Soil saturation effects on forest dynamics: scaling across a southern boreal/northern hardwood landscape John F. Weishampel1,*, Robert G. Knox2 and Elissa R. Levine2 1Department of Biology, University of Central Florida, Orlando, FL 32816-2368 2Code 923 Biospheric Sciences Branch, NASA Goddard Space Flight Center, Greenbelt, MD 20771 Keywords: aggregation, biomass, drainage class, GIS, soil maps, succession model, waterlogging *author for correspondence. Fax: (407) 823-5769 Phone: (407) 823-6634 e-mail: jweisham@pegasus.cc.ucf.edu Abstract Simulations of forest response to climatic change at the southern boreal/northern hardwood transition zone need to incorporate landscape-level patterns of soil heterogeneity. The duration and extent of soil moisture saturation are critical edaphic properties that determine the present distribution and structure of forest communities of this region. To capture this relationship for soil series found in east-central Maine, we parameterized a forest gap model using soil moisture attributes derived from a physical model of soil moisture and heat flux. Simulated average soil moisture and depth to an apparent water table were highly dependent on soil series attributes, e.g., horizon thicknesses, saturated hydraulic conductivity, porosity, and clay mass fraction. Differences among the series yielded a variety of communities comprised of species with various tolerances for soil waterlogging. Aboveground biomass accumulated more slowly in poorer draining soils, which suggests that climatically drier conditions may increase the capacity of patches, that presently consist of forested wetlands, to function as carbon sinks. When these predictions were scaled to a 600 ha landscape based on local- (1:12,000), county- (1:120,000) and state- (1:250,000) scale soil maps, the removal of the waterlogging effect caused a 20-60% increase in biomass accumulation during the first 50 years of simulation. However, this increase and the maximum level of biomass accumulation over a 200 year period varied as much as 40% depending on the source of geographic information which provided the composition of the soil series. Thus, simulated effects of soil heterogeneity depended on the scale of the soil data provided, suggesting caution when scaling up effects of potential climate change with forest patch models. 1. Introduction Individual-based forest patch models can address questions at larger spatial scales by using stratified sampling to simulate patches representing different environmental conditions and aggregating the area-weighted results to represent a landscape mosaic (Shugart et al. 1992). This form of direct extrapolation (King 1992) assumes conditions within a patch to be homogeneous and typically ignores patch to patch interactions which may be related to the spatial configuration of the landscape. Although individual-based forest models have depicted the trajectories of patches that comprise a contiguous, heterogeneous landscape (e.g., Pearlstine et al. 1985, Shao et al. 1994), aggregated, landscape-level responses from these models have not been reported. Here, we report aggregated responses for a landscape in east-central Maine. The complex patterning of communities located near the transition zone between the southern boreal and northern hardwood forests is often determined by edaphic influences, expressed largely by soil moisture regimes which may favor the development of forested and unforested bogs (Glebov and Korzukhin 1992, Pastor and Mladenoff 1992). Though the formation of broad scale wetlands is primarily dependent on climatic variables, e.g., evapotranspiration and precipitation, finer scale mosaic patterning of forested wetlands is a function of water table levels which are dependent on drainage and local soil properties (Solomon 1992). Because of the overriding importance of hydrological properties of the underlying soil, vegetation patches located near this ecotone are often coarsely classified accordingly, i.e., into wet, mesic, and dry categories (Bonan and Shugart 1989). The recognition of the importance of soil saturation in determining forest structure near this ecotone between northern hardwood and boreal forests has led to the inclusion of related hydrological controls in one series of succession models (e.g., Botkin and Levitan 1977, Botkin et al. 1989, Botkin and Nisbet 1992a,b, Botkin 1993). For more northern latitudes where there is a permafrost layer, a depth to seasonal thaw growth multiplier was formulated (Bonan 1989), which in effect, mimics the waterlogging multiplier as permafrost impedes infiltration and yields elevated soil moisture levels during the growing season (Bonan and Shugart 1989). However, other simulation efforts for this same general geographic region (e.g., Solomon 1986, Pastor and Post 1986, Pastor and Post 1988, Martin 1992, Bugmann and Solomon 1995) have concentrated on the dynamics of more upland forests and thus neglected saturation effects. This inconsistency may be due in part to the difficulty in measuring soil moisture regimes across this heterogenous landscape. Because the number of observation wells needed to adequately describe the saturation level varies with the complexity of the soil profile (USDA SCS 1971) and horizontal and vertical properties of soils which were subject to glacial activity are highly variable over relatively short distances (Levine et al. 1994, Levine and Knox, in press), many wells would need to be installed at a variety of depths for each soil series. Moreover, temperature fluxes of northern soils yield freeze/thaw periods and spring snow melt and subsequent drawdown produced by vegetative growth that create significant seasonal variation of depths to water table. These temporal patterns require wells to be frequently monitored over many years to account for interannual changes in precipitation and temperature. To estimate soil moisture conditions for a variety of soil types, we used a soil physics model (Levine and Knox, in press) which simulates water and energy dynamics in a soil column given site-specific edaphic and climatic parameters. The primary objective of the present study was to evaluate the sensitivity of an individual-based forest succession model to model-derived soil moisture parameters for soil series comprising a range of drainage classes found at the northern hardwood/southern boreal transition zone in east-central Maine. The secondary objectives were to aggregate the output from the succession model for the different soil series to calculate and compare effects of waterlogging for a specific 600 ha landscape based on edaphic information obtained from local (1:12,000), county (1:120,000), and state (1:250,000) soil maps. 2. Methods 2.1 Study area International Paper's Northern Experimental Forest (NEF) in east-central Maine (45§15' N, 68§45' W) has served as the site of an intensive field and multisensor aircraft campaign that emphasized the use of remote sensing to study forested landscapes (Goward et al. 1994). The NEF comprises ~7000 ha consisting primarily of hemlock-spruce-fir (i.e., primarily Tsuga canadensis, Picea rubens, P. mariana, and Abies balsamea), aspen-birch (i.e., primarily Populus tremuloides, P. grandidentata, Betula papyrifera, and B. allegeniensis) and hemlock- hardwood mixtures. The terrain of the NEF is relatively flat, possessing a maximum elevation change of 135 m over 10 km, which decreases the rate of lateral drainage and promotes ponding during spring melting. The snow pack is typically continuous from December to March. Due to the region's glacial history and more recent alluvial events, soil drainage classes, defined as the rate at which water is removed from the soil, drastically vary from somewhat excessively drained eskers to very poorly drained topographic depressions that possess a compacted, impermeable horizon. Trees in excessively drained soils may suffer from drought effects, whereas trees in poorly drained soils are subject to anoxic conditions in the rooting zone. Because of the differential sensitivity of tree species to different drainage patterns, the NEF consists of a mosaic of stands of different species composition, biomass levels, and stem densities which are to a certain extent correlated to spectral reflectance, i.e., NDVI (Levine et al. 1994), and radar backscatter (Ranson and Sun 1994a, 1994b) patterns exhibited across the landscape. 2. 2 Soil moisture relationships in a forest model The Forest Ecosystem Dynamics modeling environment (Levine et al. 1993), conceived as a network of individual biotic and abiotic models to simulate remotely sensed patterns associated with forest succession, included a gap model, ZELIG (Urban 1990). Though for the most part employing the same methodology from early models developed over 25 years ago, this class of individual-based models is believed to offer a reasonable approach to predict long-term responses of forest vegetation to environmental change (Shugart et al. 1992, Malanson 1993a, Hinckley et al. 1994). ZELIG represents a generic form of gap model which was designed to serve as a readily adaptable framework for cross-site comparisons of forested systems. In its original published form (Urban 1990), ZELIG included only a positive effect of increasing soil moisture to reduce drought stress. On a monthly basis, the model estimates available soil moisture by calculating actual (AET) and potential evapotranspiration (PET) for a site for given edaphic conditions, i.e., field capacity (FC = cm water at 0.1 MPa) and permanent wilting point (PWP = cm of water at 1.5 MPa ), and climatic regimes, i.e., latitudinally related radiation, temperature, and precipitation (Thornthwaite and Mather 1957, Pastor and Post 1986). From FC and accumulated potential monthly water loss, the amount of water remaining in the soil- water column is determined. If PET > precipitation, AET is equal to precipitation and the loss of the water remaining from the previous month. To predict the effects of drought stress, a drought-day index, i.e., the proportion of days during the growing season in which soil moisture is below the PWP, is calculated annually. This index is incorporated into a growth multiplier which relates basal area growth to a given species-specific tolerance class (Pastor and Post 1986, Urban 1990). For the situation where PET < precipitation, AET = PET. Excess water is added to the soil-water column. If water depth exceeds the FC, it is treated as run-off. However, the terrain of the NEF is relatively flat and drainage tends to be slow. As a result, some soils remain saturated for periods exceeding the monthly time step. Such saturation promotes anaerobic conditions which can affect root respiration, pH, decomposition of organic matter, and for less tolerant tree species, inhibit establishment and growth and increase mortality rates. Following this Thornthwaite bucket model approach and with only a positive relationship between soil moisture and growth, ZELIG would have been incapable of reproducing the heterogeneity in forest composition and structure caused by soil saturation common to the NEF. To incorporate water saturation effects, we modified and incorporated a simple waterlogging growth multiplier described in Botkin (1993) to the gap model. The original equation depicts a site wetness factor for a species (WeFi) as: WeFi = max [0, 1 - (DTmini/DT)] (1) where DTmini = minimum distance to the water table tolerable for species i and DT is the average growing season depth to the water table. Because we believed that waterlogging- sensitive species were excessively penalized at depths to the water table well below the rooting zone, the growth multiplier was slightly modified to: WeFi = {0 if Dtmini > DT, (DT - DTmini)2/[DTmini2 + (DT - DTmini)2], otherwise}. (2) This change maintains the general form of the growth multiplier, but reduces the differences between tolerance classes at high DT's (Fig. 1). With this multiplier, there is a range of responses for the species pool available at the NEF (Table 1). Species can be grouped into waterlogging tolerance classes based on their DTmin's. Tolerant species with low DTmin's at the NEF include larch (Larix laricina) and northern white cedar (Thuja occidentalis). Intolerant species with high DTmin's include eastern white pine (Pinus strobus) and quaking aspen (Populus tremuloides). 2.3 Derivation of soil moisture parameters To derive soil hydrology parameters necessary for the forest succession simulations, we used a soil process model FroST (Frozen Soil Temperatures). The FroST model (Levine and Knox, in press) was developed as a modification of the "Residue" model described by Bristow et al. (1986) and Bidlake et al. (1992). FroST (Fig. 2) is similar to Residue in that it simulates the dynamic aspects of mass and energy transfer in a soil-vegetation-atmosphere system using numerical methods to describe fundamental physical processes. It differs from other models which predict permafrost presence in soil in a number of ways. Modifications to Residue present in the FroST model included changing all input climate requirements from daily to hourly, and the addition of algorithms to simulate snow dynamics, transpiration, and forest canopy conditions. Climatic input requirements include global short-wave radiation, air temperature, average wind speed, and precipitation. The model simulates short-wave and long-wave radiative transfer, soil and canopy temperature, canopy vapor density, rainfall interception by the canopy, infiltration redistribution, evaporation, freezing, thawing, and drainage. FroST is unique from ecosystem models which predict soil moisture or temperature flux such as TCX (Bonan 1991) or EXE (Martin 1992) in that it predicts hourly soil moisture, heat, and ice content in individual horizons of a given soil profile. Specific soil properties are defined to represent differences in thickness, particle size distribution, hydraulic conductivity, water holding capacity, air entry matric potential, and bulk density in each horizon. Mass and energy move through soil layers designated by nodes at specific depths. A canopy model is included in which the top three nodes represent forest overstory, and the bottom three represent litter. Radiative and convective transfer is described for each layer. Solutions to the mass and energy balance equations use the Newton Raphson technique to solve for temperature and matric potential of the soil and temperature and vapor density of the air in the canopy at the end of each hourly time step (Campbell 1985). Using these techniques, the model has been used to determine soil heat and water budgets and is especially useful for cold climates in which continuous or discontinuous ice is present (Levine and Knox, in press). For this study, data for soil series (see Table 2a) were obtained and an input file containing the relevant information for nodes at specific depths within horizons was created. Nodes were chosen either as midpoints within a soil horizon as described in the soil profile description or to represent a specific depth. Data for each soil node included: node depth (m), distance to next upper node (m), thickness of the soil layer (m), air-entry matric potential (J/kg), bulk density (Mg/m3), slope of the log-transformed moisture characteristic, saturated hydraulic conductivity (kg.s/m3), particle density (Mg/m3), clay mass fraction (%), initial soil temperature (§C), and initial volumetric water content (%). A saturated zone was designated at a specific depth for each soil based on the hydrologic group classification given in the soil survey. The groups are defined by the Soil Survey Division Staff (1993) according to the saturated hydraulic conductivity and the depth to internal "free water" (water table), where group A has the highest conductivity and deepest occurrence of internal free water, and group D has the lowest conductivity and the shallowest occurrence of internal free water. For purposes of setting up the starting conditions for FroST, soils classified as group A were given an initial saturated horizon of 12 m, group B at 2.5 m, group C at 1.0 m, and group D at the soil surface (see Table 2b). Input data required for the canopy nodes were reflectivity, zero plane displacement height (m), and momentum roughness length (m), which were obtained from structural characteristics of the vegetation at the NEF. Other site characteristic inputs included degrees north latitude and hour of solar noon. Climate data for the years 1988-1992 collected at the NEF were used to drive the model using inputs of: day of year, time of day, total incoming solar radiation (W/m2), atmospheric air temperature (§C), atmospheric relative humidity (%), wind speed (m/s), and daily precipitation (cm). 2.4 Simulations and scaling procedures Site characteristics (e.g., monthly values of temperature and precipitation) and autecological parameters (e.g., tree age, height and diameter maxima; growth and sapling establishment rates; light, water, fertility, and temperature tolerances) were derived from empirical data and published sources (e.g., Pastor and Post 1985, Botkin 1993, Weishampel et al. 1997) for the species found at the NEF. Because the model contains stochastic elements (e.g., weather, mortality, and recruitment) replicates with different random number seeds were averaged to estimate the system's behavior. Thirty runs, each consisting of nine, 10 x 10 m plots, were simulated starting from bare ground conditions for 14 soils series which comprise approximately 95% of the NEF (Table 2). To isolate soil moisture effects, differences in nutrient status of the soil series were not included. For comparison, half of the simulations did not include waterlogging effects and used laboratory estimates of FC, the other half included waterlogging effects and used estimates of depth to water table and effective FC predicted with FroST. The laboratory estimates of FC and PWP were derived from measurements of soil moisture at 0.03 MPa and 1.5 MPa, respectively. These were obtained by pressure-plate extraction techniques (USDA NRCS 1996) for soil samples of the predominant soil series found at the NEF (USDA SCS 1990). Aboveground productivity from the simulations for the different soil types were aggregated for a 600 ha area extracted from local (1:12,000), county (1:120,000), and state (1:250,000) soil maps. The local map was created as a soil survey of part of the NEF in 1990 by the USDA Soil Conservation Service in Orono, ME (USDA SCS 1990). The soil series distributions at the county scale were obtained by digitizing the soil maps associated with the study area from a 1963 USDA county soil survey of Penobscot County, Maine (Goodman et al. 1963). Additional information regarding the local and county soil maps is available at the Forest Ecosystem Dynamics world-wide website (Fifer 1996). At the state level, soil polygon data were obtained from the State Soil Geographic data base (STATSGO) map for Maine (USDA SCS 1993). Because land area represented by a given map increases as the scale decreases, less detail is depicted on smaller scale maps, i.e., maps with coarser resolution. Thus, generalized maps of counties or states rarely consist of mapping units comprised of a single soil series. More often, these mapping units represent associations of more than one soil series or other taxonomic unit (e.g., soil family, subgroup, or great group). The 600 ha map at the local scale consisted of eight soil associations consisting of twelve soil series; at the county scale, five soil series comprised four associations; and at the state scale, thirteen soil series comprised a single association (Fig. 3). The contribution of each association to the overall productivity of the landscape was weighted according to its areal extent. Within each association, the contribution of each series was weighted based on its percent composition. Table 2a gives the taxonomic classification, drainage class, and parent material of all the soil series used at each scale in this study. Because soil series not present in the local map existed in the smaller scale maps, analogs from the local map (Table 3) were substituted when appropriate. Three associations that comprised a small fraction (<1.25%) of the local and county scale data (and for which there was insufficient information to parameterize FroST) were excluded in the weighted aggregation. In the case of the STATSGO data, two lithic (shallow to bedrock) soils, Lyman and Monson, which together comprise 15% of the soil association, did not have an appropriate analog at the local scale. In order to include these soils in the scaling exercise, new soil hydrology parameters were estimated with FroST based on their description. 3. Results 3. 1 Predictions of soil water conditions For the 13 predominant soil series at the NEF (excluding the lithic soils), the sum of laboratory field capacities was correlated with average moisture levels in FroST simulations (R2 = 0.52, P < 0.01), but many soils retained more moisture in the multilayer dynamic simulations than would be expected from simple pressure plate measurements on isolated soil samples (Fig. 4). The somewhat excessively drained soil, Adams, and the very poorly drained soil, Biddeford, possessed the lowest and the highest moisture levels, respectively, in both the summed laboratory FC and the average simulated profile water content estimates. In general, the better drained soils tended to exhibit lower field capacities than the more poorly drained soils and poorly drained soils had the larger excesses over the expected from laboratory FC with free drainage. FroST yielded a range of depths to water table for the soil series (Table 2b). To a certain extent, these predictions corresponded to drainage class, i.e., well drained soils had deeper DT's than more poorly drained soils. The somewhat excessively to moderately well drained soils, were on average 2.8 m deeper than the more poorly drained soils. Though the average depth to water table for the somewhat poorly drained class was greater than the averages for the poorly and the very poorly drained classes as expected, the average for the poorly drained class was less deep than the average for the very poorly drained class. Laboratory data for the Kinsman series were from a somewhat poorly drained profile rather than the very poorly drained "taxadjunct" Kinsman soils mapped at the NEF. Peacham soils occur in depressions with better drained series and may be difficult to simulate correctly without lateral water movement. 3.2 Waterlogging effects on species composition When running the baseline ZELIG without the waterlogging multiplier, edaphic responses are a product of drought from FC and PWP estimates. However, given the cold, humid (80-100 cm precipitation/year), continental climate of the NEF, typically AET = PET, thus species trajectories were very similar for the different soil drainage classes (Fig. 5a). Intermediately waterlogging tolerant species, primarily red spruce (Picea rubens) and paper birch (Betula papyrifera), dominated the stand throughout the 200 year simulation. Waterlogging intolerant species outcompeted tolerant species during the early stages of succession, but were supplanted by waterlogging tolerant species after ~75 years. Overall basal area on the somewhat excessively drained soil was reduced probably due to slight drought effects. When including the waterlogging multiplier, species trajectories varied based on the drainage class of the soil (Fig. 5b) as found with earlier simulations (Botkin et al. 1989, Botkin and Nisbet 1992a,b, Botkin 1993) based on the Botkin and Levitan (1977) multiplier. The well drained soils revealed little difference from those in the previous slide without waterlogging. But as drainage and depth to water table decreased, the species trajectories changed so that the importance of more waterlogging tolerant species increased. For the somewhat poorly drained soil, tolerant species accounted for roughly a third of the entire basal area after 100 years. In the scenario with the very poorly drained soil, tolerant species, primarily northern white cedar (Thuja occidentalis), were the only ones which persisted which is consistent with field data collected from the NEF (Levine et al. 1994). 3.3 Waterlogging effects on aboveground biomass Without waterlogging, the 14 soils at the NEF all exhibited a similar sigmoidal pattern of above ground biomass accumulation (Fig. 6a). Though the biomass trajectories were similar, levelling off at ~100 years, the amount of biomass at 200 years ranged between 100 and 200 Mg/ha. These differing biomass levels are primarily related to evapotranspirative-induced drought effects. The correlation coefficient between the average biomass levels between 50 and 200 yr and the differences between FC and PWP was 0.86 (P<0.001). The difference between FC and PWP, however, does not necessarily correspond to drainage class. For example, the difference in FC and PWP for the Plaisted series, although it is well drained due to its gravelly loam texture, is comparable to a poorly drained soil series such as Monarda which is classified as a silt loam. With waterlogging, drought effects were reduced in most soils as biomass trajectories segregated according to drainage classes. More poorly drained soils were less productive than better drained soils during the first 100 years of succession (Fig. 6b). One very poorly drained soil series, Biddeford, which corresponds to open bogs in the NEF landscape, possessed a depth to water table estimated at 0 m and yielded no tree growth as expected with the waterlogging function. 3.4 Scaling effects While the general pattern from waterlogging was consistent for the data from the different scaled maps, the effect was more exaggerated for projections based on the local and state soil maps which had overall lower levels of aboveground biomass than projections based on the county soil map (Fig. 7). When we aggregated simulations based on the areal extent of soil series across the 600 ha landscape, the effect of waterlogging amounted up to a ~40% reduction for the local-scale and state-scale projections and up to a ~20% reduction for the county level projections in aboveground biomass during the first 50 years. However, after 125 years, the aboveground biomass from the simulations with waterlogging converged with the simulations without waterlogging. This general waterlogging effect was fairly consistent among the various scales. The absolute levels of biomass varied substantially, maintaining up to a ~40% difference between the projections based on state and county data. The soil composition information from the county map (1:120,000 scale) yielded the highest biomass levels, nearly 200 Mg/ha after 200 years. The soil composition information from the state map (1:240,000 scale) yielded the lowest biomass levels, approximately 120 Mg/ha at 200 years. The biomass level derived with the information from the local map approached 160 Mg/ha at 200 years. 4. Discussion Successional pathways of southern boreal-northern hardwood forests are highly dependent on edaphic conditions which tend to be fairly heterogeneous (Pastor and Mladenoff 1992, Levine et al. 1994). Most global warming scenarios predict increased temperatures and evaporation of free water for this transition zone which should lead to warmer substrates and lowered water tables (Price and Apps 1993). Thus, the inclusion of water table dynamics may be critical to how forests occupying poorer draining soils respond. Although changes in depth to water table were not included as part of the climate change scenarios (Botkin et al. 1989, Botkin and Nisbet 1992a, Botkin 1993), a sensitivity analysis of a gap model by Botkin and Nisbet (1992b) showed slight changes (ñ10%) in DT produced little change in forest structure for a given soil. The parameters derived from the soil physics model for the variety of soils found at the NEF were substantially different and yielded dramatic differences in species composition and aboveground biomass levels. The differences in depths to the water table from the somewhat excessively drained, Adams, to the very poorly drained, Biddeford, as well as among most other soil series were much larger than the 2-6 cm differences tested by Botkin and Nisbet (1992b). Based on this study, it would seem that if depths to water table were to decrease to levels found in drier soil series, denizen forests would exhibit a change in species composition which corresponds to an increase in aboveground productivity during early stages of succession. Hence, more poorly drained regions may better function as aboveground carbon sinks with increasing temperatures and drier conditions. One of the major criticisms raised by the use of gap models to predict effects of climate change of forests is the unrealistic mechanism for simulating species specific temperature responses (Urban and Shugart 1989). Like the temperature relationship, species responses to waterlogging stress has been to use crude parabolic (Phipps 1979, Pearlstine et al. 1985) or sigmoidal relationships (Botkin and Levitan 1977, Botkin and Nisbet 1992) that relates to the optimal depth to water table for a species. However, as with all the growth multipliers, each species probably possesses their own uniquely shaped soil saturation response curve (Malanson 1993b). Furthermore, field measurements of forested wetlands in Minnesota found aboveground net primary productivity and biomass to be more sensitive to fluctuations in groundwater than to average depth to the water table (Grigal and Homann 1994). Thus, the general waterlogging growth multiplier approach using an annual time increment and an average DT may be less appropriate than a seasonally or monthly changing water table. In order to fully account for changes in evapotranspiration, a more complete coupling of the hydrological and ecological dynamics would be needed to adjust for shifts in AET associated with successional changes in the forest composition. When an evapotranspirative feedback was included from a gap model to a biophysical atmospheric model (Martin 1992), it was found that soil drying increased dramatically causing a shift from forests to grasslands. This implies that water table drawdown by evapotranspiration could potentially promote a change from waterlogging tolerant to less tolerant species as found on the better drained soils. Thus, there is a need to create physiologically tractable growth relationships (Bonan 1993, Bugmann and Martin 1995) in this class of individual based model perhaps along the line of Friend et al. (1993) which relate to soil saturation effects. There was no a priori expected result from this scaling exercise, other than it was not expected that "scaled-up" results would be consistent as the mapping resolution changed. Certain local areas may be wetter or drier on average than others and such fine-scale heterogeneity may not be captured with coarser mapping units. Based on the aggregated landscape, regardless of scale, the effects of waterlogging on aboveground biomass were most apparent during the early periods of succession, i.e., up to 100 years. (This model did not include differences in fertility among these soils.) Afterwhich, aboveground biomass trajectories from the aggregate landscape that included waterlogging effects converged with that which did not. This consistency was surprising as the edaphic information regarding the areal extent of drainage classes from the three different soil maps varied considerably (Table 3). At the local scale, poorly and very poorly draining soils comprised <50% of the 600 ha NEF area. Compared to the local soil map, the state map overestimated somewhat excessively drained soils for the 600 ha landscape. And compared to the local soil map, the county map underestimated and overestimated the extent of poorly and well-drained soils, respectively In this exercise, it was found that the choice of which geospatial information was used was more critical than whether or not soil saturation effects were included. As a result of how soils for an arbitrarily chosen 600 ha area from the NEF were classified, the model predictions based on data from county soil map overestimated and predictions based on the state soil map underestimated productivity levels when compared to predictions based on the more detailed local soil map. This 40% range in productivity predictions demonstrates how aggregation error from decreasing map scale is propagated by simulations models which rely on the map data. Furthermore, the fact that biomass levels increased as map resolution decreased from 1:12,000 to 1:120,000 then decreased as map resolution decreased from to 1:120,000 to 1:250,000 suggests that there are no simple scaling rules for soil heterogeneity which could be applied for this area. Acknowledgments Research was supported in part by an National Research Council (NRC) Research Associateship and by funding for the Forest Ecosystem Dynamics (FED) project from NASA's Terrestrial Ecosystem and Modeling Program. The authors would like to thank Steve Fifer for assisting with the geographical analysis and Dean Urban with help in estimating tree species parameters and providing the ZELIG source code. References Bidlake, W. R., Campbell, G. S., Papendick, R. I. and Cullum, R. F. 1992. Seed zone temperature and moisture conditions under conventional and no-tillage in Alaska. Soil Sci. Am. J. 56: 1904-1910. Bonan, G. B. 1989. Environmental factors and ecological processes controlling vegetation patterns in boreal forests. Landscape Ecology 3: 111-130. Bonan, G. B. 1991. Atmosphere-biosphere exchange of carbon dioxide in boreal forests. J. Geophys. Res. 96D: 7301-7312. Bonan, G. B. 1993. Do biophysics and physiology matter in ecosystem models? Climatic Change 24: 281-285. Bonan, G. B. and Shugart, H. H. 1989. Environmental factors and ecological processes in boreal forests. Ann. Rev. Eco. Syst. 20: 1-28. Bonan, G. B., Shugart, H. H. and Urban, D. L. 1990. The sensitivity of some high-latitude boreal forests to climatic parameters. Climatic Change 16: 9-29. Botkin, D. B. 1993. Forest dynamics: an ecological model. Oxford Univ. Press, New York. Botkin, D. B. and Levitan, R. E. 1977. Wolves, moose and trees: an age specific trophic- level model of Isle Royale National Park. IBM Research Report in Life Sciences RC 6834. Botkin, D. B. and Nisbet, R. A. 1992a. Projecting the effects of climate change on biological diversity in forests. In Consequences of the greenhouse effect for biological diversity. pp. 277- 293. Edited by R. Peters. Yale Univ. Press, New Haven, Connecticut. Botkin, D. B. and Nisbet, R. A. 1992b. Forest response to climatic change: effects of parameter estimation and choice of weather patterns on the reliability of projections. Climatic Change 20: 87-111. Botkin, D. B., Nisbet, R. A. and Reynales, T. E. 1989. Effects of climate change on forests of the Great Lake States. In The potential effects of global climate change on the U. S., Appendix D - Forests. Edited by J. B. Smith and D. A. Tirpak. U. S. EPA, Washington DC. Bristow, K. L., Campbell, G. S., Papendick, R. I. and Elliot, L. F. 1986. Simulation of heat and moisture transfer through a surface residue-soil system. Agric. For. Meterol., 36: 193-214. Bugmann, H. and Martin, P. 1995. How physics and biology matter in forest gap models. Climatic Change 29: 251-257. Bugmann, H. K. M. and Solomon, A. M. 1995. The use of a European forest model in North America: a study of ecosystem response to climate gradients. J. Biogeogr. 22: 477-484. Campbell, G. S. 1985. Soil physics with BASIC: transport models for soil-plant systems. Elsevier Pub. New York. Fifer, S. 1996. Forest Ecosystem Dynamics (FED) welcome page: Available as: http://fedwww.gsfc.nasa.gov. Friend, A. D., Shugart, H. H. and Running, S. W. 1993. A physiology-based gap model of forest dynamics. Ecology 74: 792-797. Glebov, F. Z. and Korzukhin, M. D. 1992. Transitions between boreal forest and wetland. In A systems analysis of the global boreal forest. pp. 241-266. Edited by H. H. Shugart, R. Leemans, and G. B. Bonan. Cambridge Univ. Press, New York. Goodman, V., Riley, R. M., Whitney, B. A., Arno, J. R., La Flamme, K. J., Von Day, S., Hardesty, J. S., Lovejoy, D. B. and Worcester, B. 1963. Soil Survey of Penobscot County, Maine. U.S. Govt. Printing Office, Washinton, D. C. Goward, S. N., Williams, D. L. and Peterson, D. L. 1994. NASA multisensor aircraft campaigns for the study of forest ecosystems. Remote Sens. Environ. 47: 107-108. Grigal, D. F. and Homann, P. S. 1994. Nitrogen mineralization, groundwater dynamics, and forest growth on a Minnesota outwash landscape. Biogeochemistry 27: 171-185. Hinckley, T. M., Sprugel, D. G., Batista, J. L. F., Brooks, J. R., Brubaker, L. B., Compton, J. Erickson, H. E., Little, R. L., Maguire, D. A., McCarter, J. B., McKay, S. J., Pass, D., Peterson, D. W., Reed, J. P., Tacey, W., Wilkinson, L. E. and Whytemare, A. 1994. Use of the JABOWA family of individual-tree based models for exploration of forest responses to global climate change. Report for the National Council on Air and Stream Improvement. King, A. W. 1980. Translating models across scales in the landscape. In Quantitative methods in landscape ecology. pp. 479-517. Edited by M. G. Turner and R. H. Gardner. Springer- Verlag, New York. Levine, E. R., Ranson, K. J., Smith, J. A., Williams, D. L., Knox, R. G., Shugart, H. H., Urban, D. L. and Lawrence, W. T. 1993. Forest ecosystem dynamics: linking forest succession, soil process and radiation models. Ecol. Model. 65: 199-219. Levine, E. R., Knox, R. G. and Lawrence, W. T. 1994. Relationships between soil properties and vegetation at the Northern Experimental Forest, Howland, Maine. Remote Sens. Environ. 47: 231-241. Levine, E. R. and Knox, R. G. Modeling soil temperature and snow dynamics in northern forests. J. Geophys. Re. BOREAS Special Issue. (in press) Malanson, G. P. 1993a. Comment on modeling ecological response to climatic change. Climatic Change 23: 95-109. Malanson, G. P. 1993b. Riparian landscapes. Cambridge Univ. Press, New York. Martin, P. 1992. EXE: a climatically sensitive model to study climate change and CO2 enhancement effects on forest. Aust. J. Bot. 40: 717-735. Pastor, J. and Mladenoff, D. J. 1992. The southern boreal-northern hardwood forest border. In A systems analysis of the global boreal forest. pp. 216-240. Edited by H. H. Shugart, R. Leemans, and G. B. Bonan. Cambridge Univ. Press, New York. Pastor, J. and Post, W. M. 1986. Influence of climate, soil moisture, and succession on forest carbon and nitrogen cycles. Biogeochemistry 2: 3-27. Pastor, J. and Post, W. M. 1988. Response of northern forest to CO2-induced climate change. Nature 334: 55-58. Pearlstine, L., McKellar, H. and Kitchens, W. 1985. Modelling the impacts of a river diversion on bottomland forest communities in the Santee River floodplain, South Carolina. Ecol. Model. 29: 283-302. Phipps, R. L. 1979. Simulation of wetlands forest vegetation dynamics. Ecol. Model. 7: 257-288. Price, D. T. and Apps, M. J. 1993. Integration of boreal ecosystem-process models within a prognostic carbon budget model for Canada. World Res. Rev. 5: 15-31. Ranson, K. J. and Sun, G. 1994a. Mapping biomass of a northern forest using multifrequency SAR data. IEEE Trans. Geosci. Rem. Sens. 32: 388-396. Ranson, K. J. and Sun, G. 1994b. Northern forest classification using temporal multifrequency and multipolarimentric SAR images. Rem. Sens. Environ. 47: 142-153. Shao, G., Schall, P. and Weishampel, J. F. 1994. Dynamic simulations of mixed broadleaved- Pinus koraiensis forests in the Changbaishan Biosphere Reserve of China. For. Ecol. Manage. 70: 169-181. Shugart, H. H., Smith, T. M. and Post, W. M. 1992. The potential for application of individual-based simulation models for assessing the effects of global change. Ann. Rev. Ecol. Syst. 23: 15-38. Soil Survey Division Staff. 1993. Soil survey manual. USDA Handbook No. 18. U.S. Gov't Printing Office, Washington, D.C. Solomon, A. M. 1986. Transient response of forests to CO2-induced climate change: simulation modeling experiments in eastern North America. Oecologia 68: 567-579. Solomon, A. M. 1992. The nature and distribution of past, present and future boreal forests: lessons for a research and modeling agenda. In A systems analysis of the global boreal forest. pp. 291-307. Edited by H. H. Shugart, R. Leemans, and G. B. Bonan. Cambridge Univ. Press, New York. Thornthwaite, C. W. and Mather, J. R. 1957. Instructions and tables for computing potential evapotranspiration and the water balance. Publications in Climatology 10: 183-311. Urban, D. L. 1990. A versatile model to simulate forest pattern: a user's guide to ZELIG version 1.0. University of Virginia, Charlottesville, VA. Urban, D. L. and Shugart, H. H. 1989. Forest response to climatic change: a simulation study for southeastern forests. In The potential effects of global climate change on the U. S., Appendix D - Forests. Edited by J. B. Smith and D. A. Tirpak. U. S. EPA, Washington DC. USDA NRCS. 1996. Soil survey laboratory methods manual. Soil Survey Investigations Report No. 42., Ver. 3.0. USDA, Natural Resources Conservation Service, National Soil Survey Center, Lincoln, NE. USDA SCS. 1971. Handbook of soil survey investigations field procedures. U. S. Gov. Printing Office, Washington, DC. USDA SCS. 1990. Soil survey and interpretations of selected land in Maine. International Paper Company Experimental Forest, Howland, Maine. USDA, SCS, Orono, ME. USDA SCS. 1993. State soil geographic data base (STATSGO): Data users guide. Misc. Pub. No. 1492. USDA, Natural Resources Conservation Service, Lincoln, NE. Weishampel, J. F., Knox, R. G., Ranson, K. J., Williams, D. L. and Smith, J. A. 1997. Integrating remotely sensed spatial heterogeneity with a 3-D forest succession model. In The use of remote sensing in the modeling of forest productivity at scales from the stand to the globe. pp. 109-133. Edited by H. L. Gholz, K. Nakane, and H. Shimoda Kluwer Acad. Pub. Dordrecht, Netherlands. Table 1. Tree species parameters used in implementation of ZELIG for the NEF. Scientific name (Common name) Age DBH H GR GDD- GDD+ DTmin L M N Abies balsamea (Balsam fir) 200 50 1500 69 250 2404 0.211 1 1 3 Acer pensylvanicum (Striped maple) 75 15 500 150 889 5500 0.567 2 3 3 Acer rubrum (Red maple) 150 100 3000 176 1260 6601 0.322 2 3 3 Acer saccharum (Sugar maple) 300 150 3000 89 1204 3200 0.567 1 2 2 Betula allegeniensis (Yellow birch) 250 75 3000 106 1420 3084 0.600 3 2 2 Betula papyrifera (Paper birch) 140 100 2500 160 700 2500 0.544 4 3 3 Betula populifolia (Gray birch) 250 25 1000 37 1007 2880 1.000 5 2 3 Fagus grandifolia (American beech) 366 100 3000 72 1327 5556 0.489 1 3 2 Larix laricina (Tamarack) 335 75 2500 66 280 2660 0.156 5 3 3 Picea mariana (Black spruce) 250 40 2000 70 265 1929 0.156 2 3 3 Picea rubens (Red spruce) 300 100 3000 89 500 2580 0.489 2 2 3 Pinus strobus (Eastern white pine) 450 150 3500 68 1500 3183 1.000 3 3 3 Populus grandidentata (Bigtooth aspen) 70 75 2500 316 1100 3169 0.400 5 3 2 Populus tremuloides (Quaking aspen) 125 75 2200 158 889 5556 0.700 5 3 2 Thuja occidentalis (Northern white cedar) 400 100 2400 55 1000 2188 0.100 2 3 3 Tsuga canadensis (Eastern hemlock) 650 150 3500 47 1324 3100 0.489 1 2 3 Variables are as follows: Age=maxium age (yr); DBH (cm); H=height (cm); GR=Growth rate (dimensionless); GDD- and GDD+=minimum and maximum growing degree-days (5.56§C base); DTmin (m); L=shade tolerance (rank 1=very tolerant); M=drought tolerance (1=least tolerant); N=nutrient stress tolerance (1=least tolerant) Table 2a. Dominant soil series found at the NEF at local, county and state scales. Series Drainage Class Parent Material Adams1 (Typic Haplorthod) Somewhat Excessively Outwash Biddeford2 (Histic Humaquept) Very Poorly Marine Sediments Boothbay1 (Aquic Eutrochrept) Somewhat Poorly Lacustrine and Marine Sediments Brayton3,4 (Aeric Haplaquod) Poorly Till Burnham2,3 (Typic Haplaquept) Somewhat Poorly Till Buxton3,4 (Dystric Eutrochrept) Somewhat Poorly Lacustrine and Marine Sediments Colonel1 (Aquic Haplorthod) Somewhat Poorly Till Croghan1 (Aquic Haplorthod) Somewhat Poorly Outwash Dixfield1 (Typic Haplorthod) Moderately Well Till Hermon3,4 (Typic Haplorthod) Somewhat Excessively Outwash Howland2,3 (Aquic Haplorthod) Somewhat Poorly Till Kinsman1 (Aquic Haplorthod) Very Poorly Outwash Lyman4 (Lithic Haplorthod) Somewhat Excessively Bedrock Marlow1 (Typic Haplorthod) Well Till Masardis3,4 (Typic Haplorthod) Somewhat Excessively Outwash Medomak5 (unknown) Very Poorly Alluvium Monarda2 (Aeric Haplaquod) Poorly Till Monson4 (Lithic Dystrochrept) Somewhat Excessively Bedrock Peacham1 (Histic Humaquept) Very Poorly Till Plaisted2 (Typic Haplorthod) Well Till Roundabout5 (unknown) Poorly Drained Alluvium Scantic1 (Typic Haplaquept) Poorly Lacustrine and Marine Sediments Tunbridge3,4 (Entic Haplorthod) Moderately Well Till Westbury1 (Aeric Haplaquod) Poorly Till Wonsqueak5 (unknown) Very Poorly Organic Materials 1Data derived from Soil Survey of NEF (USDA SCS 1990) 2Data dervied from Penobscot County Soil Survey Report (Goodman et al. 1963) 3Used input file from similar soil (see Tables 3b and 3c) 4Data derived from STASGO data base (USDA SCS 1993) 5Not included in this study Table 2b. Hydrologic soil grouping, measured PWP and predicted DT for selected soil series used as inputs to FroST. Series Hydrologic Group Summed Laboratory PWP (cm) Modeled DT (m) Adams A 4.8 4.00 Biddeford D 22.1 0.00 Boothbay C 12.8 0.35 Colonel C 9.6 1.23 Croghan B 10.3 1.05 Dixfield C 8.1 3.4 Kinsman C 10.8 0.84 Marlow C 8.7 3.3 Monarda C 6.3 0.34 Peacham D 15.0 0.62 Plaisted A 3.8 3.0 Scantic C 18.2 0.25 Westbury C 5.7 0.71 Lithic* A 2.9 10.00 *Category included to account for Lyman and Monson series identified at the state level in the STATSGO database without an analog at local or county levels. Table 3a. Composition of 6 ha NEF area using information from the 1:12,000 local soil map (USDA SCS, 1990). Soil Association Percent composition of landscape Series comprising soil association Percent composition of association** Colonel-Westbury 50.61 Colonel 50 Westbury 50 Colonel-Dixfield 21.59 Colonel 50 Dixfield 50 Peacham-Westbury 8.08 Peacham 70.6 Westbury 29.4 Boothbay-Scantic 6.05 Boothbay 50 Scantic 50 Croghan-Kinsman 5.22 Croghan 56.3 Kinsman 43.7 Colonel-Dixfield-Marlow 4.83 Colonel 17.6 Dixfield 47.1 Marlow 35.3 Biddeford 2.45 Biddeford 100 Bucksport-Wonsqueak* 0.93 Bucksport 60 Wonsqueak 40 Roundabout-Medomak* 0.24 Roundabout 35 Medomak 65 *not included in aggregation **does not account for inclusions Table 3b. Composition of 6 ha NEF area using information from the 1:120,000 county soil map (Goodman et al. 1963). Analogs for series are in parentheses. Soil Association Percent composition of landscape Series comprising soil association Percent composition of association** PrC 58.41 Plaisted 100 MrB 32.46 Monarda 50 Burnham (Colonel) 50 HvB 8.60 Howland (Croghan) 100 Mu* 0.54 Muck 100 *not included in aggregation**does not account for inclusions Table 3c. Composition of 6 ha NEF area using information from the 1:250,000 state soil map (USDA SCS 1993). Analogs for series are in parentheses. Soil Association Percent composition of landscape Series comprising soil association Percent composition of association** ME008 100 Brayton (Westbury) 20 Dixfield 12 Hermon (Adams) 11 Lyman (Lithic) 11 Peacham 11 Biddeford 8 Colonel 7 Marlow 6 Monson (Lithic) 4 Buxton (Boothbay) 3 Scantic 3 Masardis (Adams) 2 Tunbridge (Dixfield) 2 **does not account for inclusions Figure Legends Fig. 1. Comparison of species growth responses for a) Botkin (1993) and b) modified waterlogging growth multipliers. Fig. 2. Schematic of fluxes in the soil physics model FroST version 1.1. Fig. 3. Spatial distribution of NEF soil associations for a 600 ha area from the a) local (1:12,000), b) county (1,120,000), and c) state (1:125,000) maps. The gray shades within each 2 x 3 km rectangle correspond to associations listed in Table 3. Fig. 4. Comparison between measured and modeled field capacities for the predominant soil series (excluding the lithic soils) at the NEF. Fig. 5. Average species trajectories for three drainage classes for simulations a) without and b) with the waterlogging growth multiplier. The tolerant class includes all NEF tree species with DTmin = 0.3 m; for intermediate, 0.6 m > DTmin > 0.3 m; for intolerant, DTmin = 0.6 m. Fig. 6. Average aboveground biomass trajectories for the 14 predominant soil series a) without and b) with the waterlogging growth multiplier. Fig. 7. Aboveground biomass trajectories aggregated to a 600 ha area based on soil information from maps of the local (1:12,000), county (1,120,000), and state (1:125,000) scales for simulations with and without waterlogging.