[STT48-P14] L-band radar backscatter variation due to the Amazonian deforestation
In this research, we will use the time series L band SAR data (ALOS / PALSAR and ALOS - 2 / PALSAR - 2) obtained from 2006 to 2017 for the change of radar backscattering in the Amazon deforestation area of Brazil By doing so, we could check the variation pattern of the backscatter coefficient during these periods.
The classification of forests and non-forests by SAR has been studied a lot in the past. The backscattering coefficient observed by radar is highly sensitive to forests and non-forests, and setting thresholds is important for classification to be performed with high accuracy. Whether or not this threshold value is stably determined may depend on the method of reducing forest and the place, careful investigation is necessary through comparison with local data.
We confirmed that there are three patterns in the change of back scattering coefficient associated with deforestation in Amazon. In the future, we will increase the evaluation area and extract forest fluctuation patterns in the target area. At the same time, we set thresholds for forest classification considering seasonal variation etc. Using Landsat, it was confirmed that this area has the greatest deforestation in the 1990s.
Whether or not the threshold is stably determined as mentioned above may depend on the way and location of forest reduction.
It is necessary to comprehensively consider three fluctuation patterns. From the graph of the average value of the backscattering coefficient in the evaluation area, by using Bayesian estimation from fluctuation of cumulative distribution function etc, we will lead to improvement of accuracy of threshold setting in forest non-forest classification in Amazon.
By incorporating JERS - 1 's image data, we plan to extract and characterize forest area variation over longer periods.
The classification of forests and non-forests by SAR has been studied a lot in the past. The backscattering coefficient observed by radar is highly sensitive to forests and non-forests, and setting thresholds is important for classification to be performed with high accuracy. Whether or not this threshold value is stably determined may depend on the method of reducing forest and the place, careful investigation is necessary through comparison with local data.
We confirmed that there are three patterns in the change of back scattering coefficient associated with deforestation in Amazon. In the future, we will increase the evaluation area and extract forest fluctuation patterns in the target area. At the same time, we set thresholds for forest classification considering seasonal variation etc. Using Landsat, it was confirmed that this area has the greatest deforestation in the 1990s.
Whether or not the threshold is stably determined as mentioned above may depend on the way and location of forest reduction.
It is necessary to comprehensively consider three fluctuation patterns. From the graph of the average value of the backscattering coefficient in the evaluation area, by using Bayesian estimation from fluctuation of cumulative distribution function etc, we will lead to improvement of accuracy of threshold setting in forest non-forest classification in Amazon.
By incorporating JERS - 1 's image data, we plan to extract and characterize forest area variation over longer periods.