Pixel-based classification is considered as a classic method of extracting land-cover related information from remotely sensed imagery, and has been used in various applications, including vegetation mapping. However, several recent studies also mentioned the weakness of the pixel-based approach, including the vegetation index transformation, in mapping the structural composition of vegetation. This study aimed to test several pixel-based classification algorithms for mapping the structural composition of vegetation using Sentinel-2A (10 meters) imagery in Salatiga and its surrounding, Central Java. In this study three classification algorithms, namely Maximum Likelihood, Minimum Distance to Mean, and Support Vector Machine were compared with respect to their accuracy results in mapping the vegetation structural composition. The authors evaluated the effects of additional data in the classification process by comparing two different datasets, i.e. (i) the one using original bands only, and (ii) the one containing original bands and additional data in the form of several vegetation indices and Leaf Area Index (LAI). We collected field samples using stratified random strategy, which were separated into two sub-datasets, as a basis for structural composition classification reference and accuracy assessment. In addition, comparison was also carried out using the original results and the one which was majority filtered. The results showed that the Maximum Likelihood algorithm performed the highest accuracies at a range of 74-86% using a combination of original bands and RVI (Ratio Vegetation Index). The result that was processed using a 5x5 majority filter showed the highest accuracy 86.29%. These results demonstrated that the pixel-based classification of Sentinel 2A imagery using the Maximum Likelihood algorithm could be used to map the structural composition of vegetation in the study area.
Remote sensing has been widely used in the estimation of forest aboveground biomass (AGB) which is essential for climate change mitigation,by using either optical or radar data and its combination. This estimation of AGB from remote sensing data is now supported by the availability of the freely available dual-polarization Sentinel 1 SAR data. However, the assessment of the accuracy for measuring AGB from the VV and VH polarization in Sentinel-1 data in Indonesia is still limited. This study aims to assess the performance of VV and VH polarization and the combination with texture data from Sentinel-1 for estimating AGB in tropical forest of Barru Regency, South Sulawesi. The AGB was calculated by using backscatter value from C-band SAR dual-polarization and Grey Level of Measure (GLCM) texture data from Sentinel-1 as the independent variables, and ground inventory plots as the dependent variable. Twenty-three plots of field inventory data were collected whereas 16 plots were used in the regression models and the remaining seven plots were used to validate the result. The allometric equation was used to calculate the biomass value of the field survey data then multilinear regression models were generated by using biomass value, backscatter data from VV and VH polarization, and texture data. The performance of the resulted multilinear regression models was compared by looking at the coefficient of determination (R2) and RMSE value using cross-validation. The results demonstrated that combination of VH and GLCM texture suggest as the best to estimate the AGB based on higher value of R2 = 0.44 and SE 83.7 kg/tree. In conclusion, VH polarization usage in vegetation AGB modelling has been able to predict 3 % higher than by using VV polarisation. The inclusion of texture also had been able to increase the model performance by 5 to 7 % which demonstrated the importance of having texture variables in the analysis of AGB.
Noise in SAR imagery was produced due to different backscatter response from the objects in the earth surface. This resulted in a grainy image, usually known as “salt and pepper” noise, which reducing the capability to identify object from radar imagery. Therefore, speckle filtering was conducted to decrease this noise from SAR imagery. This study aims to assess the performance of different types of speckle filters especially when used to construct forest aboveground biomass (AGB) model from Sentinel-1 data in Barru Regency, South Sulawesi. There were 4 filters used in this study i.e. Frost, Gamma-MAP, Median, and Refined Lee. AGB model was constructed by using dual polarization C-band SAR of Sentinel1 data and ground inventory plots. 23 plots were collected in the field and the allometric equation was used to calculate the biomass value of the field survey data then cross validation models were generated by using biomass value and backscatter data from VV and VH polarization. Quality control was performed by comparing the coefficient of determination (R2 ) of those filters. The result shows that Frost filter especially on VH polarization was chosen as the bestfit model to estimate the AGB based on higher value of R2 (0.3464158) and RMSE (33.5231). The result demonstrated Frost filter as the best filter for retaining and/or enhancing the backscatter signal in Sentinel-1 data to be used in vegetation bio-physical modelling.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.