Simulating the effects of urbanization, afforestation and cropland abandonment on a regional carbon balance: a case study for Central Germany

Abstract

The newly developed model system HILLS is used to simulate recent (1990–2000) and future (up to 2020) changes in land use and carbon sequestration over Central Germany. HILLS is unique as that it integrates the spatially explicit land-use-change model LUC-Hesse with the dynamic ecosystem model Century under a GIS platform. With this new tool, the concurrent effects of urbanization, afforestation and cropland abandonment on regional carbon sequestration are analyzed for an exemplary “Business as Usual” scenario. During the simulation period, afforestation was estimated to sequester 880 Gg C and cropland abandonment 783 Gg C. Urbanization was estimated to release 336 Gg C formerly stored in soil organic matter and thereby offsets about 20% of the C sequestered by cropland abandonment and afforestation. The case study shows that urbanization can partly counteract the benefits of carbon sequestration resulting from other land-use changes and should be investigated in other carbon balances.

Appendix A Calibration and testing of the afforestation component of the carbon-module

Growth calculations

In this appendix we describe the calibration and testing of the component of the GIS-Century model that concerns forest growth calculations. Forest growth calculations are used in this paper to compute carbon fluxes in afforested areas once they are established. Since the calibration and testing of these forest growth calculations have not been published elsewhere, we describe them here.

The purpose of the calibration is to identify a set of parameters for GIS-Century that simulate the growth processes of newly established broadleaf forests (the most common type of planted tree in Hesse). Because of the lack of spatially explicit forest stand data, the model is tested with indirect indicators of forest growth, namely by comparing the simulated and observed increment of carbon stored annually in tree biomass. This comparison is carried out for an undisturbed forest rotation period (balance period), estimated from Hessian forest census data to be approximately 160 years. Observed data for this comparison are census data on yearly increment of coarse-wood (defined by Kramer 1988 as stems and branches with a diameter of more than 7 cm) which are converted to carbon equivalents using Eq. 1. Below we examine the accuracy of assuming that the increment of coarse-wood from census data are comparable to the increment of carbon stored in tree biomass.

$$ dcarbon\, = \,{\sum\limits_i {dcoarsewood_{i} } }\, \times \,wdensity_{i} \, \times \,wcontent_{i} \, \times \,warea_{i} $$

(1)

where

dcarbon :

carbon increment per year of tree species i [g/m² a]

dcoarsewood :

coarse wood increment of tree species i [m³/m² a]

wdensity :

wood density of tree species i [g/m³]

wcontent :

wood carbon content of tree species i [0...1]

warea :

share of broadleaf forest area of tree species i [0...1]

Forest growth is simulated on the grid cell level using local soil and climate data. To calibrate model parameters an initial estimate of vegetation and soil carbon pools is obtained from a spin-up simulation consisting of two steps: (1) a 1,000-year run of GIS-Century assuming a forested land cover, followed by a clear cutting event where 99% of the above-ground biomass is removed from the site, and (2) a 160-year rotation period, again followed by a clear cutting event. After this initialization process, the simulation of the balance period is carried out.

Calibration procedure

Calibration of the forest model is conducted with data from the Darmstadt-Dieburg district in the southern part of Hesse. The procedure consists of six steps performed iteratively until a feasible combination of parameter values is found:

• Make first estimate of initial parameter set.

• Calculate forest growth for all broadleaf forest cells of the district.

• Calculate mean yearly increment of carbon storage in the tree biomass component during the balance period for these grid cells.

• Calculate mean value of increment from all simulated broadleaf cells on district level.

• Compare district level results to census data and compute performance measure of results.

• Based on value of performance measure, change parameter values if necessary.

The calibration starts with a standard parameter set for broadleaf trees from Parton et al. (2001). During the calibration procedure, the parameters for maximum gross production (prdx2), the effect of temperature on production (ppdf) and the turnover rates for the wood carbon pools (wooddr) are varied within acceptable ranges. The mean absolute percent error (MA%E) is used as a performance measure (Mayer and Butler 1993). The six-step calibration process is continued until MA%E is less than or equal to 10% and the resulting parameter set is then used for model testing and the scenario model runs.

Plausibility test of the forest model assumptions

For calibration and testing we compared the carbon content in tree biomass computed by GIS-Century with the coarse-wood used in the Hessian census data. To determine whether these two measures of forest biomass are comparable, we compare the ratio of rlwodc to the total simulated amount of carbon stored in the above-ground biomass with literature values. Röhrig and Bartsch (1992) report a ration of 0.78 for a typical stand of Fagus sylvatica. By comparison a ratio of 0.8 is computed by GIS-Century when using the pre-defined parameter set for broadleaf forests. The close agreement of the ratios provides confidence that the carbon indicators of tree biomass used by GIS-Century and the Hessian forest census are comparable.

Validation of simulated forest growth

To validate the model calculations of forest growth, we apply the calibrated parameter set to simulate annual accumulation of carbon for each district and compare these calculations with forest census data (Fig. 5), considering age structure and species composition (Table 6). Since the data set for testing is small (single estimates from 21 districts) we again use the mean average percent error as a performance measure. The calculated MA%E over all districts is 6.8% which indicates good model performance under these test conditions.

Fig. 5

Test of forest growth simulation results on district level for broadleaf forest. The light bars mark reference data derived from Hessian forest census (HLFW 2003), the dark bars mark the simulated data

Table 6 Biophysical characteristics of major broadleaf tree species in Hesse