首页 > , Vol. , Issue () : -
Forest fires have broken out frequently in recent years, which severely damages the structure and function of forest ecosystem around the world. Initial assessment of burn severity after forest fires could provide a quantitative basis for rapid implementations of restoration measures in burned areas. In last decades, remote sensing-based models have become an appropriate choice to assess burn severity, which generally requires a certain amount of field survey data. However, this requirement could not be satisfied sufficiently in most cases, since the field survey work would cost a lot of time, labor, and money. Therefore, this would largely limit the efficient application of remote sensing technology to the initial assessment of burn severity. To improve the time-efficiency of traditional remote sensing-based models, a transfer learning algorithm (i.e., SSTCA, semi-supervised Transfer Component Analysis) was employed in this study to build an initial assessment model of burn severity. At first, the SSTCA algorithm was applied to project a series of new features from original spectral features of remotely sensed data. Based on these projected features, a SVR (Support Vector Regression) model was then trained using historical field survey data of source areas (i.e., Bear fire on June 27, 2002 and Mule fire on July 11, 2002). After that, this combined SSTCA-SVR model was transferred to the initial assessment of burn severity of a target area (i.e., Lushan fire on March 30, 2020). Finally, its performance was quantitatively compared with that of some traditional models (i.e., dNDVI-, dLST-, dNBR-, and SVR-based models). Results showed that original spectral features of remote sensing images over source and target areas were quite different. After the projection of SSTCA, projected features of source and target samples have similar distribution patterns in the new features-based space. Meanwhile, in the initial assessment of burn severity, dNDVI- and dNBR-based models have overestimated burn severity levels with the lowest accuracy (i.e., overall accuracy was from 20.80% to 24.80% and Kappa value was between 0.01 and 0.06). Compared to these spectral indices-based models, dLST-based model has a better performance with an overall accuracy of 34.80% and a Kappa value of 0.19. Although SVR-based model has shown a promising performance with an overall accuracy of 58.00% and a Kappa value of 0.48, this machine learning model still has overestimated burn severity in some regions of burned areas. Compared to that of mentioned-above models, assessment results of burn severity levels using the SSTCA-SVR model was more accurate with an overall accuracy of 71.20% and a Kappa value of 0.64. It could be concluded that the application of a transferring learning algorithm would be very helpful for building an assessment model of burn severity with a good transferring ability. In this way, more accurate results could be obtained in the initial assessment of burn severity and the response of post-fire management could thus be speeded up after forest fires.