Contemporary Carbon Dynamics in Terrestrial Ecosystems in the Southeastern Plains of the United States

Appendices

Appendix 1: Modification of CENTURY

Simulation of Net Primary Production as a Function of Forest Age

In the CENTURY model, NPP is not explicitly simulated as a function of forest age. Field measurements on forest growth indicate that C or biomass accumulations in young stands, particularly plantations, are highly related to forest age (Oliver 1981, Gholz and Fisher 1982, Mencuccini and Grace 1996). Biomass accumulation for plantations that are close to rotation age and for most natural mature forests will tend toward zero, with annual aboveground net primary production becoming almost entirely composed of litterfall and whole tree mortality (Ryan and others 1997).

Mencuccini and Grace (1996) studied a chronosequence of Scots pine plantations with age ranging from 7 to 59 years old. The leaf area index (LAI) reached a maximum at 20 years old when the canopy closure began. By age 40 the overstory LAI started to decrease, and LAI at age 59 was about half of that at age 20. Aboveground net primary production (ANPP) initially increased with canopy LAI, but decreased as the stands aged. Based on Meldahl and others (1998), excluding all the locations where forest ages differed by more than 5 years among plots, we found that ANPP and needle NPP decrease exponentially from about 20 to 110 years of age. The ANNP decreased at a rate faster than the needle NPP, suggesting that aboveground woody NPP was decreasing faster than the needle NPP. This pattern is general for most even-aged forests (Ryan and others 1997). The study of Meldahl and others also indicated that needle fall decreases exponentially with age from about 20 to 110 years old. The decline of fine litterfall during forest aging has been observed in other forest chronosequences as well (Waring and Running 1998). In this study, we assume that the gradual reduction of fine litterfall during aging observed at the Meldahl et al study sites applies to the forests in U.S. Southeastern Plains.

Simulation of Temporal Changes in Allocation

The fraction of carbon allocated to each tree component (i.e., leaves, branches, stems, and fine and coarse roots) changes as a forest grows (Gholz and others 1986, Waring and Running 1998, Ryan and others 1997, Meldahl and others 1998). Initially, trees tend to allocate more photosynthates to leaves or needles and fine roots to maximize the capture of solar energy and absorbance of soil nutrients. The fractions to leaves and fine roots decrease gradually with time as more and more photosynthates are allocated to woody tissues (e.g., branches, stems, and coarse roots). After reaching a maximum value, the allocation to woody tissues decreases gradually while at the same time the fractions to leaf and fine root increase gradually. Finally, the allocation fractions reach an equilibrium condition.

The original CENTURY used only two values to represent the change of allocation fraction to a specific tree component as a forest grows from young to mature. For example, allocation fraction to stems changed suddenly from 0.10 to 0.30 in the temperate mixed forest (i.e., TMPMX in CENTURY code) when a certain mature year was reached. Because most of the forests in the Southeastern Plains are relatively young and growing, it is necessary to modify the changes of allocation coefficients with curvilinear functions to better simulate the carbon accumulation in these dynamic forests. In this study, we replaced the stepwise two-value functions of CENTURY with regression equations derived from Meldahl and others (1998) to represent the dynamic changes of allocation coefficients in the forests.

Appendix 2: Data Assimilation

Initializing Forest Age and Biomass with FIA Data

The initial structure of forest age is critical for simulating carbon dynamics at the regional scale because many characteristics of forests, including standing biomass, NPP, and allocation pattern, are closely related to forest age. In this study, we assumed that the initial forest age structure within the sample blocks follows the statewide age structure characterized by the FIA databases. If the initial land cover of a site in 1973 was forest, then its initial age is randomly assigned according to the age structure of the forests of that state. Because the time (i.e., year) when the forest inventory data were collected might not be the same as the initial time of the model simulation, an offset time (i.e., the difference between the FIA inventory year and the starting year of the simulation) was added to adjust the age estimate. Once the initial age of the forest was estimated, the corresponding standing biomass was estimated based on the relationship between age and biomass as derived from FIA data.

Initializing Soil Texture, SOC, Bulk Density and Drainage with STATSGO Data

Once a soil component was determined, information regarding drainage condition, soil layers, soil texture, water holding capacity, bulk density, and soil organic matter content (converted to SOC using a factor of 0.58) were retrieved from the STATSGO attribute databases. For the variables with high (V1) and low (V2) values, the following procedure was used to assigned a value to minimize potential bias (Reiners and others 2002, Pierce and Running 1995):

where NORM is the standardized normal distribution. The above treatment assumes that the possible values of the soil characteristic follows a normal distribution with 95% of the values varying between V2 and V1. Considering the possible nonlinear impacts of texture fractions on biogeochemical cycles, a Monte Carlo approach was used to assign fractions of sand, silt, and clay in their corresponding possible ranges rather than using the mean fractions of the texture class specified in the USDA soil texture classification system.

Constructing Land Use History by Assimilating Remote Sensing, Census, and Inventory Data

It is necessary to disaggregate the agricultural land class into a combination of specific crop types for biogeochemical modeling because each type of crop has distinct biological characteristics and management practices, leading to differing impacts on carbon dynamics in vegetation and soils. Disaggregation of the agricultural land class was done stochastically based on crop composition statistics at the county level derived from USDA NRI agricultural census data (http://www.nrcs.usda.gov/technical/NRI/ ). The NRI database is a statistically based sample of land use and natural resource conditions and trends on US nonfederal lands. The inventory, covering about 0.8 million sample points across the country, was done once every five years.

Another task in generating LUCC sequences is to fill the gaps between consecutive land cover maps (i.e., the remotely sensed data). This was accomplished with crop rotation probabilities calculated from the NRI databases.

Our land cover change data was from the US Land Cover Trends (LCT) project (Loveland and others 2002). The LCT project was not specifically designed to capture all the forest harvesting activities. Optical reflectances at the clear-cut sites become indistinguishable from those of the surrounding mature forests 3 years after clear-cutting in the Southeastern Plains owing to fast recovery of understory and trees. The LCT data do not include selective harvesting activities. We derived the probability and intensity of selective cutting from FIA databases. The mean annual clear-cutting probability during a specific time period within a block was calculated using the LCT map by dividing the total mapped clear-cut area by 3 years. The probabilities of selective and clear-cutting were then used to stochastically schedule additional forest harvesting events that were not reflected in the LCT land cover maps. It was assumed in the model that a minimum age of 20 years was required for scheduling any harvesting activities in a forest.