9:45 AM - 10:00 AM
[MIS01-04] Determining quantitative and qualitative characteristics of mixed forest stands using Sentinel-1 imagery
Keywords:forest characteristics, radar imagery, forest density, standing volume, Sentinel-1, remote sensing
The study aims to develop methods for determining the forest characteristics using radar space observation data. Based on the forest inventory data and optical survey materials and considering forests' diversity and growing conditions, 190 sampling plots were established. The fieldworks were carried out in Kostroma, Vologda, Arkhangelsk, Nizhny Novgorod regions, and the Udmurt Republic (Russia).
The radar images (RI) of the Sentinel-1 satellite aged less than three years from establishing plots were analyzed. To calculate the correlation between forest characteristics and radar survey data, the images were processed according to three options:
1. Applying the incoherent accumulation filter
2. Applying the Frost filter
3. No data filtering
Radar indices and Haralik texture features were computed for each data set acquired from three amplitude radar image pre-processing variants. Data were processed with the open-source software SNAP. The Graph Builder and Batch Processing modules were used to work on the imagery. These modules allow building and applying prepared processing algorithms ("graphs") with customizable parameters.
The study results showed that forest characteristics are related to:
- specific radar cross-section (SRCS) values in gamma-zero on the VV-polarization (GammaVV),
- the radar index (the ratio of four SRCS objects in the value of gamma-zero on the VH-polarization to the sum of the SRCS objects in the value of gamma-zero on VV and VH-polarizations (RVI)
- textural features "total mean" (GLCMMean),
- the scatter of the mean value of reference and neighboring pixels combinations (GLCMVariance),
- the linear connection value of the pixel pairs brightness levels (GLCMCCorrelation).
Determining forest characteristics based on radar satellite images using various methods of its pre-processing show similar efficiency (Table 1).
To sum up, the highest correlations are obtained by applying the incoherent accumulation procedure for imagery data pre-processing. The research also revealed that the most accurate determination of the standing volume and forest density could be acquired by using multiple factor regression models.
The radar images (RI) of the Sentinel-1 satellite aged less than three years from establishing plots were analyzed. To calculate the correlation between forest characteristics and radar survey data, the images were processed according to three options:
1. Applying the incoherent accumulation filter
2. Applying the Frost filter
3. No data filtering
Radar indices and Haralik texture features were computed for each data set acquired from three amplitude radar image pre-processing variants. Data were processed with the open-source software SNAP. The Graph Builder and Batch Processing modules were used to work on the imagery. These modules allow building and applying prepared processing algorithms ("graphs") with customizable parameters.
The study results showed that forest characteristics are related to:
- specific radar cross-section (SRCS) values in gamma-zero on the VV-polarization (GammaVV),
- the radar index (the ratio of four SRCS objects in the value of gamma-zero on the VH-polarization to the sum of the SRCS objects in the value of gamma-zero on VV and VH-polarizations (RVI)
- textural features "total mean" (GLCMMean),
- the scatter of the mean value of reference and neighboring pixels combinations (GLCMVariance),
- the linear connection value of the pixel pairs brightness levels (GLCMCCorrelation).
Determining forest characteristics based on radar satellite images using various methods of its pre-processing show similar efficiency (Table 1).
To sum up, the highest correlations are obtained by applying the incoherent accumulation procedure for imagery data pre-processing. The research also revealed that the most accurate determination of the standing volume and forest density could be acquired by using multiple factor regression models.