Introduction
Tropical forests store 40–50 % of the carbon (C) in terrestrial biomass
(Phillips et al., 1998; Lewis et al., 2009) and account for one-third of
global terrestrial net primary productivity (Saugier et al., 2001; Malhi et
al., 2014b), thereby contributing significantly to the global C cycle and
climate. In addition to their influence on climate, tropical forests also
provide other important ecosystem services such as food, wood products,
erosion control, biodiversity protection and water regulation (Costanza et
al., 1997; Alamgir et al., 2016). Tropical montane forests (TMFs) cover ca.
8 % (elevation > 1000 m a.s.l.) of the total tropical forest area
(Spracklen and Righelato, 2014) and are considered particularly important for
harbouring biodiversity and water regulation (Martínez et al., 2009;
Scatena et al., 2011). Studies indicate that TMF has been underestimated with
respect to its capacity to store (Spracklen and Righelato, 2014) and
sequester (Fehse et al., 2002) C. However, the current understanding of the
role of TMFs in regulating global biogeochemical cycles is hampered by the
paucity of field data on productivity and soil C, but also biomass,
especially from the African continent (Malhi et al., 2013a, b; Spracklen and
Righelato, 2014). Our current understanding of C storage in tropical forests
is to a large extent based on studies of lowland forests in South America and
south-east Asia. Despite of being the world's second largest tropical forest
block, African tropical forests have drawn little attention in terms of C
cycling research compared to their counterparts in South America and
south-east Asia (Lewis et al., 2009; Malhi et al., 2013a, b).
A recent review indicated that the above-ground biomass (AGB) in central
African lowland forest is higher compared to lowland forests in central and
eastern Amazonia (Lewis et al., 2013), which may be related to differences in
climate, soil, biodiversity and/or legacy of disturbance. Ensslin et al. (2015)
suggested that AGB may also be higher in African TMFs compared to those in
South America. However, because of the low number of studies in Africa it is
currently difficult to draw any conclusions about the differences in biomass
of TMFs of the two continents. Even for basic forest attributes such as
biomass, stand structure, species diversity and composition, the number of
African studies of both lowland and montane forest is low (Lewis et al.,
2013; Spracklen and Righelato, 2014; Bastin et al., 2015).
There is a growing number of studies conducted in lowland tropical forests,
focusing on estimates of net primary productivity (NPP), gross primary
productivity (GPP) and C allocation (Aragão et al., 2009; Doughty et al.,
2014; Malhi et al., 2014a). Such studies are still few in TMFs, causing large
uncertainty regarding the C fluxes of these ecosystems (Girardin et al.,
2014a; Spracklen et al., 2016). In general, TMFs are thought to have low
productivity compared to lowland tropical forests (Bruijnzeel and Veneklaas,
1998; Girardin et al., 2014a; Huasco et al., 2014), but this is not always
the case (Fehse et al., 2002). Up to now, no studies of productivity have
been reported for African TMFs.
Most tropical forests are a mix of disturbed and undisturbed stands, denoted
secondary and primary (or old-growth) forests. Secondary forests are defined
as forest regenerating largely through natural processes after significant
human and/or natural disturbance (Chokkalingam and de Jong, 2001). The
detailed distribution of this forest type remains uncertain due to large
variation in how this forest type has been defined in the literature with
respect to the degree of disturbance. However, secondary forests account for
at least 40 to 60 % of the total tropical forest area and are therefore
considered to be very important, both ecologically and economically (Brown
and Lugo, 1990; FAO, 2010), as well as being an important part of the global
C cycle (Birdsey and Pan, 2015; Noormets et al., 2015). Their fraction is
expected to increase in several tropical regions as the human pressure is
likely to continue increasing over the coming decades (Lewis et al., 2015),
e.g. in central Africa (Feintrenie, 2014). In spite of their importance and
frequent occurrence, tropical secondary forests have received comparatively
little attention as the majority of studies reported so far have focused on
old-growth tropical forests (Clark et al., 2001; Malhi et al., 2014b). The C
uptake and storage of secondary forests is therefore still highly uncertain
(Pan et al., 2011). Secondary forests are characterized by a high abundance
of early successional (ES) tree species, which will be successively replaced
by late successional (LS) tree species dominating in undisturbed old-growth
forests. This replacement may take several decades to centuries
(Peña-Claros, 2003; Liebsch et al., 2008; Martin et al., 2013) and since
the ES species grow faster, but may have a lower stature and wood density
(ρ) than LS species (Lawton, 1984; Poorter et al., 2008; Gustafsson et
al., 2016), both the productivity and the forest C stock are likely to
change during successional progression. However, although studies of
secondary forests indicate that they have high above-ground productivity and
C sink strength (Sierra et al., 2012), below-ground compartments have rarely
been investigated (Berenguer et al., 2014).
Quantification of both biomass and productivity rely on established tree
allometric relations. Such relations formulated as equations may be generic
(based on multiple tree species harvested in many different sites), site and
species specific, or something in between (Chave et al., 2005, 2014; Jara
et al., 2015). Chave et al. (2005) showed that the most important parameters
in estimating biomass in pantropical forests were (in decreasing order of
importance) trunk diameter, ρ, tree height (H) and forest type. Thus,
to substantially improve allometric estimates of forest biomass in African
forests, more information on these variables is needed for key tree species in
different types of forest (Gibbs et al., 2007; Kearsley et al., 2013; Lewis
et al., 2013). Height has mostly been omitted in early estimates of tropical
forest biomass (Feldpausch et al., 2011), but when included it reduced the
standard error from 19.5 to 12.5 % in the pantropical biomass estimates
(Chave et al., 2005). Moreover, the use of pantropical allometric equations
where H was not incorporated caused a 52 % biomass overestimation in
TMFs compared to when H was incorporated (Girardin et al., 2010). When H
is incorporated in the biomass estimations, it is normally calculated from
H vs. D relationships established from measurements of subsamples of
trees. Since the D vs. H relationship may vary greatly among forest types
and regions, specific information on this trait is critical to accurately
estimate forest biomass (Feldpausch et al., 2011; Kearsley et al., 2013).
Furthermore, most studies today apply species- or genus-specific ρ data
available in a rather comprehensive database (Chave et al., 2009; Zanne et
al., 2009). However, ρ data are still lacking for many important
tropical tree species and there may also be considerable variation in this
trait within a given species, likely due to variation in environmental
conditions among sites (Muller-Landau, 2004). Species-specific on-site
information for key species may therefore be valuable, and in studies
investigating the influence of successional stages on forest biomass.
With the overall aim of reducing the knowledge gap regarding the C balance of
African TMFs, we quantified C stocks and productivity of 15 half-hectare plots
with mature trees, but with different disturbance histories and different
abundances of early and late successional tree species (ES and LS tree
species). We assessed above- and below-ground C stocks, tree
recruitment and mortality, and NPP of leaves, wood and roots. We hypothesized
that (1) tree biomass and total C stock is higher in LS compared to ES
stands; (2) trees in ES stands are smaller but have higher relative growth
rates compared to trees in LS stands, resulting in similar NPP at both
successional stages; (3) it is critical to account for variation in
allometric relations and ρ when quantifying and comparing forest
biomass at different successional stages, since ES and LS species differ in
these traits; (4) C stocks are higher in the African TMF studied here
compared to TMFs in South America.
Temperature (T) and precipitation at five locations along a 32 km
transect with 15 plots in Nyungwe tropical montane forest during July 2013 to
June 2015. Temperature at 3 m above ground was measured both below the
canopy in each plot and at four meteorology stations in open areas. SD of
plot and meteorology station data represents variation among plots and
months, respectively. Plot numbers at each location are given in brackets.
Locations (plots)
I (1–3)
II (4–6)
III (7–9)
IV (10–12)
V (13–15)
Properties
Mean
SD
Mean
SD
Mean
SD
Mean
SD
Mean
SD
Latitude
2∘31′54′′ S
2∘31′25′′ S
2∘28′54′′ S,
2∘28′30′′ S,
2∘28′38′′ S
Longitude
29∘23′20′′ E
29∘20′33′′ E
29∘14′36′′ E
29∘12′40′′ E
29∘6′53′′ E
Elevation (m a.s.l.)
2493 ± 27
2505 ± 31
2400 ± 18
2415 ± 67
1952 ± 13
Plot air temperatures, under canopy:
Mean annual air temperature (∘C)
13.9 ± 0.1
13.7 ± 0
14.2 ± 0.5
14.4 ± 0.7
15.6 ± 0.2
Lowest monthly mean daily minimum T (∘C)
11.3 ± 0.2
9.9 ± 1
10.1 ± 2
11.7 ± 1.3
10.9 ± 0.3
Highest monthly mean daily maximum T (∘C)
17.7 ± 0.1
18.6 ± 0
19.4 ± 0.6
18.3 ± 0.2
21.2 ± 0.9
Meteorology station data, open areas:
Mean annual air temperature (∘C)
14.1 ± 0.06
14.4 ± 0.05
14.6 ± 0.03
16.1 ± 0.01
Annual precipitation (mm year-1)
1657 ± 163
1755 ± 300
1860 ± 116
3016 ± 63
Material and methods
Study area
The study was conducted in Nyungwe tropical montane rainforest located in
south-western Rwanda (2∘17′–2∘50′ S,
29∘07′–29∘26′ E), ranging from 1600 to 2950 m a.s.l.
Nyungwe forest was gazetted as a national park in 2004 (Gross-Camp et al.,
2012). It covers an area of 1013 km2 and is the largest remaining
middle-elevation montane rainforest in central Africa. It hosts a large
biodiversity, supporting approximately 1105 vascular plant species (of which
230 are trees), 280 bird species and is home to 13 species of primates
(Plumptre et al., 2007). The forest contains various ecosystems ranging from
dense forest and bamboo groves to marshes. Large areas consist of a mixture
of primary and secondary forest (Fashing et al., 2007) due to its disturbance
history (Plumptre et al., 2002; Masozera and Alavalapati, 2004; Masozera et
al., 2006). The secondary forest areas are mainly created from human-induced
disturbances such as tree cutting, fire and mining, but natural disturbances
such as landslide and fallen trees are also significant. The soils were
developed on quartzite schist, mica schist, schist and granite parent
material (Cizungu et al., 2014). The mineral top soil consists of clay, sand
and silt ranging from 2 to 71, 9 to 61 and 5 to 61 % with averages of 34, 43
and 23 %, respectively (Gharahi Ghehi et al., 2014). At a meteorological
station located at Uwinka (2∘28′43′′ S,
29∘12′00′′ E, 2465 m a.s.l. elevation, Nsabimana, 2009), the
average day and night air temperatures were 15.7 and 13.5 ∘C,
the relative humidity was 81 % and annual rainfall was
1867 mm during 2007–2015. The difference between the warmest and coldest
months was 1.1 ∘C. There is a 2-month dry season, normally occurring
from mid-June to mid-August.
Plots
In late 2011 and early 2012, 15 permanent plots with a planimetric area of
0.5 ha (100 × 50 m) were established. The plots were arranged
along a 32 km-long east–west transect at an elevation of ca. 1950 to
2500 m a.s.l. (Fig. S1 in the Supplement). Forest stands ranging from a
dominance of early successional (ES) to a dominance of late successional (LS)
species were included, but areas with recent and extensive disturbance were
excluded. The most abundant ES and LS tree species were Macaranga kilimandscharica Pax and Syzygium guineense (Engl.) Mildbr.,
respectively. Each plot was subdivided into eight subplots with a size of
25 m × 25 m. All individual woody plants with a breast height
diameter (D)≥5 cm were mapped and identified to species level when
possible. The total number of identified tree species was 83. A subset of
species, consisting of those that were among the four most abundant species
with respect to basal area in any of the 15 plots, was selected for more
detailed studies to facilitate the estimation of C stock and productivity.
This subset comprised 22 species in total, representing 90 % of the basal
area and 79 % of all individual stems across all 15 plots. Plot positions
and climate are given in Table 1 and information on topography and stand
characteristics are presented in Table S1 in the Supplement and Table 2. All
forest-area-based information is related to the planimetric area.
Meteorological data
Data on air temperature, air humidity, solar radiation and precipitation
were collected every 30 min from four meteorology stations installed along
the transect of plots (Table 1). One major station was established at the
Uwinka research site in February 2007 (Nsabimana, 2009) and three minor
additional stations were established in June 2013 (Fig. S1). The Uwinka
station was installed in a 15 m tower on a hill top so it would reach above the
canopy, while the others were installed in open areas at 3 m height
(1.5–2 m above-ground vegetation). The minor stations were equipped with
sensors for measurements of temperature, relative humidity, solar radiation
and precipitation (VP-3, PYR/PAR and ECRN-100 from Decagon
Device, Inc, Pullman, WA, USA) connected to a data logger (Em50G, Decagon
Device Inc). At the major and one of the minor stations soil temperature and
moisture were also measured by a thermistor (SKTS 200, Skye instruments Ltd.,
Powys, UK), Theta probes (ML2, Delta-T Devices Ltd., Cambridge, UK) and a
combined sensor (5TM) at 10 cm depth. Air temperature and
humidity was also measured at the centre of each plot (under the canopy)
using mini-loggers (Model TinyTag Plus 2, Gemini data loggers Ltd, United
Kingdom) placed inside self-ventilating radiation shields at approximately
3 m above the ground.
Stem mass and NPP
The D of all trees with D≥5 cm was determined using diameter tape
in two censuses in (1) October–July 2011/12 and (2) October–June 2014/15,
i.e. with 3 years in between. For trees with major
irregularities (e.g. buttresses) at breast height, the point of measurements
(POM) was moved up the stem (max 6.5 m above ground). The D of these
trees was estimated using the taper function by Metcalf et al. (2009):
D=Dhe(-αh-1.30),
where α=0.31, as determined for 31 African tropical species by
Ngomanda et al. (2012) and Dh is the stem diameter at height, h. The
height (H) of 930 trees, representing the full D range of the most
abundant species in the subset defined above, was measured using a clinometer
(Vertex IV, Haglöfs Sweden AB, Långsele, Sweden). To estimate the
tree H of all individuals of these species, H vs. D relationships,
specific for each species, were established by fitting data to the following
function by Lewis et al. (2009):
H=a(1-ebDc),
where a, b and c are fitting parameters. For all other species, we used
generic parameters obtained by fitting Eq. (2) to the data representing all
measured trees. Wood density (ρ) of the most abundant tree species was
estimated by taking wood cores at breast height by using an increment borer
(Haglöf Sweden AB, Långsele, Sweden). The diameter (5.15 mm) and
length (below bark to centre of stem) of the fresh cores and the mass of the
oven dried (70 ∘C) cores were used to calculate the ρ
(g cm-3). When presented as plot mean, BA-weighted ρ was used
(ρBA). The stem mass (including branches) was estimated using
the equation of Chave et al. (2014):
Biomass properties and species composition (means ± SD and
range) of all plots (1–15) classified as early (ES; n=5) or late (LS; n=5) successional. The data are based on results from the first census
(2012) of all stems ≥ 5 cm within the plots, expect for relative
growth rate, recruitment and mortality (Eqs. 6, 10 and 11, respectively) that
were based on both census I and II data. The successional indices (based on
number of stems, #, and basal area, BA) were calculated using Eq. (9). The
listed species are used to construct the successional index and selected
based on their contribution to total stand BA. Diff represents the mean %
differences of LS in relation to ES plots and P values are the results of a
t test of the difference between ES and LS. Bold value, P<0.05.
Plot properties
All plots (1–15)
Plots of different successional stages
Mean
SD
Range
ES (n=5)
LS (n=5)
Diff (%)
P value
Mean
SD
Mean
SD
No of stems, D>5 cm (ha-1)
752 ± 398
350–1844
645 ± 205
868 ± 563
35
0.43
No of stems, D>10 cm (ha-1)
453 ± 218
220–958
421 ± 134
478 ± 277
14
0.69
Mean cross sect. area, D>5 cm (m-2)
0.047 ± 0.025
0.023–0.109
0.038 ± 0.014
0.050 ± 0.034
33
0.47
Mean cross sect. area, D>10 cm (m-2)
0.073 ± 0.036
0.036–0.158
0.054 ± 0.018
0.082 ± 0.046
50
0.25
Basal area (BA, m2 ha-1)
30.0 ± 11.2
17.6–61.0
22.7 ± 3.4
36.2 ± 17.1
59
0.12
Mean height, D>5 cm (m)
14.4 ± 1.5
12.6–18.5
14.1 ± 1.2
14.6 ± 2.4
4
0.64
Mean height, D>10 cm (m)
18.3 ± 1.7
16.5–22.8
17.5 ± 0.5
19.0 ± 2.7
9
0.25
Mean H of 100 highest trees ha-1 (m)
24.6 ± 3.1
21.7–33.0
22.2 ± 0.4
26.9 ± 4.2
21
0.040
Number of big trees, D>40 cm (ha-1)
54 ± 36
26–172
39 ± 14
76 ± 56
94
0.19
No of species
22.6 ± 10.8
11.0–49.0
17.0 ± 3.2
29.2 ± 14.0
72
0.093
Successional index#
0.16 ± 0.15
0.01–0.52
0.03 ± 0.03
0.33 ± 0.12
1159
< 0.001
Successional indexBA
0.34 ± 0.28
0.002–0.89
0.03 ± 0.02
0.65 ± 0.14
2190
< 0.001
Most abundant species (% of BA):
Harungana montana (ES)
2.9 ± 6
1–22
6.7 ± 10.0
0.8 ± 0.9
0.23
Macaranga kilimandscharica (ES)
29.0 ± 27
0–73
59.4 ± 14.8
3.6 ± 5.7
< 0.001
Polyscias fulva (ES)
3.7 ± 6
0–16
7.9 ± 7.2
0.0 ± 0.1
0.040
Carapa grandiflora (LS)
2.5 ± 3
0–10
0.2 ± 0.5
3.2 ± 3.8
0.13
Cleistanthus polystachyus (LS)
2.7 ± 7
3–26
0.0 ± 0.0
7.4 ± 11.4
0.18
Faurea saligna (LS)
3.2 ± 10
11–36
0.0 ± 0.0
9.5 ± 15.6
0.21
Ficalhoa laurifolia (LS)
1.4 ± 4
0–13
0.1 ± 0.2
4.0 ± 5.6
0.16
Ocotea kenyensis (LS)
2.9 ± 3
5–23
4.3 ± 4.7
0.0 ± 0.0
0.07
Ocotea usambarensis (LS)
3.0 ± 7
0–9
2.7 ± 3.7
4.8 ± 3.2
0.36
Syzygium guineense (LS)
25.3 ± 22
0–65
2.2 ± 2.3
38.9 ± 19.0
0.003
Sum of 10 species (% of BA)
76.6 ± 15
42–97
83.6 ± 12.9
72.2 ± 13.4
-14
0.21
MStem=0.0673(ρDBH2H)0.976,
where MStem is the biomass of individual stems in kg, and H is
tree height (m). The coarse root mass (MCR) was estimated based on
MStem using median root to shoot ratio from Cairns et al. (1997) as
follows:
MCR=0.21MStem.
To convert biomass into C mass, we assumed a C concentration of 47.4 % in
line with Martin and Thomas (2011). The net primary production of stems
(NPPStem, Mg ha-1 year-1) was calculated according to
the following equation:
NPPStem=∑MStem2-∑MStem1ΔtA-1,
where ΣMStem1 and ΣMStem2 is the sum of
MStem at censuses 1 and 2 in a plot of area A and, Δt is the
time in between the censuses (2.5 to 3.7 years). Relative growth rate
(RGRStem, % year-1) is calculated as follows:
RGRStem=lnMStem2-lnMStem1Δt×100
for each individual stem; thereafter RGRStem is averaged for a
certain area.
To monitor the stem growth of the two most abundant species M. kilimandscharica and S. guineense in detail, dendrometer bands
(Jädraås skog och mark, Jädraås, Sweden) were installed at
breast height or higher (see D measurements above) on 125 trees of each
species. A randomized block design was applied in which D classes (10 cm
intervals) and plots were used as blocks. The increments were observed
approximately every 4 months with calipers. The averages of these readings
were used to calculate the annual increment of MStem and
RGRStem.
Canopy NPP
Litter from 90 traps distributed over all plots were collected twice per
month from January 2013 to December 2014. In each plot, six of the subplots
were randomly assigned one trap that was randomly placed at one of 16 grid
points within each subplot using a 5×5 m grid. The litter traps
consisted of nylon mesh bags suspended from a circular wire frame of aluminum
(0.3 m2; Jädraås skog och mark, Jädraås, Sweden) and
mounted horizontally on wooden poles ca. 0.8 m above-ground level. The litter
from each trap was collected separately, placed in paper bags and sent to
the lab where it was oven-dried at 70 ∘C to constant mass. After
drying, each sample was sorted into five fractions: leaves, reproductive
organs (fruits, flowers, seeds), twigs, epiphytes and unidentified fine
debris, and weighed. The annual sum of all five fractions was used to
calculate the canopy NPP.
Fine root, litter and soil organic mass
Litter and soil were sampled from the centre of each subplot quadrant (480
sample points), where the litter and organic (O) soil were separately
excavated from a 0.5×0.5 m horizontal ground area. Below the
O-horizon, three consecutive cores (8 cm diameter and 15 cm depth each) of
mineral (M) soil down to a depth of 45 cm were sampled using a root auger
(Ejkelkamp soil & water, Giesbeek, the Netherlands); thereafter the four
M-samples within each subplot and depth were mixed. A subsample of 20 %
based on the fresh mass was taken from each O- and mixed M-sample; thereafter
roots were extracted from each subsample. All samples were then brought to
the lab for drying to constant mass in an oven set to 70 ∘C. The litter
and soil samples were milled in a ball mill (Model: MM 200, Retsch, Germany)
and C concentrations were determined by dry combustion using an elemental
analyser (Model: EA 1108 CHNS-O, Fisons Instruments, Italy).
Fine-root NPP
The fine-root production was measured using in-growth cores with root free
soil surrounded by mesh containers (40 cm deep, 8 cm diameter and ca.
2 mesh, i.e. 12 mm grid). The in-growth cores were installed in the soil by
drilling a vertical hole of 8 cm in diameter to the depth of 40 cm in the
middle of each subplot, using a root auger (see above). The soil from the
drilling was separated into the O- and M-horizons, and after removing the
roots it was used to fill the mesh container installed into the drilling pit,
maintaining the soil horizons. The in-growth cores were installed in
September and December 2013, and fine roots were allowed to grow into the
cores over periods of 3–6 months before being harvested in March 2014,
July–August 2014, December–January 2014/15 and July 2015. To avoid an
underestimation of root mass, because a proportion of the roots inevitably
remain uncollected (Sierra et al., 2003), this study followed the method by
Metcalfe et al. (2007), which controls for systematic underestimation of fine
roots. This was conducted by extracting the roots from the soil in the
in-growth cores during four 8 min intervals (32 min), considering O-and
M-horizons separately, and then fitting the cumulative increase of collected
root mass over time to the following equation to predict root mass as if the
extraction was continued for 120 min:
Mfr,t=alogt+b,
where Mfr,t is the fine-root mass extracted at time t; a and
b are fitting parameters. The roots were brought to the lab and cleaned
from soil by rinsing and sedimentation processes in tap water and thereafter
dried in 70 ∘C until constant mass. The annual sum of production was
calculated in proportion to the time between the harvests (without assuming
any lag-time) and used as fine-root NPP.
Understory
The understory defined as all above-ground parts of plants with
D < 5 cm (woody, herbaceous and grass species) were sampled from one
square metre plot (1×1 m) randomly placed at one of 16 grid points
within each subplot using a 5×5 m grid. All plants within the one
square metre plots were harvested at ground level and thereafter dried in
70 ∘C until constant mass, from which the dry mass per area of
understory (MUstory) was calculated. Based on MUstory and
MStem, an understory index (UI) was developed to classify the
fraction of understory biomass:
UI=MUstory(MUstory+MStem).
Successional index
A successional index (SI), ranging from 0 to a maximum of 1, was developed
to classify the successional stage of the plots from the fractions of ES and
LS trees within the plots:
SIx=LSxTx×1-ESxTx,
where T is the plot total. The subscript x denotes whether it is based on
basal area (BA) or number of tree individuals (#). In this study we based
the index on the 10 most abundant species representing 77 % of the basal
area and 59 % of the individuals of all plots (see Table 2). The
classification of which successional group the species belongs to was mainly
based on Tesfaye et al. (2002), Fischer and Killman (2008), Bloesch et
al. (2009), Kindt et al. (2014) and Rutten et al. (2015a) and we found that
three belong to ES and seven to LS (Table 2). Based on SIBA, two
groups of five plots each were defined, one with the lowest (< 0.1) and
one with the highest SIx (> 0.5), denoted ES and LS plots,
respectively. Similar ranking and grouping resulted if SI# was used
instead (Table S1). However, SI# values were generally lower since
many of the frequently occurring but small trees were not classified for
successional groups. The five plots with a SIx between the ES and LS
groups were classified as intermediate or mixed successional plots (MS).
Recruitment and mortality
The recruitment rate (λ, % year-1) and mortality rate
(μ, % year-1) was determined from the number of stems
(> 5 cm D) in census 1 (n0), in census 2 (nt) and the number
of stems that died (Dt) over the time between the two censuses (t).
The rates were calculated according to Sheil and May (1996), including a
correction factor suggested by Lewis et al. (2004), as follows:
λ=lnn0-ln(n0-Dt)t×100×t0.08μ=lnnt-ln(n0-Dt)t×100×t0.08.
Seasonal variation of monthly mean air temperature (a) and
monthly precipitation (b) at four meteorology stations located
across a 32 km east–west transect in Nyungwe tropical montane forest. Roman
numbers refer to locations presented in Table 1. The data are based on
half-hourly measurements from July 2013 to June 2015.
Statistics
The significance of the relationship between biomass and production
parameters versus the successional indices, understory index, number of big
trees and basal area were determined using the regression analysis tool in
SigmaPlot 12.5 (Systat Software Inc., San Jose, CA, USA). Differences in
forest structure, biomass, C stock and productivity between ES and LS plots
and species were analysed using a two-tailed independent-sample t test.
When differences in these parameters between ES, MS and LS plots were
analysed, one-way ANOVA was conducted, followed by a post-hoc comparison
using Tukey's HSD (honest significant difference) test when the ANOVA
indicated a significant difference (P < 0.05). Data that violated the
assumption of normality, homogeneity or when outliers were present were
log-transformed before the statistical analysis. Both a t test and ANOVA
were conducted using SPPS software (IBM SPSS Statistics for Windows,
Version 22.0. Armonk, NY: IBM Corp.).
Results
Climate
The climate was monitored along the transect of plots between July 2013 and
June 2015 and was characterized by very small seasonal variations in monthly
mean air temperature (ca. 1 ∘C; Fig. 1a), while the seasonal
variation in precipitation exhibited a short dry period in July and less than
average precipitation from May to August (Fig. 1b). The annual mean air temperature under
the canopies at plots 1 to 12 varied between 13.7 ± 0.002 to
14.5 ± 0.07 ∘C and had an annual precipitation of
1657 ± 163 to 1860 ± 116 mm (Table 1). Plots 13–15 located at
ca. 500 m lower elevation compared to the others had a mean air temperature
of 15.5 ± 0.06 ∘C and an annual precipitation of
3016 ± 63 mm during the same period. The lowest monthly mean daily
minimum temperature varied between 9.9 to 11.7 ∘C and the highest
monthly mean daily maximum varied between 17.7 to 21.2 ∘C among
plots. The average temperatures measured at climate stations in nearby open-area
plots were on average 0.3 ∘C higher than under the canopy.
The daily mean soil temperatures were similar to the air temperatures, but
with lower mean diurnal amplitudes (1.9 ∘C in the soil compared to
6.1 ∘C in air). The soil water content at the end of the dry period
varied between 0.05–0.1 m3 m-3 (at 10 cm depth) while it
normally varied between 0.25–0.4 m3 m-3 outside the dry period.
The daily mean photosynthetic photon flux density at the four meteorology
stations was 289 ± 10 µmol m-2 s-1, with slightly
elevated levels during the dry period. The 2 years of climate data presented
for the transect was similar to the 9-year average from our long-term
monitoring station.
Distribution of the mean stem numbers (a) and stem
biomass (b) per unit area in relation to D classes for three
groups of plots (n=5 for all groups) belonging to different successional
stages. The error bars are standard deviation (SD).
Wood density (basal-area-weighted average, ρBA), above-ground
biomass (AGB, stems D≥5 cm), relative growth rates, recruitment rate
and mortality rate of all trees with D>5 cm, given as means ± SD
and range of all plots (1–15) classified as early (ES; n=5) or late (LS;
n=5) successional. The AGB estimates are based on species-specific or
generic H vs. D relationship and ρ. Bold value, P<0.05.
Plot properties
All plots (1–15)
Plots of different successional stages
Mean
SD
Range
ES (n=5)
LS (n=5)
Diff (%)
P value
Mean
SD
Mean
SD
Wood density, specific (ρBA, g cm-1)
0.56 ± 0.06
0.47–0.66
0.48 ± 0.02
0.62 ± 0.02
29
< 0.001
AGB, specific H and ρ (Mg ha-1)
274 ± 165
142–793
156 ± 15
387 ± 244
147
0.022
AGB, generic H, specific ρ (Mg ha-1)
275 ± 151
148–743
164 ± 17
374 ± 222
128
0.023
AGB, specific H, generic ρ (Mg ha-1)
279 ± 136
156–699
189 ± 24
357 ± 209
89
0.080
AGB, generic Hand ρ (Mg ha-1)
275 ± 125
161–650
191 ± 30
342 ± 189
79
0.083
Relative growth rate (%, year-1)
5.2 ± 2.0
2.1–10.0
6.8 ± 1.9
3.8 ± 1.4
-44
0.022
Recruitment rate (%, year-1)
3.8 ± 3.4
0.4–14.1
6.3 ± 4.4
1.4 ± 1.0
-77
0.007
Mortality rate (%, year-1)
1.4 ± 0.5
0.5–2.7
1.1 ± 0.4
1.4 ± 0.4
26
0.26
Successional stage
The successional index used (SI; Eq. 9) to characterize the differences in
successional stages was markedly different between ES and LS plots using both
BA (0.03 and 0.65) and the number of stems (#) per area (0.03
and 0.33; Table 2) as an index basis. The highest SI in any
plot was 0.89 and the reason that the maximum SI (1) was not reached in any
of the plots is partly due to the occurrence of some ES species in all plots,
but mainly because not all tree species were classified into SI groups. The
SI is conservative towards the ES forest type and is sensitive to fact that a
high degree of non-classified species give lower values. The classification
into ES and LS groups was used to test for differences in forest structure, C
stock and productivity between stands of different successional stages
(Table 2) and in the following, the values reported for the two groups of
plots will be separated by an oblique (/) in the order ES/LS.
Forest structure
Many of the common forest structure parameters determined as means for all
trees with D > 5 cm (e.g. stem density, tree cross sectional area at
breast height, BA, H) were generally higher in the LS plots compared to ES
plots but the difference was not significant (P≥0.12; Table 2).
However, the H of the 100 highest trees per ha (22.2/26.9 m; P=0.04), indicating the canopy height, and the BA-weighted ρ (ES: 0.48,
LS: 0.62 g cm-3), P < 0.001) were significantly higher in LS
compared to ES plots, with MS plots in between (Table S1). The total number
of woody species with D > 5 cm was 83 (Table S2), with an average of
23 per plot and no significant difference between ES and LS plots (17/29,
P=0.093). However, the abundance based on BA of several species was
substantially different, but significantly so only in a few cases due to
large variation in species composition between plots also within successional
groups (Table 2).
The distribution of stem number across D classes in all forest types
described an exponential decay function with increasing D and thus
decreased linearly when a logarithmic scale for stem numbers was applied
(Fig. 2a). Notably, ES plots were lacking stems in D classes > 90 cm.
Generally, trees with D of 5 to 10 cm had a small contribution to the
total biomass (< 2.5 %; Fig. 2b). The distribution of biomass across
D classes differed between ES and LS plots, with a major part of stand
biomass in relatively small size trees in ES plots (68 % in trees
< 50 cm D) and large size trees in MS and LS plots (> 50 % in
trees > 50 cm D; Fig. 2b). Thus the tree demography varied between ES
and LS stands, reflecting a difference in disturbance history.
Carbon stocks of different ecosystem compartments (means ± SD
and plot range) for all plots (1–15) classified as early (ES, n=5) and
late (LS, n=5) successional. Diff and P value represent the mean
differences and the results of a t test of the difference between ES and
LS, respectively. For calculating carbon stock (C stock) in the different
compartments we used measured values for each fraction of litter
(37.3–51.3 % C), organic soil (26.0–49.6 % C), mineral soil
(1.3–9.9 % C) and literature values for wood (47.4 % C, Martin and
Thomas, 2011) and for others we assumed 50 % C. AGB is above-ground
biomass; BGB is below-ground biomass; C stock values are in Mg C ha-1.
Bold value, P<0.05.
Compartment
All plots (1–15)
Plots of different successional stages
Mean
SD
Range
ES (n=5)
LS (n=5)
Diff (%)
P value
Mean
SD
Mean
SD
C stockStem
130 ± 78
68–376
74 ± 7.3
183 ± 116
146
0.023
C stockUnderstory*
2.0 ± 1.3
0.2–5.4
1.7 ± 0.6
1.9 ± 1.2
11
0.98
C stockAGB
132 ± 78
69.4–377
76 ± 7.2
185 ± 115
143
0.021
C stockCoarse roots
27 ± 16
14.2–79
16 ± 2
38 ± 24
146
0.023
C stockFine roots
3.3 ± 1.7
2.0–8.1
2.8 ± 0.5
3.8 ± 1.4
37
0.17
C stockBGB
31 ± 17
17.5–83
18 ± 1.3
42 ± 25
130
0.025
C stockLitter
4.3 ± 1.4
1.6–6.3
4.8 ± 1.0
3.5 ± 1.6
-28
0.15
C stockOrganic soil
31 ± 13
9–51
26 ± 7
36 ± 15
35
0.34
C stockMineral soil
157 ± 37
86–196
173 ± 13
139 ± 45
-20
0.13
C stockSoil tot
192 ± 45
97–252
204 ± 13
178 ± 56
-13
0.27
C stockTotal
353 ± 99
232–662
299 ± 21
402 ± 147
35
0.11
AGB fraction of C stockTotal (%)
36 ± 12
23–57
25 ± 0.8
44 ± 14
73
0.019
AGB + BGB fraction of C stockTotal (%)
44 ± 14
28–68
32 ± 1.0
54 ± 17
70
0.019
* Understory data are from Ndayisabye (2014).
Height (H) vs. stem diameter at breast height D relationship for
930 trees representing the 25 most abundant species of all plots and fitted
to Eq. (2). The relationship for all measurements (red line) is compared to
the measurements of 1982 trees in African tropical forests (Lewis et al 2009,
grey dashed line) mainly at an altitude below 1000 m a.s.l. (a) as
well as to species-specific functions for the 10 most abundant species,
including (b) three early successional species (ES) and
(c) seven late successional species (LS). All species-specific data
are presented in Table S3. The mean simulated height (m) at D=10 cm
was 13.4 ± 3.3 in ES and 12.0 ± 1.4 in LS (P=0.35); D=40 cm was 23.3 ± 1.5 in ES and 25.9 ± 1.5 in LS (P=0.033);
D=80 cm was 26.4 ± 1.7 in ES and 33.4 ± 3.9 in LS (P=0.020).
Biomass and C stock
The above-ground C pool of trees was estimated from D measurements of all
stems, H measurements of 930 trees of different D classes from the most
abundant species and species-specific ρ obtained from measurements
(species representing 91.2 % of the BA) or databases (Tables S2 and S3).
The H measurements were used to determine both generic (based on all
measured trees in this study, i.e. site specific) and species-specific
parameterizations of Eq. (2) (Fig. 3). Both the measured data and the generic
parameterization clearly show that the H to D ratio was lower in this
montane forest compared to a generic equation for the central African lowland
(Fig. 3a). The output from the species-specific parameterization show that ES
compared to LS species had a significantly lower H at D of 40
(-10 %, P=0.033) and 80 cm (-21 %, P=0.020) (Fig. 3b and
c). Furthermore, the average ρ of ES compared to LS species was
19 % lower, but the difference was only marginally significant (P=0.057) for the 10 most abundant species. However, the BA-weighted ρ
of LS compared to ES plots was 29 % higher (P<0.001; Table 3). The
use of species specific rather than generic H vs. D relationships and
ρ values did, as expected, not change the estimated average stem
biomass across all plots (274 Mg ha-1) but had a large effect on the
estimated difference in stem biomass between ES and LS plots (Table S1).
Using the species-specific parameters resulted in a higher stem
biomass of 146 % in LS compared to ES plots (156/387 Mg ha-1, P=0.022) while
the generic parameters suggested only a 79 % difference
(191/342 Mg ha-1, P=0.083). The stem biomass of the MS plots was
always in between ES and LS plots and was significantly different from ES, but
not LS (Table S1).
Annual stem volume increment (a), mass
increment (b) and relative growth rate (c) for M. kilimandscharica (n=112) and S. guineense (n=119)
distributed over all available D classes and all plots. Stem production
estimates are based on nine consecutive recordings over 3 years of D with
fixed dendrometer bands. The volume increment, mass increment and relative
growth rate (RGR) of stems averaged over the six lowest D classes (10 to
70 cm) were 53 % (P=0.012), 6 % (P=0.62) and 35 % (P=0.027) larger for M. kilimandscharica compared to
S. guineense. Mean weighted values
based on D classes were used in the comparison.
The total above- and below-ground C pools (down to a depth of 45 cm in the
mineral soil, excluding standing and fallen dead wood) averaged to
353 ± 138 Mg C ha-1 across all plots, with a non-significant
difference between ES and LS pools (299/402 Mg C ha-1, P=0.11;
Table 4). However, the relationships of C stockStem with
SIBA (R2=0.67; Fig. 5a) and SI# (R2=0.42;
Fig. S2) were both highly significant (P<0.001). The woody C pools were
significantly different between ES and LS plots (C stockStem
74/183 Mg C ha-1, P=0.023; C stockCoarseRoots
16/35 Mg C ha-1, P=0.031), as were AGB (including
understory), BGB (including fine roots) and the ratio of AGB to total C
(25/44 %, P=0.019; Table 4) and total biomass (AGB + BGB) to
total stand C (32/54 %, P=0.020). The C stockStem was
negatively and positively related to understory index (Eq. 8) and the number
(#) of large trees (Fig. 6a, b), respectively, although none of these
parameters significantly differed between plots belonging to the two
successional groups.
Biomass (a), relative growth rates (RGR) (b) and
net primary production (NPP) (c) of stems including branches of each
plot in relation to its successional index, based on basal area (BA) of most
abundant early and late successional tree species (Eq. 10). Biomass, RGR and
NPP are based on C units and calculated from Eqs. (3), (6), (5),
respectively, based on measurements of DBH, height, wood density
and a stem tissue C concentration of 47.4 %. The adjusted R2 values
are 0.65 and 0.21 in (a, b).
The total C stock of litter, organic soil and mineral soil were similar in
the plots of the two successional groups (204/178 Mg C ha-1, P=0.27; Table 4). The depth of the soil organic layer was on average 11 cm
(range: 4.5–17.4 cm) with no significant difference between ES and LS plots
(P=0.57). Despite the relatively thick organic layer and high C
concentration (26–50 %), the C StockOrganicSoil was relatively
low (26/36 Mg C ha-1, P=0.34) because of a very low bulk
density (0.08 g cm-3) as the organic soil horizon mainly consisted of
a soft matrix of fine roots and decaying litter.
Net primary production (NPP, Eq. 5) of different forest compartments
(means ± SD and range) for all plots (1–15) classified as early (ES,
n=5) and late (LS, n=5) successional. Diff and P value represent
the mean differences and the results of t test of the difference between ES
and LS, respectively. For C concentrations of different compartments see
Table 4. NPP values are in Mg C ha-1 year-1. OL, organic soil
layer; ML, mineral soil layer.
All plots (1–15)
Plots of different successional stages
Compartment
Mean
SD
Range
ES (n=5)
LS (n=5)
Diff (%)
P value
Mean
SD
Mean
SD
NPPStem
2.8 ± 1.0
1.6–4.5
3.0 ± 1.2
2.7 ± 1.1
-12
0.61
NPPCoarseRoots
0.9 ± 0.4
0.5–1.5
1.0 ± 0.4
0.9 ± 0.5
-4
0.89
NPPWood (AG & BG)
3.7 ± 1.3
2.0–6.0
4.0 ± 1.6
3.6 ± 1.5
-10
0.68
NPPFineRoot (OL)
1.2 ± 0.4
0.8–1.9
1.2 ± 0.2
1.1 ± 0.4
-10
0.51
NPPFineRoot (ML)
0.8 ± 0.3
0.2–1.2
0.8 ± 0.2
0.7 ± 0.4
-19
0.41
NPPFineRoots
2.0 ± 0.6
1.2–2.9
2.0 ± 0.4
1.7 ± 0.7
-14
0.43
NPPLeaves
2.4 ± 0.6
1.5–3.4
2.3 ± 0.4
2.4 ± 0.7
5
0.77
NPPReproductive
0.5 ± 0.4
0.2–1.6
0.3 ± 0.1
0.6 ± 0.5
63
0.38
NPPTwigs
0.5 ± 0.2
0.2–1.0
0.4 ± 0.1
0.6 ± 0.3
39
0.25
NPPEpiphytes
0.2 ± 0.3
0.01–1.0
0.2 ± 0.3
0.3 ± 0.4
73
0.58
NPPOther
0.04 ± 0.04
0.02–0.2
0.1 ± 0.1
0.03 ± 0.01
-52
0.35
NPPCanopy
3.7 ± 0.9
2.2–5.6
3.3 ± 0.4
3.9 ± 1.3
18
0.35
NPPTot
9.4 ± 1.5
6.7–12.1
9.3 ± 1.7
9.2 ± 2.1
-1
0.93
NPPWood/NPPTot (%)
39 ± 10
23–51
41.9 ± 10.9
38.2 ± 9.9
-9
0.59
NPPFineRoots/NPPTot (%)
21 ± 6
11–32
22.3 ± 5.9
19.4 ± 6.7
-13
0.48
NPPCanopy/NPPTot (%)
40 ± 102
29–59
35.8 ± 5.7
42.4 ± 12.7
18
0.32
NPP, RGR, mortality and recruitment
The sum of NPP of different forest compartments (NPPTot) was on
average 9.41 ± 1.50 Mg C ha-1 year-1 across all plots
(Table 5). The variation between plots ranged from 6.7 to
12.1 Mg C ha-1 year-1, but no difference between ES and LS
plots was observed (9.3/9.2 Mg C ha-1 year-1, P=0.93).
The ratios of NPPwood, NPPFineRoots and
NPPCanopy to NPPTotal were on average 0.39, 0.21 and
0.40, and did not significantly differ between ES and LS plots (P>0.32).
NPPStem was related neither to SIBA (R2<0.01;
Fig. 5) nor to SI# (R2=0.04; Fig. S2). However,
RGRStem was negatively related to both SIBA (R2=0.27, P=0.048; Fig. 5) and SI# (R2=0.52, P=0.003;
Fig. S2), and RGRStem was 79 % higher in ES compared to LS
plots (6.8/3.8 %, year-1, P=0.022; Table 3) and
in between in MS plots (Table S1). The lack of difference in NPPstem
between ES and LS stands is probably the net result of counteracting effects
of differences in stem biomass (higher in LS stands) and RGRstem
(higher in ES stands).
To explore how production and RGR varied with tree sizes, the growth of the
most abundant ES and LS species (M. kilimanscharica and
S. guineense, respectively) were analysed in detail using
dendrometer band readings over 3 years (Fig. 4). The stem volume
increment and RGRStem were significantly higher (P=0.012 and
0.027) in M. kilimanscharica than in S. guineense within the D range of 10–70 cm. However, the stem mass
production did not differ between the two species (P=0.62; Fig. 4) since
M. kilimanscharica had lower ρ than S. guineense
(0.44 and 0.63). Species-specific tree growth within given D
classes did not significantly vary among plots or along the plot transect
(data not shown), indicating that the species-specific responses were not
constrained by changes in the plot environment.
Biomass of stems including branches in relation to understory
index (a), number of big trees with D>40 cm (b) and
mean relative growth rates (RGR) of stems in relation basal area (BA).
Understory index is calculated from Eq. (8) and biomass and RGR are expressed
in C units and calculated from Eqs. (3) and (6), respectively, based on
measurements of D, H and ρ and a stem tissue C concentration of
47.4 %. The correlation of stem biomass vs. number of big
trees (b) is also significant when the extreme value is omitted (P=0.004). The adjusted R2 values are 0.17 and 0.39 in (a, c).
The annual recruitment of new stems (6.3/1.4 %, P=0.007) was
significantly higher in ES compared to LS plots, while the annual mortality
(1.1/1.4 %, P=0.26) was similar in both successional groups
(Table 3). However, the most dominant ES species M. kilimandscharica
had significantly higher annual mortality compared to the most dominant LS
species S. guineense (2.1/0.63 %, P=0.035) at plots (n=8) where they co-occurred (≥ 10 stems of each). However, recruitment
rate (3.0/1.4 %, P=0.17) did not significantly differ between these
two species on plots where they co-occurred.
Discussion
Here, we report the first comprehensive estimates of the productivity,
biomass and C stocks of African TMFs. We generally found high C stocks and
productivity, with higher AGB at later stages compared to early successional
forests, but similar productivity across different successional stages. Our
results further demonstrated that accurate quantification of the C stocks
and dynamics of the forests in the present study required local information
on tree allometry, wood density and species composition. This highlights the
need to account for such variation in traits when estimating C stocks of
tropical rainforests at different successional stages and in different
regions, as well as when implementing REDD+ projects.
Biomass and C stocks in relation to successional status
While ES forest stands with a closed canopy and mature trees had significantly
lower AGB and BGB (59 %, P=0.023 and 52 %, P=0.025) than LS stands, there were no significant differences observed
for the soil C stock (Table 4). As a consequence, the plant (AGB + BGB)
fractions of the total C stock was significantly lower in ES compared to LS
plots (32 % in ES, 54 % in LS; P=0.020). This finding is in line
with other studies reporting relatively unaffected soil C stocks in
moderately disturbed and secondary tropical forest (Martin et al., 2013),
although disturbances may have significant effects on the soil C stock in
steep slopes and thus more severely affect TMFs. As shown by the significant
relationship between the stem biomass and the successional index (SIx)
based on BA or stem density (Figs. 5a and S2a), the difference in AGB is
connected to the species composition of both ES and LS forest stands. The
larger AGB in LS compared to ES forest stands was due to LS tree species
having significantly higher ρ and H to D ratio at a given stem
diameter (in the larger diameter classes; Fig. 3; Table 3; see Sect. 4.3), as
well as LS stands having a larger fraction of large trees than ES stands
(Figs. 2, 6b). The basal area actually did not significantly differ between ES
and LS stands. Our results agree with earlier studies showing that
information on the abundance of large trees is an important indicator and
determinant of the whole forest stand AGB (Slik et al., 2013; Bastin et al.,
2015). In summary, our data support hypothesis #1 “Tree biomass and total
C stock is higher in LS compared to ES stands”.
Productivity and C dynamics
Our results are in line with the general observation that ES species grow
faster than LS species (Poorter et al., 2008; Gustafsson et al., 2016) since
we observed a 53 % higher RGRStem (P=0.012) of the dominant ES
species (M. kilimandscharica) compared with the dominant LS
species (S. guineense) over a wide range of D classes.
Furthermore, this was expressed at the plot level as the average
RGRStem in ES plots was twice as high as in LS plots (P=0.017).
In spite of the higher average RGRStem in ES compared to LS
species/plots, we observed no significant differences in NPP of any
compartment (wood, canopy, fine roots; Table 5) between ES and LS plots.
There are likely two reasons for this apparent discrepancy: firstly the
counterbalancing effects of differences in AGB (higher in LS) and
RGRStem (higher in ES) on NPP, and secondly the larger number of
big trees (with lower RGRStem; Fig. 4c) in the LS compared to the
ES plots (Fig. 2a, b). Thus, we find support for hypothesis #2: “trees in
ES stands are smaller but have higher relative growth rates compared to trees
in LS stands, resulting in similar NPP at both successional stages”.
It is generally thought that the productivity of high-elevation rainforests
is low, likely due to low temperature and radiation (i.e. increased
cloudiness; Bruijnzeel and Veneklaas, 1998). Indeed, productivity declines
with increasing elevation have been reported from Borneo and the Andes
(Kitayama and Aiba, 2002; Girardin et al., 2010), as well as in a compilation
of NPP data from different continents and elevations (Malhi et al., 2011).
However, the TMF of the present study has high productivity. Both the above-ground and total NPP of this study were only slightly lower (-7 and
-17 %) than the average recorded NPP for lowland
(< 1000 m a.s.l.) tropical forests (Malhi et al., 2011). When compared
with studies at elevations and mean annual temperatures similar to in our
study, the Nyungwe TMF had among the highest NPP values estimated for TMFs.
Fehse et al. (2002) also reported high above-ground biomass production of
secondary tropical TMFs, although it was not expressed as NPP and therefore difficult
to compare. Our values on canopy and stem NPP were similar to what was found
in a lower montane forest (1050 m a.s.l.) in south-east Asia (Hertel et
al., 2009), while our fine-root NPP was approximately twice as high as in
that study. However, our values of stem and canopy NPP were much higher
compared to a study by Girardin et al. (2010) in the Andes at a similar
elevation to our study, while values of fine-root NPP did not differ.
Stem allometry and wood density
Parameterization of equations to establish H vs. D relationships
developed from data obtained from outside the study area causes large
uncertainties in C stock estimation of tropical forests (Feldpausch et al.,
2011; Kearsley et al., 2013). Indeed, we found significant differences in C
stock estimates when using species-specific parameters compared to generic
parameters (the mean across all species) of ρ and H to D
relationship (Table 3; Fig. 3). In particular, the use of species-specific
parameters greatly affected the comparison of C stocks of LS and ES stands.
The stem C stock was 147 % higher in LS compared to ES forest stands
using local and species-specific parameters, while the difference was reduced
to 79 % if generic parameters were used (Table 3). With species-specific
parameters the difference was significant (P=0.021), while it was not if
generic parameters were used (P=0.083). The main reasons for this
discrepancy were that LS tree species had a considerably
higher ρ and H vs. D
relationship (in larger size classes) compared to ES tree species (Table 3;
Fig. 3). These results demonstrated that species composition, together with
differences in ρ and allometry between LS and ES tree species, are
important drivers for the difference in C stocks between primary and
secondary forests.
Our results further showed that the application of lowland H vs. D
relationships (Fig. 3a; lowland parameterization of Eq. 2; Lewis et al., 2009)
for our TMF caused considerable overestimation of C stocks (Fig. 3a) as it
provided strongly overestimated H for our study area. Thus, our results
demonstrate the importance of using locally derived or validated H vs. D
relationships in TMF C stock estimation, as previously also shown in mixed
secondary forests of Indonesia (Ketterings et al., 2001).
Above-ground biomass (AGB) and forest structure (including trees
with D > 10 cm) of old-growth tropical lowland (< 1000 m a.s.l.)
and montane forest of different tropical regions. The TMF sites were selected
to be within an altitude range of 1600 to 2800 m a.s.l. and an annual mean
temperature range of 11 to 18 ∘C. The mean, min and max values are
based on the means from sites. Abbreviations: SE is south-east, C & E is
central and east, C is central; MAT is mean annual temperature, MAP is mean
annual precipitation, D is breast-height diameter, BA is basal area, ρ is wood density.
Tropical Lowland Forest
Tropical Montane Forest
SE Asiaa
C & E Amazoniab
C Africac
SE Asiad
C & S Americae
C & E Africaf
This studyg
Mean
Min
Max
Mean
Min
Max
Mean
Min
Max
Mean
Min
Max
Mean
Min
Max
Mean
Min
Max
Mean
Elevation (m a.s.l.)
249
9–991
122
41–197
456
35–874
2205
1560–2825
2208
1750–2825
2347
2230–2464
2230
MAT (∘C)
26
22–27
26
25–27
25
22–27
15
12–18
14
11–18
14
12–15
15
MAP (mm)
3127
2052–4441
2421
2009–2856
1853
1530–2837
2468
1891–3985
2976
1487–5000
2281
2240–2322
2322
AGB (Stems, D≥10 cm)
456
196–779
341
251–387
431
147–749
248
119–307
224
78–408
327
275–380
380
BA (m2 ha-1)
37
22–49
29
23–34
32
14–47
41
34–53
36
27–51
42
35–49
35
ρ (g cm-3)
0.60
0.56–0.64
0.68
0.65–0.72
0.64
0.45–0.84
0.58
0.56–0.61
0.54
0.52–0.56
0.62
0.62
Stem density (ha-1)
584
326–1337
597
–
426
181–650
1467
697–2943
1343
477–2753
428
378–478
478
No of sites/plots/total area (ha)
56/79/235
4/17/29
51/193/253
4/19/3
8/52/12
2/10/4
1/5/2.5
a Slik et al. (2010) – (Borneo –
Brunei; Malaysia; Indonesia). b Baker et al. (2004); Quesada et al.
(2010); www.ctfs.si.edu/group/Ecosystems+and+Climate/Data+Resources –
(Brazil). c Lewis et al. (2013) – (Cameroon; Central African
Republic; Democratic Republic of Congo; Gabon; Nigeria; Republic of the
Congo). d Aiba et al. (2005); Culmsee et al. (2010); Dossa et
al. (2013); Edwards and Grubb (1977); Kitayama and Aiba (2002); Sawada et
al. (2016) – (Malaysia and Indonesia; Papua New Guinea). ρ and stem
density only from 3 sites. e Álvarez-Arteaga et al. (2013);
Delaney et al. (1997, 1998); Girardin et al. (2010, 2014b); Grimm and
Fassbender (1981); Lieberman et al. (1996); Moser et al. (2011); Leuschner et
al. (2007); Spracklen and Righelato (2005); Unger et al. (2012) – (Costa
Rica; Ecuador; Mexico; Peru; Venezuela). BA, ρ and stem density only
from 3 sites. f This study; Ensslin et al. (2015); Rutten et
al. (2015b); Hemp (2006) – (Rwanda; Tanzania). ρ only from this study.
g Only LS plots, stems with D≥10 cm.
In the high tree diameter range, we found that LS trees were significantly
taller than ES trees at a given tree diameter, while it was the opposite in
the low tree diameter range (Fig. 3b, c). ES forest stands are characterized
by pioneers, with low stature and light wood (Swaine and Whitmore, 1988;
Muller-Landau, 2004). For example, the dominant ES species M. kilimandscharica (59.4 ± 14.8 % of BA in ES plots; Table 2) rarely
grows taller than 25 m (Fig. 3c). Such differences between ES and LS species
have been observed earlier (e.g. King, 1996) and may reflect a trade-off
between fast height extension and mechanical stability during forest
succession (Lawton, 1984). ES species have faster height growth in the
beginning of successional progression but will eventually be surpassed by LS
trees with higher ρ and stability.
High relative growth rates and low ρ of ES tree species agree with the
plant economics spectrum of fast-growing species (Woodall et al., 2015;
Reich, 2014). The trade-off between ρ and growth rates has also been
observed in earlier tropical studies (Poorter et al., 2008, 2010;
Keeling et al., 2008). Chave et al. (2009) further argued that mortality
rates in tropical forests are associated with low-density wood, but this was
not confirmed by our results, in which we did not find any significant difference
in mortality rates between ES and LS plots (Table 3). However, M. kilimandscharica, the most abundant ES species, had a significantly higher
mortality rate than S. guineense, the most abundant LS species.
Overall the observations in this study support hypothesis #3: “it is
critical to account for variation in allometric relations and ρ when
quantifying and comparing forest biomass at different successional stages,
since ES and LS species differ in these traits”. It is therefore likely
that a full recovery of the C stock not will be achieved before the first
generation of ES trees have been replaced by LS trees.
Biomass and C stocks in relation to studies of old-growth TMF
To put our results into perspective, we compiled data from studies of AGB
(trees with D > 10 cm) of old-growth tropical lowland forest
(< 1000 m a.s.l.) and TMF (elevations of 1600 to 2800 m a.s.l. and
mean annual temperature of 11 to 18 ∘C, ca. ±300 m and
±2.5 ∘C of our lowest and highest elevation and temperature;
Table 6). The late successional (LS) stands of the Nyungwe TMF were shown to
have somewhat higher AGB than the average old-growth lowland tropical forests
of central/eastern Amazonia (+11 %) and a bit lower than lowland
forests of central Africa (-11 %) and Borneo (-15 %; Table 6).
Average stem density, basal area and ρ in our LS stands were similar to
those observed in lowland tropical forests. These findings support the
suggestion by Spracklen and Righelato (2014) that TMF biomass may store more
C than expected.
Compared to the average AGB of the TMFs in south-east Asia and Central and
South America, our LS plots had substantially higher AGB (Table 6). Only two
plots from the studies in Central and South America (Delaney et al., 1997;
Álvarez-Arteaga et al., 2013) had similar AGB to our average LS AGB (395
and 408 Mg ha-1). However, the AGB in these studies may be
overestimated as none of them used H as a parameter in the AGB calculation,
an omission which may overestimate the biomass in TMFs (Girardin et al.,
2010). When we applied the allometric equations from Delaney et al. (1997) to
our data we obtained a 20 % higher AGB than when using the equation from
Chave et al. (2014), including H as a parameter (see further discussion in
Sect. 4.3). The only study of undisturbed TMF at the selected altitudes and
temperatures from the African continent that we found (Ensslin et al., 2015;
Rutten et al., 2015b) reported lower AGB (275 Mg ha-1) than in our LS
plots but higher than average AGB in the American and Asian TMF studies
(Table 6). Although AGB data from African TMFs are still scarce, the general
pattern of higher AGB in Africa compared to America observed in tropical
lowland rainforests (Banin et al., 2012; Feldpausch et al., 2012; Lewis et
al., 2013) may thus hold also for TMFs, as also suggested by Ensslin et
al. (2015).
TMF soils are considered to be richer in soil C content than lowland tropical
forest (Roman et al., 2011), which has on average 46 % of the total
ecosystem C stock in the soil down to 1 m (Malhi et al., 2015). Taking into
account that an average tropical forest soil has around 25 % of the C
content between our sampling depth (average 0.56 m) and 1 m (Jobbágy
and Jackson, 2000), the soil C fraction in the LS plots was 51 %, which
was similar to an study in an old-growth TMF in Venezuela observing 52 %
(Delaney et al., 1997), but much higher than in a natural submotane
rainforest in Sulawesi having only 32 % of the C in the soil (Kessler et
al., 2012). Only a few TMF studies have reported the C stock of both below
and above-ground compartments, which makes it difficult to assess the
portioning of C in these ecosystems. In addition, comparing the below-ground C stock of
different studies is probably less reliable than comparing above-ground C
stocks. This is because of inconsistent sample methodology (e.g. regarding
sampling depth and intensity; Roman et al., 2011) combined with the
likelihood of large local variation in below-ground C stocks, especially in
mountain areas where the vertical soil profile may be very variable due to
variation, e.g. in sloping degree.
Based on available data, our total below-ground C stock is similar to the
levels reported for studies in South American TMFs, when adjusted for
differences in sampling depth (Delaney et al., 1997; Grimm and Fassbender,
1981; Girardin et al., 2010; Moser et al., 2011). However, the partitioning
of below-ground C in the present study differs greatly from studies in the
Andes by Girardin et al. (2010) and Moser et al. (2011) which were the only
studies providing detailed below-ground compartmental C stock information. In
our study, the C stock of the mineral soil down to 30 cm depth was more than
three times as large as in the Andean TMFs, while the C stock of the soil
organic layer plus fine roots was only half. The main reasons for this
difference in soil C partitioning were that the mineral soil in the present
study had very high C concentrations (4–7 % in top 30 cm) while the
organic soil layer was more shallow (11 cm) compared to the Andean studies
(31 cm).
Overall, the observation of high above- and below-ground C stocks in our LS
plots, as well as in Ensslin et al. (2015) support hypothesis #4: “C
stocks are higher in the African TMF studied here compared to TMFs in South
America”. However, ecosystem C stock data from African TMF studies are still
scarce and there is a risk for biased results among TMF studies as the
average plot size in TMF studies is small (0.22 ha in our compilation in
Table 6).