Search

High-resolution land cover mapping is crucial in various disciplines but is often hindered by the lack of accurately matched labels. Our study introduces an innovative deep learning methodology for effective land cover mapping, independent of matched labels. The approach comprises three main components: (1) An advanced fully convolutional neural network, augmented with super-resolution features, to refine labels; (2) The application of an instance-batch normalization network (IBN), leveraging these enhanced labels from the source domain, to generate 2-m resolution land cover maps for test sites in the target domain; (3) Noise assessment tests to evaluate the impact of varying noise levels on the model’s mapping accuracy using external labels. The model achieved an overall accuracy of 83.40% in the target domain using endogenous super-resolution labels. In contrast, employing exogenous, high-precision labels from the National Land Cover Database in the source domain led to a notable accuracy increase of 2.55%, reaching 85.48%. This improvement highlights the model’s enhanced generalizability and performance during domain shifts, attributed significantly to the IBN layer. Our findings reveal that, despite the absence of native high-precision labels, the utilization of high-quality external labels can substantially benefit the development of precise land cover mapping, underscoring their potential in scenarios with unmatched labels.

Fine fuel load (FFL) is a crucial variable influencing the occurrence of wildfire. Accurate knowledge of the distribution of FFL in mountainous forests is essential for ongoing wildfire risk management and the stability of mountain ecosystems. Traditional methods of estimating forest fuel load typically involve ground surveys combined with remote sensing, which can be costly and inefficient. Therefore, low-cost, large-scale FFL estimation remains challenging. In this study, Sentinel-2A satellite imagery from the Changsha forest region was used as the data source. Firstly, different feature variables were constructed based on false-color (B843), true-color (B432), four-band (B8432) combinations, and the Normalized Difference Water Index (NDWI). Subsequently, a machine learning approach based on random convolution was employed to estimate FFL. This study also included accuracy assessments of the estimation results and the creation of FFL maps for the study area. The results showed that the FFL estimation based on the B8432 band combination achieved the highest accuracy, with RMSE and R2 values of 5.847 t·hm−2 and 0.656, respectively. FFL estimation results based on false-color imagery followed, with true-color imagery and NDWI index-based estimation results exhibiting lower accuracy. This study offers critical FFL insights using random convolution techniques applied to Sentinel-2A imagery, enhancing the ability to monitor and manage forest fuel conditions effectively, thereby facilitating more informed regional wildfire risk management strategies.

Change detection is a crucial task in remote sensing that finds broad application in land resource planning, forest resource monitoring, natural disaster monitoring, and evaluation. In this paper, we propose a change detection model for cross-domain recognition, which we call CrossCDNet. Our model significantly improves the modeling ability of the change detection on one dataset and demonstrates good generalization on another dataset without any additional operations. To achieve this, we employ a Siamese neural network for change detection and design an IBNM (Instance Normalization and Batch Normalization Module) that utilizes instance normalization and batch normalization in order to serve as the encoder backbone in the Siamese neural network. The IBNM extracts feature maps for each layer, and the Siamese neural network fuses the feature maps of the two branches using a unique operation. Finally, a simple MLP decoder is used for end-to-end change detection. We train our model on the LEVIR-CD dataset and achieve competitive performance on the test set. In cross-domain dataset testing, CrossCDNet outperforms all the other compared models. Specifically, our model achieves an F1-score of 91.69% on the LEVIR-CD dataset and an F1-score of 77.09% on the WHU-CD dataset, where the training set was LEVIR-CD.

Abstract Cliff ecosystems are considered the ‘Last Unknown’ because cliffs may host ancient and unique species that are located in extremely hostile environments and are difficult to reach. However, comprehensive and systematic information and databases of cliff flora are unavailable; and obtaining information on these undisturbed ecosystems and cliff flora is fundamental and is of high priority for botanists. We present the first systematically sampled expert‐verified checklist of cliff flora from the ground to the top of cliffs in these harsh environments based on high‐resolution visible imagery captured with close‐range photogrammetry by a smart remote‐controlled unmanned aerial system (UAS). Using approximately 9200 records retrieved from high‐resolution images of 197 cliffs in the Wulingyuan global cultural heritage area, a mid‐subtropical region in China, we identified 267 vascular plant species, 140 genera and 58 families, excluding herbaceous plants. The maximum diversity was attributable to a few dominant families and genera, and most of the genera and families were represented by only a few species. Our findings highlight the unique floristic patterns as well as morphology on these cliffs that warrant further study and the potential for the use of a smart UAS to investigate cliff flora at large and global scales.

Camellia oleifera is a vital economic crop of southern China. Accurate mapping and monitoring of Camellia oleifera plantations are essential for promoting sustainable operations within the Camellia oleifera industry. However, traditional remote sensing interpretation methods are no longer feasible for the large-scale extraction of plantation areas. This study proposes a novel deep learning-based method that utilizes GF-2 remote sensing imagery to achieve precise mapping and efficient monitoring of Camellia oleifera plantations. First, we conducted a comparative analysis of the performance of various semantic segmentation models using a self-compiled dataset of Camellia oleifera plantations. Subsequently, we proceeded to validate the prediction results obtained from the most effective deep-learning network model for Camellia oleifera plantations in Hengyang City. Finally, we incorporated DEM data to analyze the spatial distribution patterns. The findings indicate that the U-Net++ network model outperforms other semantic segmentation methods when applied to our self-generated dataset of Camellia oleifera plantations. It achieves a recall rate of 0.89, a precision rate of 0.92, and an mIOU of 0.83, demonstrating the effectiveness of the proposed method in identifying and monitoring Camellia oleifera plantations. By combining the predicted results with the data from DEM, we discovered that these plantations are typically situated at elevations ranging from 50 to 200 m, with slopes below 25°, and facing south or southeast. Moreover, a significant positive spatial correlation and clustering phenomenon are observed among the townships in Hengyang City. The method proposed in this study facilitates rapid and precise identification and monitoring of Camellia oleifera plantations, offering significant theoretical support and a scientific foundation for the management and ecological conservation of Camellia oleifera plantations.

The cliff ecosystem is one of the least human-disturbed ecosystems in nature, and its inaccessible and often extreme habitats are home to many ancient and unique plant species. Because of the harshness of cliff habitats, their high elevation, steepness of slopes, and inaccessibility to humans, surveying cliffs is incredibly challenging. Comprehensive and systematic information on cliff vegetation cover is not unavailable but obtaining such information on these cliffs is fundamentally important and of high priority for environmentalists. Traditional coverage survey methods—such as large-area normalized difference vegetation index (NDVI) statistics and small-area quadratic sampling surveys—are not suitable for cliffs that are close to vertical. This paper presents a semi-automatic systematic investigation and a three-dimensional reconstruction of karst cliffs for vegetation cover evaluation. High-resolution imagery with structure from motion (SFM) was captured by a smart unmanned aerial vehicle (UAV). Using approximately 13,000 records retrieved from high-resolution images of 16 cliffs in the karst region Guilin, China, 16 models of cliffs were reconstructed. The results show that this optimized UAV photogrammetry method greatly improves modeling efficiency and the vegetation cover from the bottom to the top of cliffs is high-low-high, and very few cliffs have high-low cover at the top. This study highlights the unique vegetation cover of karst cliffs, which warrants further research on the use of SFM to retrieve cliff vegetation cover at large and global scales.

As one of the four most important woody oil-tree in the world, Camellia oleifera has significant economic value. Rapid and accurate acquisition of C. oleifera tree-crown information is essential for enhancing the effectiveness of C. oleifera tree management and accurately predicting fruit yield. This study is the first of its kind to explore training the ResU-Net model with UAV (unmanned aerial vehicle) images containing elevation information for automatically detecting tree crowns and estimating crown width (CW) and crown projection area (CPA) to rapidly extract tree-crown information. A Phantom 4 RTK UAV was utilized to acquire high-resolution images of the research site. Using UAV imagery, the tree crown was manually delineated. ResU-Net model’s training dataset was compiled using six distinct band combinations of UAV imagery containing elevation information [RGB (red, green, and blue), RGB-CHM (canopy height model), RGB-DSM (digital surface model), EXG (excess green index), EXG-CHM, and EXG-DSM]. As a test set, images with UAV-based CW and CPA reference values were used to assess model performance. With the RGB-CHM combination, ResU-Net achieved superior performance. Individual tree-crown detection was remarkably accurate (Precision = 88.73%, Recall = 80.43%, and F1score = 84.68%). The estimated CW (R2 = 0.9271, RMSE = 0.1282 m, rRMSE = 6.47%) and CPA (R2 = 0.9498, RMSE = 0.2675 m2, rRMSE = 9.39%) values were highly correlated with the UAV-based reference values. The results demonstrate that the input image containing a CHM achieves more accurate crown delineation than an image containing a DSM. The accuracy and efficacy of ResU-Net in extracting C. oleifera tree-crown information have great potential for application in non-wood forests precision management.

Dynamic detection of forest change is the fundamental method of monitoring forest resources and an essential means of preserving the accuracy and timeliness of forest land resource data. This study focuses on a deep learning-based method for dynamic forest change detection using Sentinel-2 satellite data, especially within mountainous areas. First, the performance of various deep learning models (U-Net++, U-Net, LinkNet, DeepLabV3+, and STANet) and various loss functions (CrossEntropyLoss(CELoss), DiceLoss, FocalLoss, and their combinations) are compared on a self-made dataset. Next, the best model and loss function is used to predict the annual forest change in Hunan Province from 2017 to 2021, and the detection results are evaluated in 12 sample areas. Finally, forest changes are detected in Sentinel-2 images for each quarter of 2017–2021. In addition, a dynamic detection map of forest change in Hunan Province from 2017 to 2021 is drawn. The results reveal that the U-Net++ model and the CELoss performed the best on the self-made dataset, with a Precision of 0.795, a Recall of 0.748, and an F1-score of 0.771. The results of annual and quarterly forest change detection were consistent with the changes in the Sentinel-2 images with accurate boundaries. This result demonstrates the high practicality and generalizability of the method used in this paper. This paper achieves a rapid and accurate extraction of multi-temporal Sentinel-2 image forest change areas based on the U-Net++ model, which can be used as a benchmark for future large territorial areas monitoring and management of forest resources.

The use of remote sensing images to detect forest changes is of great significance for forest resource management. With the development and implementation of deep learning algorithms in change detection, a large number of models have been designed to detect changes in multi-phase remote sensing images. Although synthetic aperture radar (SAR) data have strong potential for application in forest change detection tasks, most existing deep learning-based models have been designed for optical imagery. Therefore, to effectively combine optical and SAR data in forest change detection, this paper proposes a double Siamese branch-based change detection network called DSNUNet. DSNUNet uses two sets of feature branches to extract features from dual-phase optical and SAR images and employs shared weights to combine features into groups. In the proposed DSNUNet, different feature extraction branch widths were used to compensate for a difference in the amount of information between optical and SAR images. The proposed DSNUNet was validated by experiments on the manually annotated forest change detection dataset. According to the obtained results, the proposed method outperformed other change detection methods, achieving an F1-score of 76.40%. In addition, different combinations of width between feature extraction branches were analyzed in this study. The results revealed an optimal performance of the model at initial channel numbers of the optical imaging branch and SAR image branch of 32 and 8, respectively. The prediction results demonstrated the effectiveness of the proposed method in accurately predicting forest changes and suppressing cloud interferences to some extent.

Increasing the area of planted forests is rather important for compensation the loss of natural forests and slowing down the global warming. Forest growing stem volume (GSV) is a key indicator for monitoring and evaluating the quality of planted forest. To improve the accuracy of planted forest GSV located in south China, four L-band ALOS PALSAR-2 quad-polarimetric synthetic aperture radar (SAR) images were acquired from June to September with short intervals. Polarimetric characteristics (un-fused and fused) derived by the Yamaguchi decomposition from time series SAR images with different intervals were considered as independent variables for the GSV estimation. Then, the general linear model (GLM) obeyed the exponential distribution were proposed to retrieve the stand-level GSV in plantation. The results show that the un-fused power of double bounce scatters and four fused variables derived from single SAR image is highly sensitive to the GSV, and these polarimeric characteristics derived from the time series images more significantly contribute to improved estimation of GSV. Moreover, compared with the estimated GSV using the semi-exponential model, the employed GLM model with less limitations and simple algorithm has a higher saturation level (nearly to 300 m3/ha) and higher sensitivity to high forest GSV values than the semi-exponential model. Furthermore, by reducing the external disturbance with the help of time average, the accuracy of estimated GSV is improved using fused polarimeric characteristics, and the estimation accuracy of forest GSV was improved as the images increase. Using the fused polarimetric characteristics (Dbl×Vol/Odd) and the GLM, the minimum RRMSE was reduced from 33.87% from single SAR image to 24.42% from the time series SAR images. It is implied that the GLM is more suitable for polarimetric characteristics derived from the time series SAR images and has more potential to improve the planted forest GSV.

Spectral reflectance distortions caused by terrain and solar illumination seriously reduce the accuracy of mapping forest carbon density, especially in mountainous regions. Many models have been developed for mitigating or eliminating the terrain effects on the quality of remote sensing images in hilly and mountainous areas. However, these models usually use global parameters, which may lead to overcorrections for regions with poor illumination and steep slopes. In this study, we present a local parameter estimation (LPE) method based on a pixel-moving window for topographic correction (TC), which can be considered as a general optimization framework for most semiempirical TC models. We set seven kernel sizes for the presented framework, which are 15 pixels, 25 pixels, 50 pixels, 100 pixels, 250 pixels, 500 pixels, and 1000 pixels, respectively. The proposed method was then applied to four traditional TC models, Minnaert (MIN), C Correction (CC), Sun Canopy Sensor + C (SCSC) and Statistical Empirical Correction (SEC), to form four new TC models. These new models were used to estimate forest carbon density of a mountainous area in Southern China using field plot data and a Landsat 8 image. Four evaluation methods, including correlation analysis, the stability of land covers, comparison of reflectance between sunlit and shaded slopes, and accuracy assessment of forest carbon density, were employed to evaluate the contributions of moving window sizes, and assess the performance of the TC models for forest carbon density estimation. The results show that the four TC models with LPE perform much better than the traditional TC models in reducing the topographic effects and improving the estimation accuracy of forest carbon density for the study area. Among the traditional TC models, SEC performs slightly better than SCSC, CC, and MIN. Therefore, the SEC-based model with LPE, that is, LPE-SEC, gets greater R2 and smaller relative RMSE values in estimating forest carbon density than other models. Moreover, all the means of the predicted forest carbon density values fall in the confidence interval of the validation data at a significant level of 0.05. Overall, this study implies that the proposed method with LPE provides great potential to improve the performance of TC and forest carbon density estimation for the study area. It is expected that the improved TC method can be applied to other mountainous areas to improve the quality of remotely sensed images.

For the planning and sustainable management of forest resources, well-managed plantations are of great significance to mitigate the decrease of forested areas. Monitoring these planted forests is essential for forest resource inventories. In this study, two ALOS PALSAR-2 quad-polarimetric synthetic aperture radar (SAR) images and ground measurements were employed to estimate growing stem volume (GSV) of Chinese fir plantations located in a hilly area of southern China. To investigate the relationships between forest GSV and polarization characteristics, single and fused variables were derived by the Yamaguchi decomposition and the saturation value of GSV was estimated using a semi-exponential empirical model as a base model. Based on the estimated saturation values and relative root mean square error (RRMSE), the single and fused characteristics and corresponding models were selected and integrated, which led to a novel saturation-based multivariate method used to improve the GSV estimation and mapping of Chinese fir plantations. The new findings included: (1) All the original polarimetric characteristics, statistically, were not significantly correlated with the forest GSV, and their logarithm and ratio transformation fused variables greatly improved the correlations, thus the estimation accuracy of the forest GSV. (2) The logarithm transformation of surface scattering resulted in the greatest saturation, value but the logarithm transformation of double-bounce scattering resulted in the smallest RRMSE of the GSV estimates. (3) Compared with the single transformations, the fused variables led to more reasonable saturation values and obviously reduced the values of RRMSE. (4) The saturation-based multivariate method led to more accurate estimates of the forest GSV than the univariate method, with the smallest value (29.64%) of RRMSE achieved using the set of six variables. This implied that the novel saturation-based multivariate method provided greater potential to improve the estimation and mapping of the forest GSV.

There is a large amount of remote sensing data available for land use and land cover (LULC) classification and thus optimizing selection of remote sensing variables is a great challenge. Although many methods such as Jeffreys–Matusita (JM) distance and random forests (RF) have been developed for this purpose, the existing methods ignore correlation and information duplication among remote sensing variables. In this study, a novel approach was proposed to improve the measures of potential class separability for the selection of remote sensing variables by taking into account correlations among the variables. The proposed method was examined with a total of thirteen spectral variables from a Gaofen-1 image, three class separability measures including JM distance, transformed divergence and B-distance and three classifiers including Bayesian discriminant (BD), Mahalanobis distance (MD) and RF for classification of six LULC types at the East Dongting Lake of Hunan, China. The results showed that (1) The proposed approach selected the first three spectral variables and resulted in statistically stable classification accuracies for three improved class separability measures. That is, the classification accuracies using three or more spectral variables statistically did not significantly differ from each other at a significant level of 0.05; (2) The statistically stable classification accuracies obtained by integrating MD and BD classifiers with the improved class separability measures were also statistically not significantly different from those by RF; (3) The numbers of the selected spectral variables using the improved class separability measures to create the statistically stable classification accuracies by MD and BD classifiers were much smaller than those from the original class separability measures and RF; and (4) Three original class separability measures and RF led to similar ranks of importance of the spectral variables, while the ranks achieved by the improved class separability measures were different due to the consideration of correlations among the variables. This indicated that the proposed method more effectively and quickly selected the spectral variables to produce the statistically stable classification accuracies compared with the original class separability measures and RF and thus improved the selection of the spectral variables for the classification.

Combining sample plot and image data has been widely used to map forest carbon density at local, regional, national and global scales. When mapping is conducted using multiple spatial resolution images at different scales, field observations have to be collected at the corresponding resolutions to match image values in pixel sizes. Given a study area, however, to save time and cost, field observations are often collected from sample plots having a fixed size. This will lead to inconsistency of spatial resolutions between sample plots and image pixels and impede the mapping and product quality assessment. In this study, a methodological framework was proposed to conduct mapping and accuracy assessment of forest carbon density at four spatial resolutions by combining remotely sensed data and reference values of sample plots from a systematical, nested and clustering sampling design. This design led to one field observation dataset at a 30 m spatial resolution sample plot level and three other reference datasets by averaging the observations from three, five and seven sample plots within each of 250 m and 500 m sub-blocks and 1000 m blocks, respectively. The datasets matched the pixel values of a Landsat 8 image and three MODIS products. A sequential Gaussian co-simulation (SGCS) and a sequential Gaussian block co-simulation (SGBCS), an upscaling algorithm, were employed to map forest carbon density at the spatial resolutions. This methodology was tested for mapping forest carbon density in Huang-Feng-Qiao forest farm of You County in Eastern Hunan of China. The results showed that: First, all of the means of predicted forest carbon density values at four spatial resolutions fell in the confidence intervals of the reference data at a significance level of 0.05. Second, the systematical, nested and clustering sampling design provided the potential to obtain spatial information of forest carbon density at multiple spatial resolutions. Third, the relative root mean square error (RMSE) of predicted values at the plot level was much greater than those at the sub-block and block levels. Moreover, the accuracies of the up-scaled estimates were much higher than those from previous studies. In addition, at the same spatial resolution, SGCSWA (scaling up the SGCS and Landsat derived 30 m resolution map using a window average (WA)) resulted in smallest relative RMSEs of up-scaled predictions, followed by combinations of Landsat images and SGBCS. The accuracies from both methods were significantly greater than those from the combinations of MODIS images and SGCS. Overall, this study implied that the combinations of Landsat 8 images and SGCSWA or SGBCS with the systematical, nested and clustering sampling design provided the potential to formulate a methodological framework to map forest carbon density and conduct accuracy assessment at multiple spatial resolutions. However, this methodology needs to be further refined and examined in other forest landscapes.