Appendix A. Model description and color photographs depicting fire in miombo woodlands.
FIG. A1. Gap model schematic. Parameters are shown in red (see Tables A2 and A3 for values, source and sensitivity of the parameters). Green shows allometric calculations based on Table A1. Model driving data is shown in blue (see Fig. A2). Stochastic processes are shown with a dashed line. The model produces annual output, but the light and C assimilation functions run on hourly time steps (see text). 
The model is a 1D patch model, explicitly representing each stem and its light environment, carbon assimilation, growth and mortality. Each model patch has twentyfive 1 m vertical layers and is not spatially explicit. The model is written in FORTRAN 90.
Light and seasonal and diurnal model drivers
Within each patch, the model simulates light interception by the canopy and the resulting photosynthetic assimilation. The model is driven with a 3.5 year radiation climatology from a weather station (Skye Instruments, UK) at Chitengo, 25 km from the study site. We generated a mean diurnal cycle of downwelling PAR, at hourly resolution, for each calendar month. The model is also driven with an observed monthly fraction of peak leaf area index (LAI), determined from 3.5 years of monthly hemispherical photos at two nearby permanent sample plots.
FIG. A2. Data used to drive the model. A seasonal cycle of leaf area index was derived from replicated monthly hemispherical photos on two 1 ha permanent sample plots. PAR measurements are from a weather station 25 km away. 3.5 years of data were averaged by hour and month to generate a representative diurnal cycle for each month. 
Plant Area Index (PAI ) was estimated from hemispherical photos (Nikon Coolpix 4500 with a FCE8 fisheye converter) collected each month at two 1 ha permanent sample plots (PSP). Nine photos were acquired on a 20 m grid on each plot, each month. The photos were thresholded to separate plant and sky pixels using the algorithm of Ridler and Calvard (1978), implemented in MATLAB to determine gap fraction, which was converted to an PAI estimate using the method of Licor (1989), with the images segmented in a similar manner to van Gardingen et al. (1999). As the trees are fully deciduous, PAI was converted to LAI by subtracting the lowest observed PAI from each observation point. The monthly LAI fraction represents the deciduousness of the vegetation and is an explicit representation of limits of the growing season of trees in this environment (Fig. A2). Each year the vertical leaf area distribution is calculated based on functions which relate each stem’s DBH to its total height, canopy depth and leaf area (Shinozaki et al. 1964) using the allometric relationships shown in Table A1. Leaves were assumed to be spread evenly throughout the canopy depth. A simple radiative transfer scheme passes light down through the canopy layers. The amount of PAR absorbed is calculated according to leaf area density using the BeerLambert law (Jones 1992), with an assumed spherical leaf angle distribution (k). All radiation was assumed diffuse and the leaves had no albedo or transmittance.
Carbon assimilation and allocation
The absorbed PAR was converted to assimilated C using photosynthetic light response curves parameterised from two studies in Zimbabwean miombo (Tuohy and Choinski 1990; Tuohy et al. 1991). Two parameters describe the light response curve: the maximum rate of assimilation, P_{max} and the amount of light needed to achieve half this rate, k_{p}. For each stem, mass of photosynthate was summed for each canopy layer, for each hour of the 12 diurnal cycles representative of each month, and scaled up to a yearly total.
Carbon allocation routines are resolved each year for each stem. We assumed that 50% of assimilate is used for respiration (Waring et al. 1998, R_{a} in Table A2). Of the remainder, the required amount is allocated to leaves to produce the observed phytomass of peak LAI. Leaf carbon specific area (LCA gC m^{2}) is derived from field measurements (Nottingham 2004). We assumed that allocation to fine roots, A_{fr}, is equal to allocation to leaf mass. Any remaining C is allocated to woody growth, which is partitioned above and belowground according to a local DBHspecific root:shoot allometric. The increase in aboveground stem C is converted to a change in DBH using allometric equations (Table A1), which in turn allows the increase in height and leaf area to be calculated for the next year.
Mortality and the effect of fire
Once growth is completed, each stem is exposed to a random chance of mortality, which is either an intrinsic (nonfire) mortality or a fireinduced mortality derived from our fire experiments. These fireinduced mortality rates are DBH and fire intensity classspecific and are based on piecewise functions of the field results (Table A3). The intrinsic mortality rate, M_{i}, was unknown and is very poorly constrained in this ecosystem. We used a nominal value of 2% regardless of size (similar to Desanker and Prentice 1994; Desanker 1996), and undertook sensitivity analysis (Table A2). In the model, the chance of a fire occurring was determined from a user supplied control variable, the fire return interval (FRI), and a random number generator. The chance of a fire occurring was determined as 1/FRI.
Resprouting
Because of the importance of resprouting in fireprone ecosystems (Bond and Midgley 2001; Chidumayo 2004; Mlambo and Mapaure 2006), aboveground stem mortality (topkill) was decoupled from belowground rootstock mortality. Stems > 2 cm DBH (parameter S_{resprout}) which are killed, have a probability or resprouting (1  S_{mort}), which is based on data from the fire experiments, and we assume they achieve a 2 cm DBH in their first year. In addition, every year there is a nominal 3% chance (parameter P_{recruit}) of a recruitment event occurring and 200 new seedlings (parameter S_{new}) being established in the patch. We justify this low level of recruitment from studies which show that most regeneration is from sprouts and that seedling survival is low in these woodlands (Chidumayo 1992, 1997).
Model output and initiation
After calculating mortality, the patch LAI (for stems > 1.5 m tall), basal area, aboveground C stock, and stocking density (for stems > 5 cm DBH), and stem size distribution are calculated for all live stems.
The model was initiated from bare ground with 200 seedlings and tended to reach equilibrium after ~200 years. All model runs in this study used 50 patches of 0.02 ha (based on an 8 m crown radius for a typical large stem) and were 1000 years long. Output for the final 500 years is reported as the 50patch 500year mean with the standard deviation between the 50 patches.
Model sensitivity analysis
Method
We ran a sensitivity test of all model parameters. Each parameter was varied in turn by a factor of 0.5, 0.75, 1.5, and 2 from its nominal value (Table A2). The sensitivity, S, is defined as S_{x} = ([R_{adj}  R_{n}]/ R_{n}) / ([P_{adj}  P_{n}]/ P_{n}), where x is the factor by which the nominal value is changed, R_{adj} is the response for the model run with the adjusted value, Rn is the response with the nominal value, and P_{n} and P_{adj} are the parameter values for the nominal and adjusted cases respectively. Patch basal area was used as the response variable. The parameters used in the model and their source, magnitude and sensitivity, are shown in Table A2.
Results
The model output were much more sensitive to the growth parameters compared to those controlling regeneration (Table A2). In order of sensitivity, they were: the maximum rate of photo synthesis (P_{max}); the fraction of assimilate allocated to respiration (R_{a}); the halfsaturation rate of photosynthesis (k_{p}); carbon mass per unit leaf area (LCA); allocation to fine roots (A_{fr}); the leaf angle distribution (k); and the leaf area:basal area ratio (LA:BA). These are basic physiological parameters, which are well constrained by local field observations (P_{max}, and k_{p}, by Tuohy et al. 1991; LCA and LA:BA by our data) or to a lesser degree, globally (R_{a}, Waring et al. 1998 and k, Norman and Campbell 1989). A_{fr} is badly constrained, particularly in savanna environments. Of the mortality and regeneration parameters, the intrinsic mortality rate (M_{i}) was the only one that significantly affected the final patch basal area.
TABLE A1. Allometric equations used in the model and derived from local field data. Equation forms are defined as follows: power, Y = aX^{b}; linear, Y = aX ± b. n is the number of samples of field data used to fit the function.
Dependent (Y) variable 
Independent (X) variable 
Form 
a 
b 
n 
r^{2} 
min X 
max X 
Notes 
C stock of stem (kg C) 
DBH, m 
power 
4222 
2.6 
29 
0.93 
0.05 
0.72 

C stem : C root+stem 
DBH, m 
linear 
0.32 
0.6 
23 
0.26 
0.05 
0.72 

Height of top of tree (m) 
DBH, m 
linear, with saturation 
42.6 
80 
0.67 
0.07 
0.86 
where DBH > 60 cm, ht = 25 m 

Height of base of canopy (m) 
DBH, m 
linear, with saturation 
22.3 
80 
0.58 
0.07 
0.86 
where DBH > 67 cm, canopy base = 15 m 

Leaf area, m^{2} 
Basal area, m^{2} 
linear 
1330 
10 
0.29 
4.4 
13.9 
plot level data 
TABLE A2. Parameters in the model, their source, nominal value and sensitivity, S_{x}, see text for symbol definitions. For the sensitivity analysis, we use the 50patch mean basal area as the response variable. Each variable was changed in turn.
Parameter description 
Parameter name 
Nominal parameter value, P_{n} 
S_{0.5} 
S_{0.75} 
S_{1.5} 
S_{2} 
Source of nominal parameter value 
Growth parameters 

Fraction of GPP used for autotrophic respiration 
R_{a} 
0.5 
1.8 
1.9 
2.4 
1.0 
Waring et al. (1998) 
Extinction coefficient for BeerLambert law 
k 
0.5 
0.0 
0.6 
0.2 
0.1 
Norman and Campbell (1989) 
Amount of C allocated to fine roots, as a fraction of allocation to leaves 
A_{fr} 
1 
0.4 
0.2 
1.0 
0.6 
Consistent with Castellanos et al. (2001). Similar to Table 5 of Hendricks et al. (2006), which is for temperate systems. 
Maximum rate of photosynthesis (µmol C·m^{2}·s^{1}) 
P_{max} 
10 
2.0 
2.0 
2.5 
2.3 
Tuohy and Choinski (1990); Tuohy et al. (1991) 
PAR intensity at which 0.5 P_{max} is obtained (µmol·s^{1}·m^{2}) 
k_{p} 
250 
1.7 
1.7 
1.1 
0.7 
Tuohy and Choinski (1990); Tuohy et al. (1991) 
Leaf carbon per leaf area (g C/m^{2}) 
LCA 
50 
1.5 
1.1 
1.7 
0.9 
Nottingham (2004); Chidumayo (1997) 
Leaf area per unit basal area (m^{2}/m^{2}) 
LA:BA 
1330 
0.0 
0.6 
0.1 
0.1 
See Table A1 
Mortality and Regeneration parameters 

Intrinsic mortality rate 
M_{i} 
0.02 
0.8 
1.3 
1.3 
0.5 
Estimated, similar to Desanker and Prentice (1994) 
Diameter, m, at which a seedling is considered to have developed a rootstock and the ability to resprout 
S_{resprout} 
0.02 
0.0 
0.2 
0.1 
0.0 
estimated 
Number of seedlings per ha established in a recruitment year 
S_{new} 
100 
0.0 
0.5 
0.2 
0.1 
estimated 
Probability of a recruitment year occurring 
P_{recruit} 
0.03 
0.1 
0.1 
0.2 
0.0 
estimated 
Probability of a root stock failing to resprout after fire 
S_{mort} 
0.04 
0.2 
0.3 
0.1 
0.0 
this study 
TABLE A3. Piecewise parameterisation of fireinduced mortality rates shown in Fig. A3. These are used to calculate stem mortality in the model in years when a fire occurs.
Fire intensity (TA tercile) 

High 
Med 
Low 

a_{x} 
70 
70 
60 
b_{x} 
5 
4 
2 
b_{x_sat} 
2 
3 
3.8 
Where: 

DBH < 10 cm 
Log odds of topkill = a_{x} DBH+ b_{x} 
DBH > 10 cm 
Log odds of topkill = b _{x_sat} 
TABLE A4. Metrological conditions during each burn. The values are the mean for the period of the fire.
Plot No. 
1 
2 
3 
4 
5 
6 
7 
8 
Temp (°C) 
no burn 
31.4  29.9  35.4  25.0  33.3  32.8  33.3 
Relative Humidity (%) 
33  32  23  55  15  28  21  
Wind speed (m/s) ^{1} 
2.4  1.4  0.9  1.1  1.7  1.6  0.9 
Where:  
DBH < 10 cm 
Log odds of topkill = a_{x} DBH+ b_{x} 
DBH > 10 cm 
Log odds of topkill = b_{x_sat} 
TABLE A5. Metrological conditions during each burn. The values are the mean for the period of the fire.
Plot No. 
1 
2 
3 
4 
5 
6 
7 
8 
Temp (°C) 
no burn 
31.4  29.9  35.4  25.0  33.3  32.8  33.3 
Relative Humidity (%) 
33  32  23  55  15  28  21  
Wind speed (m/s) 
2.4  1.4  0.9  1.1  1.7  1.6  0.9 
TABLE A6. Tree and sapling species composition on the experimental plots. Trees are defined as live standing woody species > 5 cm DBH. Saplings are defined as woody species < 5 cm DBH and > 0.3 cm D_{10}. Species names area abbreviated as follows: Brachystegia boehmii (Bb), Brachystegia spiciformis (Bs), Burkea africana (Ba), Combretum adenogonium (Ca), Crossopteryx febrifuga (Cf), Dalbergia boehmii subsp. boehmii (Dbb), Diplorhynchus condylocarpon (Dc), Erythrophleum africanum (Ea), Julbernardia globiflora (Jg), Lannea schimperi (Ls), Millettia stuhlmannii (Ms), Ormocarpum trichocarpum (Ot), Pterocarpus angolensis (Pa), Pterocarpus rotundifolius subsp. Polyanthus (Prp), Sclerocarya birrea subsp. caffra (Sbc), Xeroderris stuhlmannii (Xs). Blanks indicate unidentified saplings.
Large tree data 

Top 5 species by basal area. 1. 
Bb 
Bb 
Bb 
Bb 
Bb 
Bb 
Bb 
Jg 
2. 
Dc 
Dc 
Dc 
Dc 
Dc 
Dc 
Ba 
Dc 
3. 
Jg 
Ba 
Ba 
Jg 
Jg 
Sbc 
Bs 
Bb 
4. 
Ba 
Bs 
Sbc 
Prp 
Prp 
Prp 
Dc 
Bs 
5. 
Ea 
Jg 
Ea 
Pa 
Xs 
Ca 
Ea 
Pa 
Top 5 spp % of total basal area 
76% 
90% 
81% 
82% 
92% 
85% 
84% 
77% 
Sapling data 

Top 5 species by basal area. 1. 
Bb 
Bb 
Dc 
Dc 
Bb 
Dc 
Dc 
Dc 
2. 
Dc 
Jg 
Ot 
Ca 
Dc 
Jg 
Jg 
Ba 
3. 
Jg 
Dc 
Ls 
Bb 
Ba 
Dbb 
Bb 
Xs 
4. 
Ca 
Ls 
Bb 
Jg 
Cf 
Ba 
Ms 

5. 
Ba 
Ba 
Ms 
Bb 
Pa 
Ca 

Top 5 spp % of total basal area 
86% 
78% 
95% 
86% 
92% 
70% 
87% 
93% 
FIG. A3. Model output of patch characteristics for a lowfire scenario (FRI = 100, lowintensity fires) from initiation with bare ground at year zero to 1000 years. The mean of the 50 patches is shown with a black line. One standard deviation of the 50 patches is shown above and below the mean with a gray shade. The white triangles on the right Y axis indicate the field data for the Marrameu forest plot (Site 2), the gray triangles indicate the 50patch mean of the final 500 model years. 
FIG. A4. A damaged tree bursts into flames. 
FIG. A5. Another collapses after the bole has combusted. 
LITERATURE CITED
Bond, W. J., and J. J. Midgley. 2001. Ecology of sprouting in woody plants: the persistence niche. Trends in Ecology & Evolution 16(1):45–51.
Castellanos, J., V. J. Jaramillo, R. L. Sanford, and J. B. Kauffman. 2001. Slashandburn effects on fine root biomass and productivity in a tropical dry forest ecosystem in México. Forest Ecology and Management 148(1–3):41–50.
Chidumayo, E. N. 1992. Seedling ecology of two miombo woodland trees. Plant Ecology 103(1):51–58.
Chidumayo, E. N. 1997. Miombo ecology and management : an introduction. London, IT Publications in association with the Stockholm Environment Institute.
Chidumayo, E. N. 2004. Development of BrachystegiaJulbernardia woodland after clearfelling in central Zambia: Evidence for high resilience. Applied Vegetation Science 7(2):237–242.
Desanker, P. V. 1996. Development of a MIOMBO woodland dynamics model in Zambezian A_{fr}ica using Malawi as a case study. Climatic Change 34(2): 279–288.
Desanker, P. V., and I. C. Prentice. 1994. Miombo  a Vegetation Dynamics Model for the Miombo Woodlands of Zambezian A_{fr}ica. Forest Ecology and Management 69(1–3):87–95.
Hendricks, J. J., R. L. Hendrick, C. A. Wilson, R. J. Mitchell, S. D. Pecot, and D. Guo. 2006. Assessing the patterns and controls of fine root dynamics: an empirical test and methodological review. Journal of Ecology 94:40–57.
Jones, H. G. 1992. Plants and microclimate : a quantitative approach to environmental plant physiology, Cambridge University Press, Cambridge, UK.
LiCor. 1989. LAI2000 Plant Canopy Analyzer Operating Manual. Lincoln, NE, LiCor.
Mlambo, D., and I. Mapaure. 2006. Postfire resprouting of Colophospermum mopane saplings in a southern A_{fr}ican savanna. Journal of Tropical Ecology 22:231–234.
Norman, J., and G. Campbell. 1989. Canopy structure. Pages 301–325 in R. W. Pearcy, J. R. Ehleringer, H. A. Mooney, and P. W. Rundel, editors. Plant Physiological Ecology: Field Methods and Instrumentation. Chapman and Hall, London, UK.
Nottingham, A. 2004. The characterisation of foliar Nconcentrations and presence of Nfixers in Miombo woodland, Mozambique. MRes Environmental Biology, University of Dundee, UK.
Ridler, T. W., and S. Calvard. 1978. Picture Thresholding Using an Iterative Selection Method. IEEE Transactions on Systems Man and Cybernetics 8(8):630–632.
Shinozaki, K., K. Yoda, K. Hozumi, and T. Kira. 1964. A Quantitative Analysis of Plant FormThe Pipe Model Theory: ii. Further Evidence of the Theory and its Application in Forest Ecology. Japanese Journal of Ecology 14(4):133–139.
Tuohy, J. M., and J. S. Choinski. 1990. Comparative Photosynthesis in Developing Leaves of BrachystegiaSpiciformis Benth. Journal of Experimental Botany 41(229):919–923.
Tuohy, J. M., J. A. B. Prior, and G. R. Stewart. 1991. Photosynthesis in Relation to Leaf Nitrogen and PhosphorusContent in Zimbabwean Trees. Oecologia 88(3):378–382.
van Gardingen, P. R., G. E. Jackson, S. HernandezDaumas, G. Russell, and L. Sharp. 1999. Leaf area index estimates obtained for clumped canopies using hemispherical photography. Agricultural and Forest Meteorology 94(3–4):243–257.
Waring, R. H., J. J. Landsberg, and M. Williams. 1998. Net primary production of forests: a constant fraction of gross primary production? Tree Physiology 18(2):129–134.