissions of N2O gas at coverage of surface properties and land use patterns the site during the ®rst year following deforestation, over 10 to 15 year time periods. which continues with only a slight decrease 68%) Besides the need to improve the satellite NDVI during the second year following deforestation, as signal for tracking changes over time in regional the mineral nitrogen released through deforestation productivity, at least two other limitations arise in is slowly immobilized in soil pools. the process of scaling up to the Amazon region an ecosystem model like the one described in this paper. 1. Individual model input variables are not accu- 7. Implications for scaling to the Amazon region rately mapped, or are incompletely represented, in terms of either the ``lumped'' parameter values, or One of the principal objectives of the LBA is to the frequency distribution of values within a quantify and predict how energy, water, carbon, trace speci®ed geographic area. gas, and nutrient cycles in the Amazon region respond 2. Combinations of model input variables are not to deforestation, agricultural practices and other land accurately geo-recti®ed to one another, or are use changes, and how these responses are in¯uenced available only at widely differing spatial and/or by climate 6LBA, 1996). This objective is attainable temporal resolutions. with the proper coupling and comparison of observational data, ®eld measurements, and dynamic ecosys- Land cover, land use, and soils classi®cations for tem models that can include land use change, like the the entire Amazon region are subject to both these one described in this paper. If ecosystem models can limitations. Regional land cover maps of the Amazon be re®ned and adequately validated against measure- produced by Skole and Tucker 61993) and Stone et al. ments at several site locations for the major land cover 61994) have been checked for accuracy in broadly- and use types in the Amazon basin, an important de®ned land use types in Amazonia, which may help research task remaining is to scale-up the model's minimize limitation 61) above. While these maps predictions to the regional level. provide useful starting points for regional ecosystem Ideally, any ecosystem model's set of algorithms simulations, their data sources for land use patterns are controlling terrestrial energy, water, carbon, nitrogen now about 10 years out-of-date. New maps of Amazon and, trace gas ¯uxes will be nearly as accurate when land cover, based on seasonal satellite observations applied at a relative coarse spatial resolution, for from the AVHRR 6Eidenshink and Faundeen, 1994), example, at the 8 km resolution used in our previous still require extensive ground truth using classi®ed regional modeling studies of Amazonia 6Potter et al., Landsat images and ®eld site veri®cation, along with 1998; Nepstad et al., 1999), as at the scale of the tens methods for improving geo-recti®cation to ground of square meters of an intensive study site used for control points. With respect to dynamic land use model validation. In regional modeling, the direct use modeling, at least two such regional maps of AVHRR of satellite sensor data to capture and integrate terres- land use, separated in time by 2 or 3 years, would be trial surface properties 6e.g. FPAR, LAI) and land necessary to implement the NASA±CASA model cover changes over large areas has been shown to deforestation sequence described above for simulating meet these requirements for accurate scaling of bio- land use change effects on water, carbon, trace gas sphere-atmosphere ¯uxes of water and carbon to ¯uxes over the entire Amazon basin. regional and global levels 6Maisongrande et al., There are additional input data sets required for 1995; Goetz and Prince, 1996; Malmstrom et al., the Amazon region that will need to be developed 1997; Potter et al., 1999). Assuming that adjustments further to serve as model drivers and surface property for atmospheric smoke and related aerosols can be settings for basin-wide simulations. Table 2 presents 114 C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117

Table 2 Required data sets for regional simulations of water, carbon, nitrogen and, trace gas ¯uxes over the Amazon region

Data sets Remote sensing sourcea Other sources

Geographic Elevation 6m) US Defense Mapping Agency Water table depth 6m)/inundation 6%) JERS Climate drivers 6daily) Surface air temperature 6min/max) 68C) University of East Anglia 6New et al., 2000) Precipitation 6cm) TRMM DNAEE 6Divisao Nacional de Aguas e Energia Eletrica) Surface relative humidity 6%) University of East Anglia 6New et al., 2000) Surface radiation flux 6W mÀ2) GOES Surface wind speed 6m sÀ1) University of East Anglia 6New et al., 2000) Vegetation FPAR/LAI AVHRR/MODIS Leaf nitrogen content 6C:N ratio) Rooting depth into mineral soil 6m) Disturbance history 6years since) Landsat, AVHRR/MODIS Land use history 6type) Landsat, AVHRR/MODIS Mineral soils RADAM in Potter et al. 61998) Soil type 6class) Bulk density 6g cmÀ3) Texture 6%sand:silt:clay)

a Summary of data sources and satellite sensor descriptions available from USGCRP 61999) and the Data Management Working Group of the Subcommittee on Global Change Research. these data sets, with speci®cations for running a daily Nonetheless, in a previous model sensitivity analysis, model of water, carbon, nitrogen, and trace gas ¯uxes we found that estimated NPP of Amazon primary like NASA±CASA. The requirement for daily climate forests can increase by almost 6% when the deep inputs at the regional level is a need that apparently rooting depth 610 m) is set, in place of a more shallow cannot be ful®lled without resorting to predicted out- rooting depth 62 m) used for forests in several global put from atmospheric general circulation models ecosystem models. Moreover, sensitivity analysis of 6AGCMs) or statistical interpolation techniques using this type may show even larger effects of rooting depth monthly observational data sets. Based on our site settings with consideration of periodic drought effects modeling studies reported in this paper, daily data sets on NPP, and the possibility that deep rooting and for surface radiation ¯ux and precipitation rates are the drought tolerance by trees in evergreen forests of most important model drivers to evaluate for regional the Amazon maintains primary production during and temporal bias. extremely dry years 6Nepstad et al., 1995, 1999). Among the other input data sets listed in Table 2, In summary, an improved understanding of how plant rooting depth into mineral soil merits special tropical forest conversion in¯uences regional water, consideration. As we reported in a previous modeling carbon, nutrient dynamics and trace gas ¯uxes in study of water, carbon, and trace gas ¯uxes for the Amazonia will depend upon bringing to bear and Amazon region 6Potter et al., 1998), improvements in improving all available methods of ecosystem process the large-scale modeling of below ground biomass monitoring and inventory, including remote sensing, dynamics and water balance will depend in part on ground-based sampling, tower ¯ux measurements, more extensive information and ®eld measurements ecosystem modeling, statistical analysis and geo- for forest rooting depths across the entire region. graphic information systems. Errors from each of Such measurements are time-consuming and labor- these methods must be reduced to the greatest extent intensive, which helps explain the relatively small data possible by systematic intercomparisons of their inde- set now available for rooting patterns of tropical trees. pendent estimates for hydrology and biogeochemistry. C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117 115

Nonetheless, predicted carbon balance from mod- Methane, Nitrogen Oxides and Halomethanes. American eling results presented in this paper implies that Society for Microbiology, Washington, DC, pp. 219±235. Davidson, E.A., Trumbore, S.E., 1995. Gas diffusivity and relatively undisturbed Amazon forest ecosystems production of CO2 in deep soils of the eastern Amazon. Tellus may be strong net sinks for atmospheric CO2. Follow- 47B, 550±565. ing large losses of biomass from simulated burning, Davidson, E.A., Verchot, L.V., Cattanio, J.H., Ackerman, I.L., the model also predicts, however, that these forest sites Carvalho, J.E.M., 2000. Effects of soil water content on soil switch from being a net sink for CO to a substantial respiration in forests and cattle pastures of eastern Amazonia. 2 Biogeochemistry 48, 53±69. terrestrial source, with annual emission ¯uxes of Eck, T.F., Holben, B.N., Slutsker, I., Setzer, A., 1998. Measure- carbon to the atmosphere equivalent in magnitude ments of irradiance attenuation and estimation of aerosol to about 10 years of undisturbed NEP sink ¯uxes. single-scattering albedo for biomass burning aerosols in These implied patterns in forest carbon dynamics Amazonia. J. Geophys. Res. 103, 31865±31878. remain subject to veri®cation over the entire Amazon Eichner, M.J., 1990. Nitrous oxide emissions from fertilized soil: region. Improvements in model process algorithms for summary of available data. J. Environ. Quality 19, 272±280. Eidenshink, J.C., Faundeen, J.L., 1994. The 1 km AVHRR global carbon and nitrogen trace gas ¯uxes across gradients land data set-®rst stages in implementation. Int. J. Remote of soil types in the Amazon basin will require further Sensing 15, 3443±3462. study as part of new LBA ®eld campaigns. Firestone, M., Davidson, E., 1989. Microbial basis of NO and N2O production and consumption. In: Andreae, M.O., Schimel, D.S. 6Eds.), Exchange of Trace Gases Between Ecosystems and the Atmosphere. Wiley, Dahlem Koferenzen. Acknowledgements Fageria, N.K., Baligar, U.C., Jones, C.A., 1991. Growth and Mineral Nutrition of Field Crops. Marcel Dekker, New York, We thank two anonymous reviewers for helpful 476 pp. comments on an earlier version of the manuscript. Goetz, S.J., Prince, S.D., 1996. Remote sensing of net primary production in boreal forest stands. Agric. For. Meteorol. 78, This work was supported by grants from the NASA 149±179. LBA-Ecology Program and the NASA Land Surface Grace, J., Lloyd, J., McIntyre, J., Miranda, A.C., Meir, P., Miranda, Hydrology Program. H.S., Nobre, C., Moncrieff, J., Massheder, J., Malhi, Y., Wright, I., Gash, J., 1995. Carbon dioxide uptake by an undisturbed rain forest in southwest Amazonia. Science 270, 778±780. Guild, L.S., Kauffman, J.B., Ellingson, L.J., Cummings, D.L., References Castro, E.A., 1998. Dynamics associated with total aboveground biomass, C, nutrient pools, and biomass burning of Bonan, G.B., 1989. A computer model of the solar radiation, soil primary forest and pasture in RondoÃnia, Brazil during SCAR-B. moisture and soil thermal regimes in boreal forests. Ecol. J. Geophys. Res. 103, 32091±32100. Modelling 45, 275±306. Hodnett, M.G., Oyama, M.D., Tomasella, J., Marques Filho, A. de Bristow, K.L., Campbell, G.S., 1984. On the relationship between O., 1996. Comparisons of long term soil water storage incoming solar radiation and daily maximum and minimum behaviour under pasture and forest in three areas of Amazonia. temperature. Agric. For. Meteorol. 31, 159±166. In: Gash, J.H.C., Nobre, C.A., Roberts, J.M., Victoria, R.L. Brooks, R.H., Corey, A.T., 1964. Hydraulic properties of porous 6Eds.), Amazonian Deforestation and Climate. Wiley, Chiche- media. Hydrology Paper 3. Colorado State University, Fort ster, UK, pp. 57±77. Collins, CO, 27 pp. Hutchinson, G.L., Davidson, E.A., 1993. Processes for production Calvert, J.G., Stockwell, W.R., 1983. Acid generation in the and consumption of gaseous nitrogen oxides in soil. Agricul- troposphere by gas-phase chemistry. Environ. Sci. Technol. 17, tural Ecosystem Effects on Trace Gases and Global Climate 428±443. Change. American Society of Agronomy, Special Publication Campbell, G.S., 1977. An Introduction to Environmental Biophy- no. 55, Madison, WI, pp. 79±93. sics. Springer, New York, 159 pp. Jipp, P.H., Nepstad, D.C., Cassel, D.K., Reis de Cavalho, C., 1998. Carvalho Jr., J.A., Higuchi, N., Araujo, T.M., Santos, J.C., 1998. Deep soil moisture storage and transpiration in forests and Combustion completeness in a rainforest clearing experiment in pastures of seasonally dry Amazonia. Climate Change 39, Manaus, Brazil. J. Geophys. Res. 103, 13195±113199. 395±412. Cicerone, R.J., 1987. Changes in stratospheric ozone. Science 237, Kauffman, J.B., Cummings, D.L., Ward, D.E., Babbitt, R., 1995. 35±42. Fire in the Brazilian Amazon: biomass, nutrient pools, and Davidson, E.A., 1991. Fluxes of nitrous and nitric oxide from losses in slashed primary forests. Oecologia 104, 397±408. terrestrial ecosystems. In: Rogers, J.E., Whitman, W.B. 6Eds.), Khalil, M.A.K., Rasmussen, R.A., 1992. The global sources of Microbial Production and Consumption of Greenhouse Gases: nitrous oxide. J. Geophys. Res. 97, 14651±14660. 116 C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117

Keller, M.E., Reiners, W.A., 1994. Soil-atmosphere exchange of Nepstad, D., de Carvalho, C.R., Davidson, E., Jipp, P., Lefebvre, P., nitrous oxide, nitric oxide and methane under secondary forest Negreiros, G.H., da Silva, E.D., Stone, T., Trumbore, S., Vieira, succession of pastures to forest in the Atlantic lowlands of S., 1994. The role of deep roots in the hydrologic and carbon Costa Rica. Global Biogeochem. Cycles 8, 399±409. cycles of Amazonia forests and pastures. Nature 372, 666±669. Keller, M., Goreau, T.J., Wofsy, S.C., Kaplan, W.A., McElroy, Nepstad, D., Jipp, P., Moutinho, P., Negreiros, G.H., Vieira, S., M.B., 1983. Production of nitrous oxide and consumption of 1995. Forest recovery following pasture abandonment in methane by forest soils. Geophys. Res. Lett. 10, 1156±1159. Amazonia: canopy seasonality, ®re resistance, and ants. In:

Keller, M., Kaplan, W.A., Wofsy, S.C., 1986. Emissions of N2O, CH4 Rappaport, D.J., Gaudet, C.L., Calow, P. 6Eds.), Evaluating and and CO2 from tropical soils. J. Geophys. Res. 91, 11791±11802. Monitoring the Health of Large-Scale Ecosystems. NATO ASI Keller, M., Vledkamp, E., Weitz, A.M., Reiners, W.A., 1993. Effect Series, Vol. I28, Springer, Berlin, pp. 333±349. of pasture age on soil trace-gas emissions from a deforested Nepstad, D.C., VerõÂssimo, A., Alencar, A., Nobre, C., Lima, E., area of Costa Rica. Nature 365, 244±246. Lefebvre, P., Schlesinger, P., Potter, C., Moutinho, P., Mendoza, LBA Science Planning Group, 1996. The Large Scale Biosphere- E., Cochrane, M., Brooks, V., 1999. Large-scale impoverish- Atmosphere Experiment in Amazonia 6LBA) Concise Experi- ment of Amazonian forests by logging and ®re. Nature 398, mental Plan. Cachoeira Paulista SP, Brazil, 44 pp. 505±508. Linn, D.M., Doran, J.W., 1984. Effect of water-®lled pore space on New, M., Hulme, M., Jones, P., 2000. Representing twentieth carbon dioxide and nitrous oxide production in tilled and century space-time climate variability. II. Development nontilled soils. Soil Soc. Am. J. 48, 1267±1272. of 1901±1996 monthly grids of terrestrial surface climate. Livingston, G.P., Vitousek, P.M., Matson, P.A., 1988. Nitrous oxide J. Climate 13, 2217±2238. ¯uxes and nitrogen transformations across a landscape gradient Nobre, C.A., Gash, J.H.C., Roberts, J.M., Victoria, R.L., 1996. in Amazonia. J. Geophys. Res. 93, 1593±1599. Conclusions from ABRACOS. In: Gash, J.H.C., Nobre, C.A., Logan, J.A., Prather, M.J., Wofsy, S.C., McElroy, M.B., 1981. Roberts, J.M., Victoria, R.L. 6Eds.), Amazonian Deforestation Tropospheric chemistry: a global perspective. J. Geophys. Res. and Climate. Wiley, Chichester, UK, pp. 577±595. 86, 7210±7254. Papendick, R.I., Campbell, G.S., 1980. Theory and measurement of Luizao, F., Matson, P.A., Livingston, G., Luizao, R., Vitousek, water potential. Water Potential Relations in Soil Microbiology. P.M., 1989. Nitrous oxide ¯ux following tropical land clearing. Soil Science Society of America, Madison, WI, pp. 1±22. Global Biogeochem. Cycles 3, 281±285. Parton, W.J., McKeown, B., Kirchner, V., Ojima, D., 1992. Maisongrande, P., Ruimy, A., Dedieu, G., Saugier, B., 1995. CENTURY Users Manual. Natural Resource Ecology Labora- Monitoring seasonal and interannual variations of gross tory, Colorado State University, Fort Collins, CO. primary productivity, net primary productivity, and net Potter, C.S., 1997. An ecosystem simulation model for methane ecosystem productivity using a diagnostic model and remotely production and emission from wetlands. Global Biogeochem. sensed data. Tellus 47B, 178±190. Cycles 11, 495±506. Malhi, Y., Nobre, A.D., Grace, J., Kruijt, B., Pereira, M.G.P., Culf, Potter, C.S., 1999. Terrestrial biomass and the effects of A., Scott, S., 1998. Carbon dioxide transfer over a central deforestation on the global carbon cycle. BioScience 49, Amazonian rain forest. J. Geophys. Res. 103, 31593±31612. 769±778. Malhi, Y., Baldocchi, D.D., Jarvis, P.G., 1999. The carbon balance Potter, C.S., Klooster, S.A., 1999. Dynamic global vegetation of tropical, temperate, and boreal forests. Plant Cell Environ. modeling 6DGVM) for prediction of plant functional types and 22, 715±740. biogenic trace gas ¯uxes. Global Ecol. Biogeogr. Lett. 8 66), Malmstrom, C.M., Thompson, M.V., Juday, G.P., Los, S.O., 473±488. Randerson, J.T., Field, C.B., 1997. Interannual variation in Potter, C.S., Randerson, J.T., Field, C.B., Matson, P.A., Vitousek, global scale net primary production: testing model estimates. P.M., Mooney, H.A., Klooster, S.A., 1993. Terrestrial ecosys- Global Biogeochem. Cycles 11 63), 367±392. tem production: a process model based on global satellite and Matson, P.A., Vitousek, P.M., 1990. An ecosystem approach to the surface data. Global Biogeochem. Cycles 7 64), 811±841. development of a global nitrous oxide budget. BioScience 40, Potter, C.S., Davidson, E.A., Verchot, L., 1996a. Estimation of 672±677. global biogeochemical controls and seasonality in soil methane Monteith, J.L., 1972. Solar radiation and productivity in tropical consumption. Chemosphere 32, 2219±2246. ecosystems. J. Appl. Ecol. 9, 747±766. Potter, C.S., Matson, P.A., Vitousek, P.M., Davidson, E.A., 1996b. Monteith, J.L., Unsworth, M.H., 1990. Principles of Environmental Process modeling of controls on nitrogen trace gas emissions Physics, Second Edition. Edward Arnold, London. from soils world-wide. J. Geophys. Res. 101, 1361±1377. Neill, C., Piccolo, M.C., Steudler, P.A., Melillo, J.M., Feigl, B.J., Potter, C.S., Riley, R.H., Klooster, S.A., 1997. Simulation Cerri, C.C., 1995. Nitrogen dynamics in soils of forests and modeling of nitrogen trace gas emissions along an age active pastures in the western Brazilian Amazon Basin. Soil gradient of tropical forest soils. Ecol. Modelling 97 63), Biol. Biochem. 27, 1167±1175. 179±196. Neill, C., Melillo, J.M., Steudler, P.A., Cerri, C.C., de Moraes, Potter, C.S., Davidson, E.A., Klooster, S.A., Nepstad, D.C., de J.F.L., Piccolo, M.C., Brito, M., 1997. Soil carbon and nitrogen Negreiros, G.H., Brooks, V., 1998. Regional application of an stocks following forest clearing for pasture in the southwestern ecosystem production model for studies of biogeochemistry in Brazilian Amazon. Ecol. Appl. 7, 1216±1225. Brazilian Amazonia. Global Change Biol. 4 63), 315±334. C. Potter et al. / Forest Ecology and Management 152 /2001) 97±117 117

Potter, C.S., Klooster, S.A., Brooks, V., 1999. Interannual Thompson, A.M., 1992. The oxidizing capacity of the Earth's variability in terrestrial net primary production: exploration of atmosphere: probable past and future changes. Science 256, trends and controls on regional to global scales. Ecosystems 2 1157±1165. 61), 36±48. Tomasella, J., Hodnett, M.G., 1998. Estimating soil water retention Priestly, C.H.B., Taylor, R.J., 1972. On the assessment of surface characteristics from limited data in Brazilian Amazonia. Soil heat ¯ux and evaporation using large-scale parameters. Sci. 163, 190±202. Monthly Weather Rev. 100, 81±92. Trumbore, S.E., Davidson, E.A., de Camargo, P.B., Nepstad, D.C., Prinn, R., Cunnold, D., Rassmussen, R., Simmonds, P., Alyea, F., Martinelli, L.A., 1995. Belowground cycling of carbon in Crawford, A., Fraser, P., Rosen, , 1990. Atmospheric emissions forests and pastures of eastern Amazonia. Global Biogeochem. and trends of nitrous oxide deduced from 10 years of ALE- Cycles 9, 515±528. GAGE data. J. Geophys. Res. 95, 18369±18385. US Global Change Research Program 6USGCRP), 1999. Our Ramanathan, V., Cicerone, R.J., Singh, H.B., Kiehl, J.T., 1985. Changing Planet: The FY 2000 Global Change Research Trace gas trends and their potential role in climate change. J. Program. Committee on Environment and Natural Resources, Geophys. Res. 90, 5547±5566. National Science and Technology Council, Washington, DC. Riley, R.H., Vitousek, P.M., 1995. Nutrient dynamics and nitrogen Verchot, L.V., Davidson, E.A., Cattanio, J.H., Ackerman, I.L., trace gas ¯ux during ecosystem development in montane rail 2000. Land use change and biogeochemical controls of forest. Ecology 76, 292±304. methane ¯uxes in soils in eastern Amazonia. Ecosystems 3, Robertson, G.P., 1989. Nitri®cation and denitri®cation in humid 41±56. tropical ecosystems: potential controls on nitrogen retention. Verchot, L.V., Davidson, E.A., Cattanio, J.H., Ackerman, I.L., In: Proctor, J. 6Ed.), Mineral Nutrients in Tropical Forest and Erickson, H.E., Keller, M., 1999. Land use change and Savanna Ecosystems. Special Publication Number 9 of the biogeochemical controls of nitrogen oxide emissions from British Ecological Society, Blackwell Scienti®c Publications, soils in eastern Amazonia. Global Biogeochem. Cycles 13, Oxford, pp. 55±69. 31±46. Saxton, K.E., Rawls, W.J., Romberger, J.S., Papendick, R.I., 1986. Williams, E.J., Guenther, A., Fehsenfeld, F.C., 1992. An inventory Estimating generalized soil-water characteristics from texture. of nitric oxide emissions from soils in the United States. Soil Sci. Soc. Am. J. 50, 1031±1036. J. Geophys. Res. 97, 7511±7519. Schimel, D.S., Alves, D., Enting, I., Heimann, M., Joos, F., Williams, M., Malhi, Y., Nobre, A., Rastetter, E.B., Grace, J.,

Raynaud, D., Wigley, T., 1996. CO2 and the carbon cycle. In: Pereira, M.G.P., 1998. Seasonal variation in net carbon Houghton, J.T. 6Ed.), IPCC Second Scienti®c Assessment of exchange and evapotranspiration in a Brazilian rain forest: a Climate Change. Cambridge University Press, UK, pp. 76±86. modelling analysis. Plant Cell Environ, 21, 953±968. Schuster, M., Conrad, R., 1993. Metabolism of nitric oxide and Woodward, F.I., 1987. Climate and Plant Distribution. Cambridge nitrous oxide during nitri®cation and denitri®cation in soil at University Press, Cambridge, 174 pp. different incubation conditions. Microbiol. Ecol. 101, 133±143. Wright, I.R., Gash, J.H.C., da Rocha, H.R., Roberts, J.M., 1996a. Skole, D., Tucker, C., 1993. Tropical deforestation and habitat Modeling surface conductance for Amazonian pasture and fragmentation in the Amazon: satellite data from 1978 to 1988. forest. In: Gash, J.H.C., Nobre, C.A., Roberts, J.M., Victoria, Science 260, 1905±1910. R.L. 6Eds.), Amazonian Deforestation and Climate. Wiley, Steudler, P.A., Melillo, J.M., Feigl, B.J., Neill, C., Piccolo, M.C., Chichester, UK, pp. 437±458. Cerri, C.C., 1996. Consequence of forest-to-pasture conversion Wright, I.R., Nobre, C.A., Tomasella, J., da Rocha, H.R., Roberts,

on CH4 ¯uxes in the Brazilian Amazon. J. Geophys. Res. 101, J.M., Vertamatti, E., Culf, A.D., Alvala, R.C.S., Hodnett, M.G., 18547±18554. Ubarana, V.N., 1996b. Toward a GCM surface parameterization Stone, T.A., Schlesinger, P., Houghton, R.A., Woodwell, G.M., of Amazonia. In: Gash, J.H.C., Nobre, C.A., Roberts, J.M., 1994. A map of vegetation of South America based on satellite Victoria, R.L. 6Eds.), Amazonian Deforestation and Climate. imagery. Photogrammetric Eng. Remote Sensing 60, 541±551. Wiley, Chichester, UK, pp. 474±504. Striegl, R.G., 1993. Diffusional limits to the consumption of Zobler, L., 1986. A world soil ®le for global climate modeling, atmospheric methane by soils. Chemosphere 26, 715±720. NASA Tech. Memo 87802, 32 pp.