Tezin Türü: Yüksek Lisans
Tezin Yürütüldüğü Kurum: Karadeniz Teknik Üniversitesi, Orman Fakültesi, Orman Mühendisliği, Türkiye
Tezin Dili: İngilizce
Öğrenci: Matthew Wojcik
Danışman: Uzay Karahalil
Özet:
Conducting a forest inventory is a crucial component of any forest management plan.
Traditional methods of taking field measurements are both time and labor intensive, in addition
to being cost prohibitive. Progressive developments in remote sensing technology have
provided an enhanced opportunity to collect increasingly precise forest stand parameter
estimations with less need for field surveys. This study makes use of spectral data and LiDAR
data from Sentinel-2 and ICESat-2 missions, respectively, as well as TIN-derived topographic
features to estimate stand volume and stand density in the Kocayaren Planning Unit, Turkey.
First, spectral data for 109 total plots extracted from Sentinel-2 including band reflectance
values and vegetation indices were compared. A combination of band reflectance values and
vegetation indices provided the best results. Spectral data was better able to predict stand
volume than tree density. The best model developed for stand volume using only spectral data
included the RVI and the NDMI, and bands 2, 3, 8a, 9, and 11, and yielded an adjusted R2 of
0.669 and RMSE of 39.125 m3 ha-1. The best model developed for number of trees included
the SAVI, EVI, RVI, IRECI, and bands 5 and 6, with a resulting adjusted R2 of 0.483 and
RMSE of 274 n ha-1. Secondly, further models including TIN-derived topographic features
were examined. The best models developed from variables that included topographic features
exhibited adjusted R2 values of 0.658 and 0.514 and RMSEs of 39.795 m3 ha-1 and 265 n ha-1
for stand volume and number of trees, respectively. Slope was the only topographic feature that
showed up in both models. Through the process of validation, it was determined that a
statistically significant relationship exists between the explanatory variables and stand volume,
but not with the number of trees, thereby rendering models for number of trees unsuitable.
Lastly, select plots that harbored ICESat-2 LiDAR data points were separated and models
including these data were created. It was found that the ICESat-2 LiDAR was not effective in
tree height estimation, and therefore did not have a significant relationship with stand volume
or the number of trees.