The aim of this paper is to propose the convolution neural network—VGG16 structure in Keras to classify pavement defects. In this paper, we present a

This study tested solutions aiming to improve the estimates of tropical forest stand volume (SV) for the large area using SPOT 5 data and field data. Firstly, forests were stratified into four forest strata using a hybrid classification of ISODATA (Iterative Self - Organization Data Analysis) unsupervised, and Maximum likelihood supervised (MLC) based on SPOT 5 image and field data. A set of 111 sample plots was distributed in these four forest strata, which represented disturbed forests under different levels by human-induced. Within the sample plots, diameter at breast height (DBH) and tree height (H) were measured to calculate standing volume using the SV equation from the previous study for the area. The method of k-NN (k-Nearest Neighbour) was applied to estimate the stand volume for different datasets of SPOT 5 bands, normalized difference vegetation index (NDVI), and combination of SPOT 5 bands and NDVI. The estimates were separately tested for the whole area and each forest stratum. Leave one out cross-validation method was employed to validate the quality of the predictions. The results indicated the accuracy of the estimate was significantly improved when applying for each stratum comparing to for the whole area with both SPOT 5 and NDVI data. The lower errors were found in the forest strata that are fewer disturbances than the heavily degraded stratum. Among the image data, the estimates were based on the NDVI give a lower accuracy compared to others.

Bao HD, Huong NTT. 2025. Estimating forest above ground biomass in Dak Lak Province, Vietnam. Asian J For 9: 115-123. Accurately estimating forest Above Ground Biomass (AGB) is essential for assessing carbon stocks, monitoring forest health, and guiding sustainable management practices. This study examined the potential of Landsat 8 satellite imagery for AGB estimation in Dak Lak Province, located in Vietnam's Central Highlands. Field data collected from 415 sample plots across diverse forest types including evergreen broadleaf, semi-deciduous, and dry dipterocarp forests revealed AGB values ranging from 15.16 to 299.33 tons/ha. Multivariate linear regression (MLR) and random forest (RF) models were applied to predict AGB using spectral bands and vegetation indices derived from Landsat 8 imagery. The MLR model demonstrated limited predictive capability (R² = 0.131), indicating that linear relationships between spectral data and AGB were insufficient to capture the system's complexity. In contrast, the RF model exhibited superior predictive performance, achieving an R² of 0.653 and a concordance R² of 0.571 when using spectral bands alone. Incorporating vegetation indices alongside spectral bands further improved the RF model’s accuracy (R² = 0.671, concordance R² = 0.586). Spatial analysis revealed considerable variability in AGB across forest types, with evergreen broadleaf forests exhibiting the highest biomass values. These findings highlight the effectiveness of satellite remote sensing and machine learning for cost-effective biomass estimation. Moreover, they highlight the urgent need for such approaches in forest management and climate change mitigation efforts in tropical regions, including Vietnam.

Understanding land use and land cover change (LULCC) is essential for evaluating landscape dynamics and supporting sustainable land management, particularly in rapidly changing tropical environments. This study analyzed long-term LULCC patterns in Dak Nong Province, Vietnam, during 2010–2023 using multi-temporal Landsat imagery (Landsat 5/7/8), spectral indices (NDVI, NDWI, BSI), a digital elevation model (DEM), and the Random Forest (RF) algorithm. Two classification scenarios were tested, and the extended feature set (Landsat bands + NDVI, NDWI, BSI, DEM) significantly improved classification performance compared with using Landsat bands and NDVI alone. Overall accuracy increased from 74.92% to 83.77%, while the Kappa coefficient improved from 0.70 to 0.81 for the 2023 map. Applying the optimized model across the time series produced stable classification accuracies (OA: 81.41–89.52%; Kappa: 0.77–0.87). Results indicate substantial natural forest loss, decreasing from 122,023 ha in 2010 to 77,943 ha in 2023, mainly due to conversion to plantation forests and agricultural land. Plantation forests expanded rapidly, while rubber plantations peaked in 2011 before declining. The findings reveal complex land-use transitions and highlight the ecological risks associated with natural forest decline. This study provides a consistent long-term LULC dataset and identifies spatial hotspots of forest conversion, offering important evidence for forest management and sustainable land-use planning in Vietnam’s Central Highlands.

This study describes a new species of Camellia section Piquetia, C. dakplaoensis, from Dak Nong Province, Vietnam. Morphologically, it closely resembles C. langbianensis but differs in key floral and fruit traits, including plain yellow flowers, pedicels with 4 or 5 larger lanceolate bracteoles, adaxially glabrous sepals with a cuspidate apex, ovate to widely orbicular and adaxially glabrous petals, wholly yellow and glabrous filaments, and distinctly ridged capsules. Conservation assessments classify C. dakplaoensis as Critically Endangered (CR) under IUCN criteria. During a taxonomic re-evaluation of species within the section, C. campanulata and C. longii were proposed as heterotypic synonyms of C. cattienensis due to a lack of significant morphological distinctions and publication priority. With these findings, section Piquetia now includes 15 species, comprising eight with yellow flowers and seven with red to purple flowers.