International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
At the current state Forestime lacks some key features and
functionality is still missing (e.g. the necessary user interaction
in the estimation process modeling) However, with a
reasonable effort Forestime can be developed into a fully-
featured operative forest variable estimation tool.
The results from the forest variable estimation tests show that
the relative errors stem volume and stem diameter estimates are
relatively small, while stem number and tree species
proportions estimates give higher errors. The upper limit for
stem volume estimates with the Forestime is around 300 m3/ha,
while with previous estimates the limit has been somewhere
between 200 m3/ha and 250 m3/ha. The large errors in the tree
species proportions are partly due to the fact that the proportion
of the broad-leaved trees is quite small in the ground data set.
The ground data validity is essential: the removal of erroneous
data points, and points residing too close to the border of very
different segments improved the estimate accuracy drastically.
The plot data obtained using a regular sampling grid is not
optimal for a system using segmentwise averaged feature
vectors as its inputs. The usage of this kind of ground data
requires additional manipulation and probably also removal of
relatively many sample data points. The best option would be
accurate standwise data with the average stand size close to
average Forestime segment size. In the future the availability of
accurate reference data obtained by airborne imaging may
prove useful. One has to remember also that the amount of the
ground data together with the range of the target variables
affects directly to the accuracy of the estimates.
5. CONCLUSIONS
In HighForest study VTT implemented the prototype version of
a modular software tool for forest variable estimation
(Forestime v. 1.0). Even if the functionality, speed and the
resource consumption of the implemented software are still
very limited, the operational tests have shown the feasibility of
the system. The software was tested as a stand-alone system, as
well as an external application program from commercial
software through the API. The operation of Forestime was as
expected and the objective for the software implementation was
well achieved.
Summarizing the results of forest variable estimation with
Forestime v 1.0, it can be stated that the objective to reduce the
reflectance saturation effect was achieved partly, as well as the
objective for accurate tree species estimation. The best results
were achieved using the spectral channels and the Haralick
entropy as input features. These are the recommended inputs for
the use of the system as its present form. The estimation speed
can still be improved, if only the spectral channels are used.
In the produced forest variable estimations, the target data
variance in the clusters is relatively large, which leads to
averaged estimates. One main issue of the future studies will be
the possibility to reduce the variance using different kind of
clustering, e.g. by taking the target variable information into the
process, and its effect on estimate accuracy. The future
development of the system should also focus on taking full
advantage of the contextual features in segmentation, and on
introducing means for easy ground reference data exploration.
The development of feature bank concept for the wide
operational use of the system is also regarded as one key
development areas of Forestime system.
ACKNOWLEDGEMENTS
The project was conducted in co-operation of four Finnish
parties: the Technical Research Centre of Finland (VTT), the
University of Helsinki (HY), Stora Enso Forest Consulting
(SEFC), and Gisnet Solutions Finland Oy. The HighForest
project had two parallel studies: the satellite image study (VTT)
and the aerial image study (HY). The Suonenjoki Research
Station of the Finnish Forest Research Institute (FFRI) provided
the sample plot ground data (measured in summer 2001) for the
system verification. Sixty percent of the funding was granted by
the National Technology Agency of Finland (Tekes).
REFERENCES
Haralick., R.M.; Shanmugam, K.; Dinstein, I. 1973: Textural
features for image classification, IEEE Transactions on
Systems, Man and Cybernetics, 3:610-621
Jain, AK. and Farrokhnia, F. 1991: Unsupervised Texture
Segmentation Using Gabor Filters, Pattern recognition, Vol. 24,
No. 12, pp. 1167-1186
Narendra, P.M., Goldberg, M. 1980. Image segmentation with
directed trees. IEEE Transaction on Pattern Analysis and
Machine Intelligence, Vol. PAMI-2, Nr. 2, March 1980.
Hime, T.; Stenberg, P.; Rauste, Y. A methodology to estimate
forest variables at sub-pixel level. - Conference on Remote
Sensing and Forest Monitoring. Rogow, PL, 1 - 3 June 1999.
European Communities (2000), s. 451 - 461.
Hime, T.; Stenberg, P.; Andersson, K.; Rauste, Y.; Kennedy,
P.; Folving, S.; Sarkeala, J.. AVHRR-based forest proportion
map of the Pan-European area. - Remote Sensing of
Environment. (2001) 77, s. 76 - 91.
Parmes, E.; Lohi, A.; Häme, T.; Holm, M.; Väätäinen, S. 2000.
Pienkuvioittaisen tiedon tuottaminen metsistä
kaukokartoituksen avulla. Loppuraportti VTT Automaatio,
2000.
Parmes, E. 1992. Segmentation of Spot and Landsat satellite
imagery. The Photogrammetric Journal of Finland, Vol. 13, Nr.
1, 1992. pp. 52 — 58.
Wulder, M.; Niemann, K.O.; Goodenough, D.G. 2002: error
reduction methods for local maximum filtering of high spatial
resolution imagery for locating trees, Canadian Journal of
Remote Sensing, Vol. 28, No. 5, pp. 621-628
References from websites:
Space Imaging; IKONOS Relative Spectral Response and
Radiometric Cal Coefficients ©2004 Space Imaging, Inc.,
http://www.spaceimaging.com/products/ikonos/spectral.htm
340
PROT
“ Corresr
° Dept. of
KEY WO
ABSTRA!
The protec
changed di
In this stuc
Combining
elevation 1
accurate at
using Mici
inquiry, m:
making fo
hectares to
areas such
the protect
The govert
function fu
achieved.
Taipei, th:
surroundec
ranging fi
headquarte
Taipei city
center of 1
natural haz
major thre
September
425mm pr
lives and
properties
and water
importance
effective m
Protection
protective
health, pla
protecting
salt (IUCN
initially es
with a tota
water, soil:
drawing m:
lands. Alt
provide sp
