University of Twente Student Theses

This website will be unavailable due to maintenance December 1st between 8:00 and 12:00 CET.

Estimation of forest canopy height using single polarized TanDEM-X across different forest biophysical characteristics in temperate forests

Gebremeskel, Haftom Hagos (2022) Estimation of forest canopy height using single polarized TanDEM-X across different forest biophysical characteristics in temperate forests.

[img] PDF
Abstract:Canopy height measurement is important to understand the forest's vertical structure and biomass quantification. In comparison to aerial-based, satellite-based canopy height estimation is useful in terms of covering large area and frequent measurements. TanDEM-X, a spaceborne active remote sensing, has been recently used to estimate canopy height. A simplified RVoG model was suggested to compute canopy height from TDX interferometric coherence. The model showed encouraging results in a well studied boreal forests and some temperate forests. Hence, the aim of this study is to further understand the simplified RVoG model (linear and sinc) on canopy height estimation of temperate forests considering forest type, slope class and canopy cover percentage. A single polarized TDX coherence data from five acquisitions with different height of ambiguity (HoA) was tested for the canopy height estimation. Two of the acquisitions were from October (with HoA of 35.2 m and 44.7 m) and three of acquisition were from January (with HoA of 41.3 m, 68.3 m, and 91.5 m). LiDAR point clouds were used to generate LiDAR canopy height and used it as reference. Root mean square error (RMSE), relative RMSE (RMSEr), coefficient of determination (R2 ), absolute error (AE), and relative AE (AEr) were used to assess the accuracies. Mann Witney-U test and Kruskal-Wallis test were used to test difference between forest types and among slope classes, respectively. Linear regression was used to assess the impact of canopy cover percentage. A regression test was also held for one selected acquisition to analyse the impact of slope, and canopy cover estimation error. The results showed the RMSE and R2 were different among acquisitions depending on the HoA and season of acquisition (leaf condition, precipitation, temperature). The RMSE ranged from 4.3 m to 5.7 m for the linear model and from 5.2 m to 16 m for the sinc model. The R2 for both models were similar ranged from 0.14 to 0.48. The RMSEr and R2 showed that coniferous forests had better estimation accuracies than broadleaved forests during the two October acquisitions, and one January acquisition that had high precipitation and temperature. In addition, for all acquisitions, the AEr showed that coniferous forest had significantly lower AEr than broadleaved forests. The RMSE and R2 did not show a trend across slope classes, for all acquisitions. Whereas the AEr showed that, gentle slope had significantly lower AEr than steep slope for the two acquisitions with highest HoA. For similar acquisitions, the AEr significantly decreases with increasing canopy cover. The regression analysis showed that slope (coefficient = 0.24) and canopy cover percentage (coefficient = -0.54) significantly (p<0.01) explained 3% of the variation in absolute error for broad leaved forest. However, for the coniferous forests, only canopy cover percentage had significant (p<0.05) influence on absolute error, with an R2 nearly to zero. Overall, canopy height estimation from single polarized TDX coherence gives moderate accuracy (RMSEr < 27%) across different biophysical characteristics of temperate forests. To obtain better accuracy for broadleaved forests, leaf season and weather conditions should be taken into consideration.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Subject:43 environmental science
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page