RMSE ¼
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PN
i¼1ðPi OiÞ
2
N
s
(2)
3. Results
3.1. Carbon stocks and changes
The mean C stock of tree biomass across the study region in 2011
was 6.6 kg m2
, ranging from 0e14.1 kg m2
, according to the
model simulations (Table 1). The C stock of biomass was slightly
lower in the south coast compared to the inland (Fig. 3a). It was
spatially autocorrelated up to about 400e500 m among the case
study areas according to the visual interpretation of the maps and
the semivariograms (Fig. 3a and Appendix 4). The simulated estimate
of the mean C stock of soil was 7.9 kg m2
, ranging from
3.4e14.9 kg m2 (Table 1). The soil C stock showed a similar
regional trend to the C stock of biomass (Fig. 3b). It was spatially
autocorrelated up to about 300e600 m (Appendix 5).
The C stocks of biomass and soil increased in the study region in
2011 according to the simulated estimates indicating that the forests
were a sink of C (Table 1). The mean change in the C stock of
biomass was 0.032 kg m2 a1
, ranging from 11.9e1.02 kg m2 a1
(Table 1). The change in the C stock of biomass showed no trend
across a climatic gradient from south to north (Fig. 3c). It was
spatially autocorrelated up to about 60e100 m (Fig. 3c and
Appendix 6). The mean change in the C stock of soil was
0.022 kg m2 a1
, ranging from 0.388e5.40 kg m2 a1 (Table 1).
The change in the C stock of soil showed no regional trend either. It
was spatially autocorrelated up to 100e300 m (Fig. 3d and
Appendix 7).
3.2. Mapping framework performance
The simulated time-series of the C stock of biomass followed the
measurement-based means over the stand age (Fig. 4). The simulated
estimates were, however, generally higher than the
measurement-based ones. The simulated and the measurementbased
means of the biomass C stock were clearly correlated
(R2 ¼ 0.75) per municipality across the study region (Fig. 5a). The
correlation was even higher at the level of individual grid cells
(R2 ¼ 0.77, data not shown). The simulated means showed a tendency
for overestimation compared to the measurement-based
means being on average 1.4 kg m2 higher at grid cell level and
1.6 kg m2 higher at the municipality level (Fig. 5a).
The simulated means of the C stock of soil were highly correlated
with the measured means per forest site type (R2 ¼ 0.93)
across the study region after removing one obvious outlier (Fig. 5b).
The measured mean of the soil C stock equal to 3.58 kg m2 for the
fertile, Oxalis-Maianthemum type (OMaT) forest was regarded as a
measurement error. The simulated means showed a slight tendency
for overestimation compared to the measured means being
on average 1.1 kg m2 higher (Fig. 5b). The simulated estimates of
harvests in the municipalities of the study region in 2011 were
highly correlated (R2 ¼ 0.88) with the reported harvests (Fig. 5c).
The RMSEweighted of the simulated vs. measured harvests was
44 300 m3
.
4. Discussion
A framework to map the current status and spatial variation of
the C budget of boreal forested landscapes was developed in this
study. The C stocks of biomass and soil and the annual change in
these stocks were quantified at the level of individual grid cells and
the results were scaled up to the regional level. The C stocks
increased from south coast to inland whereas the changes in these
stocks were more uniform. The spatial patches of C stocks were
larger than those of C stock changes. The simulated estimates were
very similar to measurement-based estimates indicating a good
mapping framework performance.
4.1. Spatial variation of C budget
The spatial variation in the C budget maps was a result of connecting
the simulated estimates of C stocks and changes to detailed,
spatially explicit information about forest characteristics. The
spatial variation reflected the variation in site type, main tree
species and age of the forest, and the management actions related
to age. The visual overview of the maps showed that the C stocks of
biomass and soil were slightly lower in the south coast compared to
the inland. A similar regional trend has been observed in previous
research (Liski and Westman, 1997). It is likely explained by the soil
properties reflected in the site type. The south coast is mostly
covered by dry heath (Calluna site type (CT) according to Cajander’s
(1949) classification) which has lower forest growth and litter
production rates compared to more fertile sites in the inland. The
changes in the C stocks had no regional trends.
The spatial interpretation of the simulated estimates revealed
that the C stocks of biomass and soil were spatially autocorrelated
up to 400e600 m. This is probably explained by the fine-scale
variation in the site type and tree species composition controlling
the biomass and litter production estimates (Liski et al., 20