Gašparović Mateo, PhD.

Digital terrain models (DTMs) present important data source for different applications in
environmental disciplines including forestry. At regional level, DTMs are commonly created
using airborne digital photogrammetry or airborne laser scanning (ALS) technology. This
study aims to evaluate the vertical accuracy of DTMs of different spatial resolutions derived
from high-density ALS data and existing photogrammetric (PHM) data in the dense lowland
even-aged pedunculate oak forests located in the Pokupsko basin in Central Croatia. As expected,
the assessment of DTMs’ vertical accuracy using 22 ground checkpoints shows higher
accuracy for ALS-derived than for PHM-derived DTMs. Concerning the different resolutions
of ALS-derived (0.5 m, 1 m, 2 m, 5 m) and PHM-derived DTMs (0.5 m, 1 m, 2 m, 5 m,
8 m) compared in this research, the ALS-derived DTM with the finest resolution of 0.5 m
shows the highest accuracy. The root mean square error (RMSE) and mean error (ME) values
for ALS-derived DTMs range from 0.14 m to 0.15 m and from 0.09 to 0.12 m, respectively,
and the values decrease with decreasing spatial resolution. For the PHM-derived DTMs, the
RMSE and ME values are almost identical regardless of resolution and they are 0.35 m and
0.17 m, respectively. The findings suggest that the 8 m spatial resolution is optimal for a
given photogrammetric data, and no finer than 8 m spatial resolution is required. This research
also reveals that the national digital photogrammetric data in the study area contain certain
errors (outliers) specific to the terrain type, which could considerably affect the DTM accuracy.
Thus, preliminary evaluation of photogrammetric data should be done to eliminate possible
outliers prior to the DTM generation in lowland forests with flat terrain. In the absence
of ALS data, the finding in this research could be of interests to countries, which still rely on
similar photogrammetric data for DTM generation.

The Airborne Laser Scanning (ALS) technology has been implemented in operational forest
inventories in a number of countries. At the same time, as a cost-effective alternative to ALS,
Digital Aerial Photogrammetry (PHM), based on aerial images, has been widely used for the
past 10 years. Recently, PHM based on Unmanned Aerial Vehicle (UAV) has attracted great
attention as well. Compared to ALS, PHM is unable to penetrate the forest canopy and, ultimately,
to derive an accurate Digital Terrain Model (DTM), which is necessary to normalize
point clouds or Digital Surface Models (DSMs). Many countries worldwide, including Croatia,
still rely on PHM, as they do not have complete DTM coverage by ALS (DTMALS). The
aim of this study is to investigate if the official Croatian DTM generated from PHM (DTMPHM)
can be used for data normalization of UAV-based Digital Surface Model (DSMUAV) and estimating
plot-level mean tree height (HL) in lowland pedunculate oak forests. For that purpose,
HL estimated from DSMUAV normalized with DTMPHM and with DTMALS were generated and
compared as well as validated against field measurements. Additionally, elevation errors in
DTMPHM were detected and eliminated, and the improvement by using corrected DTMPHM
(DTMPHMc) was evaluated. Small, almost negligible variations in the results of the leave-oneout
cross-validation were observed between HL estimated using proposed methods. Compared
to field data, the relative root mean square error (RMSE%) values of HL estimated from DSMUAV
normalized with DTMALS, DTMPHM, and DTMPHMc were 5.10%, 5.14%, and 5.16%, respectively.
The results revealed that in the absence of DTMALS, the existing official Croatian DTM
could be readily used in remote sensing based forest inventory of lowland forest areas. It can
be noted that DTMPHMc did not improve the accuracy of HL estimates because the gross errors
mainly occurred outside of the study plots. However, since the existence of the gross errors in
Croatian DTMPHM has been confirmed by several studies, it is recommended to detect and
eliminate them prior to using the DTMPHM in forest inventory.

High temporal resolution of synthetic aperture radar (SAR) imagery (e.g., Sentinel-1 (S1) imagery) creates new possibilities for monitoring green vegetation in urban areas and generating land-cover classification (LCC) maps. This research evaluates how different pre-processing steps of SAR imagery affect classification accuracy. Machine learning (ML) methods were applied in three different study areas: random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGB). Since the presence of the speckle noise in radar imagery is inevitable, different adaptive filters were examined. Using the backscattering values of the S1 imagery, the SVM classifier achieved a mean overall accuracy (OA) of 63.14%, and a Kappa coefficient (Kappa) of 0.50. Using the SVM classifier with a Lee filter with a window size of 5×5 (Lee5) for speckle reduction, mean values of 73.86% and 0.64 for OA and Kappa were achieved, respectively. An additional increase in the LCC was obtained with texture features calculated from a grey-level co-occurrence matrix (GLCM). The highest classification accuracy obtained for the extracted GLCM texture features using the SVM classifier, and Lee5 filter was 78.32% and 0.69 for the mean OA and Kappa values, respectively. This study improved LCC with an evaluation of various radiometric and texture features and confirmed the ability to apply an SVM classifier. For the supervised classification, the SVM method outperformed the RF and XGB methods, although the highest computational time was needed for the SVM, whereas XGB performed the fastest. These results suggest pre-processing steps of the SAR imagery for green infrastructure mapping in urban areas. Future research should address the use of multitemporal SAR data along with the pre-processing steps and ML algorithms described in this research.