Research Highlights: In this study, we classified natural forest into four forest types using time-series multi-source remotely sensed data through a proposed semi-supervised model developed and validated for mapping forest types and assessing forest transition in Vietnam. Background and Objectives: Data on current forest state and changes detection are always essential for forest management and planning. There is, therefore, a need for improved tools to classify and evaluate forest dynamics more accurately and effectively. Our objective is to develop such tools using a semi-supervised model and landscape metrics to classify and map changes in natural forest types by using multi-source remotely sensed data. Materials and Methods: A combination of Landsat data with PALSAR and PALSAR-2 was used for forest classification through the proposed semi-supervised model. This model turned a kernel least square into a self-learning algorithm, trained by a small number of samples with given labels, and then used this classifier to assign labels to the unlabeled data. The overall accuracy, kappa, user’s accuracy, and producer’s accuracy were used to evaluate the classification accuracy by comparing the classified image with the results of ground truth interpretation. Based on the classified images, forest transition was evaluated using certain landscape metrics at the class and landscape levels. Results: The multi-source data approach achieved improved discrimination of forest types compared to only using single data (optical or radar data). Good classification accuracies were obtained, with kappas of 0.81, 0.76, and 0.74 for the years 2007, 2010, and 2016, respectively. The analysis of landscape metrics indicated that there were different behaviors in the four forest types, as well as provided much information about the trends in spatial pattern changes. Conclusions: This study highlights the utilization of a semi-supervised model in forest classification, and the analysis of forest transition using landscape metrics. However, future research should include a comparison of different models to estimate the improvement of the proposed model. Another important study that should be conducted is to test the proposed method on larger areas.

There is a demand for better information on forest biomass in tropical regions for use in carbon accounting. This needs robust above-ground biomass (AGB) estimation in different forest types. Our study sought to improve biomass estimation by selecting the best regression models based on observations of the contribution of radar signals to AGB in five forest types in Vietnam. Data from PALSAR and PALSAR-2, which covered the forest area, were used to extract 16 polarimetric radar (PolSAR) parameters in 2007 and 2016. This study was designed as a comparative experiment of four regression models: linear, polynomial, support vector machine (SVR) and random forest. First, the contribution of PolSAR data to AGB estimation was evaluated using two approaches: the sample data from all forest types, and the five individual forest types (rich, medium, poor, restoration and bamboo forest). Second, we examined the improvement of AGB prediction by selecting the important variables and assessing the best models for different forest types. The results showed an improvement in the value of R-squared and RMSE using the five individual forest types compared to the combined forest types. In particular, using a multivariate model, RMSE values were enhanced by 9-18% for the rich forest, and by 80-85% for the remaining forest types in all models. SVM provided the best performance for medium and poor forest (RMSE of 8.27 tons ha -1 and 12.38 tons ha -1 , respectively), random forest for bamboo (RMSE of 23.18 tons ha -1 ), and the polynomial regression for the restoration forest (RMSE *of 10.11 tons ha -1 ). Further research is required to derive a more robust AGB estimation model for the rich forest.

Abstract. In this study, above-ground biomass (AGB) performance was evaluated by PALSAR-2 L-band and Landsat data for bamboo and mixed bamboo forest. The linear regression model was chosen and validated for forest biomass estimation in A Luoi district, Thua Thien Hue province, Vietnam. A Landsat 8 OLI image and a dual-polarized ALOS/PALSAR-2 L-band (HH, HV polarizations) were used. In addition, 11 diferrent vegetation indices were extracted to test the performance of Landsat data in estimating forest AGB Total of 54 plots were collected in the bamboo and mixed bamboo forest in 2016. The linear regression is used to evaluate the sensitivity of biomass to the obtained parameters, including radar polarization, optical properties, and some vegetation indices which are extracted from Landsat data. The best-fit linear regression is selected by using the Bayesian Model Average for biomass estimation. Leave-one-out cross-validation (LOOCV) was employed to test the robustness of the model through the coefficient of determination (R squared – R2) and Root Mean Squared Error (RMSE). The results show that Landsat 8 OLI data has a slightly better potential for biomass estimation than PALSAR-2 in the bamboo and mixed bamboo forest. Besides, the combination of PALSAR-2 and Landsat 8 OLI data also has a no significant improvement (R2 of 0.60) over the performance of models using only SAR (R2 of 0.49) and only Landsat data (R2 of 0.58–0.59). The univariate model was selected to estimate AGB in the bamboo and mixed bamboo forest. The model showed good accuracy with an R2 of 0.59 and an RMSE of 29.66 tons ha−1. The comparison between two approaches using the entire dataset and LOOCV demonstrates no significant difference in R (0.59 and 0.56) and RMSE (29.66 and 30.06 tons ha−1). This study performs the utilization of remote sensing data for biomass estimation in bamboo and mixed bamboo forest, which is a lack of up-to-date information in forest inventory. This study highlights the utilization of the linear regression model for estimating AGB of the bamboo forest with a limited number of field survey samples. However, future research should include a comparison with non-linear and non-parametric models.