Jianyang Liu

Against the backdrop of climate change, forest fires increasingly undermine ecosystem stability and reshape species distributions in Heilongjiang Province. Therefore, quantifying the drivers of fire occurrence and conducting long-term fire risk forecasting holds critical value for regional ecological security. Centered on the forested regions of Heilongjiang Province, this study systematically assessed the relative contributions of multi-source factors—including topography, vegetation, and meteorological conditions—to fire occurrence and compared the predictive performance of three models: Deep Neural Network with Residual Connections (ResDNN), Artificial Neural Network (ANN), and Support Vector Machine (SVM). Modeling results based on historical fire records indicated that the ResDNN model achieved the highest accuracy (85.6%). Owing to its robust nonlinear mapping capability, it performed better in capturing complex feature interactions than ANN and SVM. These results demonstrate its strong applicability to forest fire prediction in Heilongjiang Province. Building on these findings, the study employed the best-performing ResDNN model in conjunction with CMIP6 multi-model climate projections to simulate and map the spatiotemporal probability of forest fire occurrence from 2030 to 2070. The results provide an intuitive representation of long-term fire-risk trajectories under future climate scenarios and offer scientific support for regional fire prevention, monitoring, early-warning systems, and forest management under climate change.

In the realm of global climate change and environmental protection, the precise estimation of forest ecosystem carbon density is essential for devising effective carbon management and emission reduction strategies. This study employed forest inventory, soil carbon, and remote sensing data combined with three models—Random Forest (RF), Geographically Weighted Regression (GWR), and the innovative Geographically Weighted Random Forest (GWRF) model—integrated with remote sensing technology to develop a framework for assessing the regional spatial distribution of the forest vegetation carbon density (FVC) and forest soil carbon density (FSC). The findings revealed that the GWRF model outperformed the other models in estimating both the FVC and FSC. The data indicated that the FVC in Heilongjiang Province ranged from 4.91 t/ha to 72.39 t/ha, with an average of 40.88 t/ha. In contrast, the average FSC was 182.29 t/ha, with a range of 96.01 t/ha to 255.09 t/ha. Additionally, the forest ecosystem carbon density (FEC) varied from 124.36 t/ha to 302.18 t/ha, averaging 223.17 t/ha. Spatially, the FVC, FSC, and FEC exhibited a consistent growth trend from north to south. The results of this study demonstrate that machine learning models that consider spatial relationships can improve predictive accuracy, providing valuable insights for the future spatial modeling of forest carbon storage.

Information of burned areas (BAs) is particularly important for research on forest fires. Mapping BAs contributes to managing and preventing further destruction. BA mapping has long been constrained by the temporal and spatial resolution of satellite data and classification accuracy, while also neglecting the spectral characteristics of optical satellites. To address this issue, a novel method for mapping and extracting BAs using a semantic segmentation model has been proposed herein. Leveraging medium spatial resolution satellites, Landsat-8 and Sentinel-2, and after determining the optimal spectral band combination, a spectral–spatial module was integrated into the U-Net; this results in a new model, known as the SSU-Net. SSU-Net precisely extracted BAs from a single post-fire image. The optimal band combinations for Landsat-8 and Sentinel-2 were bands 7-5-4 and 12-8-4, respectively. SSU-Net achieved an F1 score of 0.9296 and an intersection over union (IoU) of 0.8707 on Landsat-8 data. It achieved slightly higher performance on Sentinel-2 with an F1 score of 0.9314 and IoU of 0.8752. Compared with U-Net, SSU-Net improved the F1 scores by 4.8% and 3.4% for Landsat-8 and Sentinel-2, respectively, and IoU by 7.7% and 5.9%. In terms of generalisation, SSU-Net achieved an F1 score of 0.9151 when trained on Landsat-8 data and tested on Sentinel-2 data, and 0.9313 vice versa, with improvements of 2.1% and 5.1% over U-Net, respectively. SSU-Net also demonstrated higher accuracy and generalisation in BA mapping, supporting the integration of multisource remote sensing data and emphasising its crucial role in sustainable forest management on a global scale.

Timely and reliable estimation of forest stock volume is essential for sustainable forest management and conservation. Light detection and ranging (LiDAR) data can provide an effective depiction of the three-dimensional structure information of forests, but its large-scale application is hampered by spatial continuity. This study aims to construct a LiDAR sampling framework, combined with multi-sensor imagery, to estimate the regional forest stock volume of natural secondary forests in Northeast China. Two sampling approaches were compared, including systematic sampling and classification-based sampling. First, the forest stock volume was mapped using a combination of field measurement data and full-coverage LiDAR data. Then, the forest stock volume obtained in the first step of estimation was used as a reference value, and optical images and topographic features were combined for secondary modeling to compare the effectiveness and accuracy of different sampling methods, including 12 systematic sampling and classification-based sampling methods. Our results show that the root mean square error (RMSE) of the 12 systematic sampling approaches ranged from 55.81 to 57.42 m³/ha, and the BIAS ranged from 21.55 to 24.89 m³/ha. The classification-based LiDAR sampling approach outperformed systematic sampling, with an RMSE of 55.56 (3/ha) and a BIAS of 20.68 (3/ha). This study compares different LiDAR sampling approaches and explores an effective LiDAR sample collection scheme for estimating forest stock, while balancing cost and accuracy. The classification-based LiDAR sampling approach described in this study is easy to apply and portable and can provide a reference for future LiDAR sample collection.

The accurate classification of single tree species in forests is important for assessing species diversity and estimating forest productivity. However, few studies have explored the influence of canopy morphological characteristics on the classification of tree species. Therefore, based on UAV LiDAR and hyperspectral data, in this study, we designed various classification schemes for the main tree species in the study area, i.e., birch, Manchurian ash, larch, Ulmus, and mongolica, in order to explore the effects of different data sources, classifiers, and canopy morphological features on the classification of a single tree species. The results showed that the classification accuracy of a single tree species using multisource remote sensing data was greater than that based on a single data source. The classification results of three different classifiers were compared, and the random forest and support vector machine classifiers exhibited similar classification accuracies, with overall accuracies above 78%. The BP neural network classifier had the lowest classification accuracy of 75.8%. The classification accuracy of all three classifiers for tree species was slightly improved when UAV LiDAR-extracted canopy morphological features were added to the classifier, indicating that the addition of canopy morphological features has a certain relevance for the classification of single tree species.

Forests are the largest terrestrial ecosystem with major benefits in three areas: economy, ecology, and society. However, the frequent occurrence of forest fires has seriously affected the structure and function of forests. To provide a strong scientific basis for forest fire prevention and control, Ripley’s K(d) function and the LightGBM algorithm were used to determine the spatial pattern of forest fires in four different provinces (Heilongjiang, Jilin, Liaoning, Hebei) in China from 2019 to 2021 and the impact of driving factors on different ecosystems. In addition, this study also identified fire hotspots in the four provinces based on kernel density estimation (KDE). An artificial neural network model (ANN) was created to predict the probability of occurrence of forest fires in the study area. The results showed that the forest fires were spatially clustered, but the variable importance of different factors varied widely among the different forest ecosystems. Forest fires in Heilongjiang and Liaoning Provinces were mainly caused by human-driven factors. For Jilin, meteorological factors were important in the occurrence of fires. Topographic and vegetation factors exhibited the greatest importance in Hebei Province. The selected driving factors were input to the ANN model to predict the probability of fire occurrence in the four provinces. The ANN model accurately captured 93.17%, 90.28%, 83.16%, and 89.18% of the historical forest fires in Heilongjiang, Jilin, Liaoning, and Hebei Provinces; Precision, Recall, and F-measure based on the full dataset are 0.87, 0.88, and 0.87, respectively. The results of this study indicated that there were differences in the driving factors of fire in different forest ecosystems. Different fire management policies must be formulated in response to this spatial heterogeneity.

Although low-intensity forest fires are a necessary part of healthy echo system, high-intensity forest fires continue to affect the diversity of forest ecosystems and species. Therefore, it is necessary to study the driving factors of forest fires and explore their possible locations and probabilities in complex forest terrain. In this article, we determined the relative influences of different types of factors on the occurrence of forest fires in Heilongjiang forest areas and compared the performance of artificial neural networks and logistic regression for wildfire prediction. By analyzing Heilongjiang forest fire data from 2002 to 2015 and constructing a model, we found that climate factors, topographical factors, and vegetation type factors play a crucial role in Heilongjiang’s wildfires. During the fire prevention period, temperature and wind speed have a more significant influence than other factors. According to the variable screening that we conducted, the model built by the variables that we used can predict 84% of forest fires in Heilongjiang Province. For recent wildfires (2019–2020) in most areas, we can use artificial neural networks for relatively accurate verification (85.2%). Therefore, artificial neural networks are very suitable for the prediction of forest fires in Heilongjiang Province. Through the prediction results, we also created a probability distribution map of fire occurrence in the study area. On this basis, we also analyzed the changes in the probability of natural fires under the weather changing trend, which can effectively aid in fire prevention and extinguishment.