Advances in Remote Sensing for Forestry: Theory, Methods, Applications, and Validation

1. Introduction

Forests comprise the largest ecosystem in the world and provide significant value to natural ecosystems and human society [1]. They provide a habitat for a wide range of animal, plant, and microbial species and are invaluable in the maintenance of biodiversity [2]. Furthermore, they provide ecological and social services to human society, furthering our sustainable development. On the one hand, forests are a vital source of timber and various other products that are valuable to human society; on the other hand, forests deliver a range of essential ecological services, including soil and water conservation, windbreaks, sand stabilization, forest therapy, and health-enhancing recreational opportunities [3]. The forest ecosystem also plays a critical role in the fight against climate change. Its capacity to sequester atmospheric carbon dioxide via photosynthesis alleviates the effects of global warming [4], and forests are generally understood to be among the most economically viable and effective means of carbon sequestration. This is because the worldwide forest carbon sink is equal to almost half of the global emissions from burning fossil fuels [5]. However, recent research has indicated that the worldwide forest sink has been weakened by deforestation and advancing forest age [6]. Clarifying the precise nature of the forest carbon sink is thus of particular importance.

Remote sensing technology has become central to forestry research [7], having played a significant role in forestry investigation, forest cover, forest health, carbon storage, forest fire and pest monitoring, and forest dynamics over vast regions, even on global scale, and it is likely to become yet more essential in the future. This Special Issue, titled “Advances in Remote Sensing in Forestry: Theory, Methods, Applications and Validation,” comprises 11 papers, focussing on emerging theories and methodologies in forest surveillance and monitoring, the application of remote sensing in forest ecosystem evaluation and biodiversity monitoring, and new technologies for monitoring carbon sinks in forest ecosystems. A couple of the studies included herein concern themselves with the theory underpinning remote sensing in forestry context [8,9]. Some of studies evaluate new sensors and equipment, such as Airborne or UAV LiDAR, which aid in forest resource investigation by reducing labor and time costs [10,11,12]. Additionally, emerging methods, such as deep learning, have been used to extract certain forest features from remote sensing data [13]. The most important application of remote sensing technology in forestry is the solving of real problems that need to be addressed in the forestry industry, such as the recognition of forest types, the detection of clear-cutting activities, and restoration assessment [14,15,16]. In meeting the challenges of global warming, forestry can also play a key role; scholars have estimated the forest carbon stock using remote sensing in a large region and have evaluated the significant role played by forests as a global carbon sink [17,18]. The studies included herein showcase a number of theoretical innovations and the application of new methods to solve real problems in forestry. Our hope is that this Special Issue will serve as a crucial reference in promoting the application of remote sensing technology in the forestry domain.

2. Overview of Published Articles

Liu et al.’s article (Contribution 1) focuses on innovations in the application of remote sensing in forestry. The research address heterogeneous and discontinuous canopy structures and proposes an innovative inversion method combining radiative transfer models (RTMs), namely, PRO4SAIL, PRO4SAIL2 and PROGeoSail, with the random forest (RF) algorithm to estimate the canopy water content. A Gaussian copula model and prior knowledge are used to constrain the model parameters to alleviate the problem or poor positioning in remote sensing inversion. In the end, the authors compare the estimated results of the coupled PRO4SAIL, PRO4SAIL2, and PROGeoSail models. The results show that the coupled PROSPECT-5B and GeoSail models achieve the best performance, with an R2 of 0.68 and a RMSE of 0.15 kg·m⁻². The forest canopy water content is estimated from MODIS products using the Google Earth Engine (GEE). The estimated results reveal the spatiotemporal heterogeneity of forest CWC across the contiguous US from west to east and show an increasing trend from January to August and a decreasing trend thereafter. This research demonstrates the superiority of 3D RTMs in forest canopy parameter inversion and provides an efficient, feasible technical scheme for forest CWC estimation in heterogeneous forest canopies, offering important references for forest drought stress monitoring and ecological protection against a background of global climate change.

Zhang et al.’s article (Contribution 2) is also focused on the estimation of forest features. The authors successfully described the three-dimensional (3D) distribution characteristics of Leaf Mass per Area (LMA) in the canopies of broad-leaved tree species, combining LiDAR and hyperspectral remote sensing data. In this study, the authors collected the leaf samples of eight dominant broad-leaved species by layer and direction and established four prediction models (PLS, LMM, SVM, and RF), integrating vegetation indices, DBH, and 3D leaf position parameters. The results show that the LMA indicates significant spatial heterogeneity: it will decrease with increasing canopy depth and increase with horizontal distance from the tree trunk. The authors also obtained some interesting results with respect to the LMA distribution in the canopy, revealing that it has no significant correlation with the orientation of leaf. Among the four models, random forest (RF) achieved the highest accuracy, with an R² ranging from 0.716 to 0.939 according to a 10-fold cross-validation method. The study presents an interesting study of high-precision 3D LMA prediction, and it could be extended to large-scale 3D LMA monitoring via LiDAR and optical remote sensing. At the same time, it offers a new perspective in clarifying canopy photosynthetic capacity and nutrient allocation mechanisms and could prove helpful in optimizing the vegetation biophysical parameter models used to estimate the carbon cycle study.

Su et al.’s article (Contribution 3) aims to solve the problems of incomplete lower canopy structure and sparse distribution in medium–low-density UAV point clouds caused by self-occlusion. The study proposes a tree DBH via the dual enhancement strategies of hierarchical random sampling and a spatially constrained loss function. The results show that this strategy avoids the loss of information in sparse areas and improves the structural bias of traditional methods. Moreover, it provides better results in trunk structure recovery and DBH measurement accuracy, as validated on FOR-instance and the Xiong’an dataset. This study provides a reference method for correcting the structural bias in medium–low-density point cloud processing.

You et al.’s article (Contribution 4) proposes a tree skeletonization method based on DBSCAN clustering to address the noise and occlusion defects of terrestrial laser scanning (TLS). The results show that the average positional deviation is 0.418 cm, and the average directivity deviation is 8.474 degrees. This method can be used to describe the constructed skeleton and tree’s geometric structure and branch hierarchies, which is useful in accurately retrieving the tree parameter, the 3D model of the tree, and the forest inventory.

Li et al.’s article (Contribution 5) focuses on the application of new sensors, such as Airborne LiDAR, for estimating the forest inventory. They propose an error-in-variable simultaneous equation (SEq) for constructing compatible forest inventory estimation models. SEqs ensures consistency with forest mensuration principles by integrating stand volume, basal area, mean height, and diameter at breast height. The results show that SEqs’ rRMSE increases by less than 2% across several forest species. Furthermore, this work showcases a forest inventory estimating method that can maintain both biophysical and mathematical compatibility. It will enhance the robustness and applicability of forest resource inventory and management models.

Ru et al.’s article (Contribution 6) proposes an automated forest monitoring system based on satellite imagery and deep learning to monitor the global deforestation. The authors integrate spatial pyramid pooling (SPP) modules into the U-Net architecture to segment forest areas without having to consider varying forest sizes, colors, and related factors. Compared with FCN, SegNet, and TernausNet, the model proposed in this study performs well in multi-scale feature extraction, with accuracies of 86.71%, 75.59%, and 82.88% on high-difficulty datasets. This work provides a reliable tool for detecting illegal logging and evaluating forest conservation efforts.

Xu et al.’s article (Contribution 7) considered an automatic extraction method for mapping the spatial distribution of Picea schrenkiana in the Tianshan Mountains. This work addresses the challenge of accounting for sample points, remote-sensing images, and classification features during high-resolution, large-scale mapping in complex terrain conditions by integrating Jeffries–Matusita (JM) distance and the random forest classifier in the Google Earth Engine (GEE). The results show that the approach achieves high accuracy, with an overall accuracy (OA) of 91.93% and a Kappa coefficient of 0.89. Furthermore, it provides a reliable method for coniferous forest mapping in complex terrains.

Pedraza et al.’s research (Contribution 8) generates annual cloud-free mosaic data from optical remote sensing data and employs a random forest algorithm for forest/non-forest classification to establish a forest dynamic monitoring and restoration assessment system for the period 1996–2021. The results indicate that the reduction in forest area was about 5587 ha over 25 years. This work provides a reference for quantifying forest dynamics and supporting forest conservation.

Choi et al.’s article (Contribution 9) also focus on the detection of forest dynamics. The authors propose a forest clear-cutting detection method via random forest (RF) classifier by integrating national spatial data and Sentinel-2 satellite imagery. Roughly 8164.0 ha of clear-cut areas, which accounted for 2.48% of the total forest area, were detected. These results fill a gap in the official statistical data and provide support for carbon sink management and sustainable forest management.

Mohamedi et al.’s article (Contribution 10) focuses on the issues of carbon peak and carbon neutrality in China and simulates China’s forest net ecosystem productivity (NEP) from 2013 to 2023 using a 3-PGS model. The results show that the forests in question sequestered 1.71 ± 0.09 PgC over the decade, with an annual average of 0.156 ± 0.0071 PgC, and evergreen broadleaf forests showed the highest NEP, as well as significant spatial–temporal heterogeneity across climatic regions in China. This research enriches the study of China’s estimated forest carbon sink, provides evidence for the effectiveness of ecological restoration policies, and promotes the achievement of China’s dual-carbon goal.

Liu et al.’s article (Contribution 11) concerns the spatial–temporal variations and driving mechanisms of vegetation NPP in mainland China from 2000 to 2022, employing the gravity center model, third-order partial correlation analysis, and geographical detector methods. The results show that the NPP decreases from southeast to northwest, and the gravity centers migrate southward in Northeast, Northwest, and North China. On the other hand, other regions show the opposite results, with gravity centers migrating northward. This research indicates that human activities dominated NPP growth in 2000–2010, while climate change became the primary driver in 2011–2022. It offers a reference for the impact of natural and anthropogenic factors in monitoring ecosystem restoration and ecological protection. It also contributes to the realization of China’s carbon peak and carbon neutrality policies.

3. Conclusions

This Special Issue, titled “Advances in Remote Sensing in Forestry: Theory, Methods, Applications and Validation,” collects 11 high-quality papers focussing on important scientific issues, systematically presenting the latest research on the application of remote sensing technology in forestry. Represented in this Special Issue are the core fields of forest canopy parameters inversion, forest structure reconstruction, the monitoring of forest dynamics, forest carbon sink estimation, and so on, alongside remote sensing equipment/platforms such as LiDAR and UAV hyper-spectral sensors. These studies not only enrich the field of remote sensing application in forestry but also offer solutions and references for the key issues in forestry management and ecological protection. This Special Issue also provides a number of scientific references for the furtherance of carbon peak and carbon neutrality policies. The studies published in this Special Issue demonstrate the latest applications of remote sensing technology in forestry, but it should be borne in mind that they cover only a fraction of the research being undertaken in this field. New technologies, such as artificial intelligence algorithms, big data mining technologies, and integrated space–air–ground monitoring technologies, have expanded the research horizon of the forestry domain. Our expectation is that this Special Issue and related work will motivate more scholars to engage with the ongoing research on remote sensing in forestry and promote the continuous innovation of new theories and methods in this domain.