Authors：Chi-Chuan Cheng

Year：2014

Key words：net primary productivity, remote sensing, vegetation indices

This study aimed to apply remote sensing to estimate the forest net primary productivity
(NPP) of Nanzhuang National Forest in Taiwan. The research processes included
calculating vegetation indices from SPOT images of 2 seasons in 2003, estimating the
fraction of photosynthetically active radiation (FPAR) and photosynthetically active
radiation absorbed by the different forest types (APAR), estimating the NPP, and finally
analyzing NPP variations from different seasons and forest types. Furthermore, the
shadow effect, simulation of the maximum light use efficiency for different forest types,
and the problem of image acquisition for NPP estimation in Taiwan were also investigated.
The results are as follows. Under the consideration of the shadow effect and simulation of
the maximum light use efficiency for different forest types, the NPP estimation on the dry
season image was 361.22 g C m-2 yr-1 with shadow retention and 293.19 g C m-2 yr-1 with
shadow correction, while the wet season image was 545.07 g C m-2 yr-1 with shadow
retention and 572.45 g C m-2 yr-1 with shadow correction. As for using dry- and wet-season
images, NPP values were 452.5 and 432.43 g C m-2 yr-1 with shadow retention and shadow correction, respectively. A comparison between the estimated NPP and the field-measured
carbon amount derived from forest inventory data (i.e., 430 g C m-2 yr-1) indicated that the
NPP estimated from 2 seasonal images had the best result because of the smallest bias.
Meanwhile, the seasonal analysis of NPP variations was significant in the study area. The
majority of NPP accumulation was about 86% of the annual NPP and was mainly distributed
between April and October. In addition, we propose that among the 3 shadow processes,
shadow removal cannot be applied to estimate the NPP because a lower FPAR was
generated when estimating the FPAR due to the linear transformation of vegetation indices.
We concluded that remote sensing is a timely, effective, feasible, and large-scale approach
for estimating the forest NPP and provides the NPP for a spatiotemporal variation analysis. Meanwhile, the shadow effect and simulation of the maximum light use efficiency for forest
types affect the estimation of forest NPP. Therefore, their effects should be considered
when applying  SPOT vegetation indices to estimate forest NPP. In addition, an alternative
approach using seasonal images is also feasible to eliminate the problem with image
acquisition in Taiwan.