Deep Learning Based Retrieval of Forest Aboveground Biomass from Combined LiDAR and Landsat 8 Data

Assessing carbon stock and BVOCs emissions from dominant tree species in Beijing

Urban areas face environmental pollution and greenhouse emissions challenges, demand collaborative efforts to mitigation. Urban forests play a crucial role in absorbing CO2 emissions and contributing to carbon sequestration potential, but they also release biogenic volatile organic compounds (BVOCs), which contribute to the formation of tropospheric ozone and secondary organic aerosols (SOA). This study aimed to understand the role of urban forests in carbon stock and BVOCs emission by establishing optimal biomass models for six typical tree species (Robinia pseudoacacia, Quercus, Populus, Pinus tabulaeformis, Betula platyphylla, and Larix gmelinii) in Beijing. Biomass models were developed using field surveys and remote sensing data, with R2 values ranging from 0.364 to 0.921. Applying these models to forest resource inventory data, carbon stock and BVOCs emission models were constructed. In 2021, the total carbon stock for these pure forest tree species was estimated at 5.638 million tons, with a carbon density of 58.86 t/ha. The carbon density ranking for pure forest tree species was: Robinia pseudoacacia > Populus tomentosa > Betula platyphylla > Quercus Linn > Pinus tabulaeformis > Larix gmelinii. Total BVOCs emission in 2021 from the studied species were calculated at 25,789.72 t, with an average emission of 0.27 t/ha. Populus tomentosa had the highest BVOCs emission per unit area, followed by Robinia pseudoacacia, and Larix gmelinii had the smallest. Betula platyphylla and Robinia pseudoacacia were identified as species with high carbon stock and low BVOCs emissions in Beijing, offering insights for future urban forest planning and eco-friendly urban environment development strategies.

Harnessing Artificial Intelligence, Machine Learning and Deep Learning for Sustainable Forestry Management and Conservation: Transformative Potential and Future Perspectives

Plants serve as the basis for ecosystems and provide a wide range of essential ecological, environmental, and economic benefits. However, forest plants and other forest systems are constantly threatened by degradation and extinction, mainly due to misuse and exhaustion. Therefore, sustainable forest management (SFM) is paramount, especially in the wake of global climate change and other challenges. SFM ensures the continued provision of plants and forests to both the present and future generations. In practice, SFM faces challenges in balancing the use and conservation of forests. This review discusses the transformative potential of artificial intelligence (AI), machine learning, and deep learning (DL) technologies in sustainable forest management. It summarizes current research and technological improvements implemented in sustainable forest management using AI, discussing their applications, such as predictive analytics and modeling techniques that enable accurate forecasting of forest dynamics in carbon sequestration, species distribution, and ecosystem conditions. Additionally, it explores how AI-powered decision support systems facilitate forest adaptive management strategies by integrating real-time data in the form of images or videos. The review manuscript also highlights limitations incurred by AI, ML, and DL in combating challenges in sustainable forest management, providing acceptable solutions to these problems. It concludes by providing future perspectives and the immense potential of AI, ML, and DL in modernizing SFM. Nonetheless, a great deal of research has already shed much light on this topic, this review bridges the knowledge gap.

Heavy metal concentrations prediction of marine sediments by visible-near infrared spectroscopy based on attention mechanism

Monitoring heavy metal concentrations in marine sediments is important for assessing marine environmental quality, protecting ecological health, and preventing human health risks. Visible-near infrared spectroscopy technology can overcome the shortcomings of traditional methods. Taking marine sediments from Jiaozhou Bay, Qingdao as an example, this paper collected visible-near infrared spectroscopy of the marine sediments and innovatively proposed a two-scale LSTM with attention mechanism method to predict heavy metal concentrations in marine sediments. By extracting two scale spectral features and combining the attention mechanism to fuse according to the difference in contribution degree of each feature, the method strengthens the effective feature extraction and improves the prediction accuracy of heavy metal concentration in Marine sediments. The results demonstrate that compared to multiple existing model methods, the prediction performance using the two-scale LSTM with attention mechanism method was improved. Specifically, for the sediment Cu concentrations model, Rc2 = 0.820, RMSEC= 1.911, Rp2 = 0.720, RMSEP= 2.849, RPD= 1.907; for the sediment As concentrations model, Rc2 = 0.925, RMSEC= 0.893, Rp2 = 0.838, RMSEP= 1.315, RPD= 2.492; and for the sediment Zn concentrations model, Rc2 = 0.815, RMSEC= 4.741, Rp2 = 0.702, RMSEP= 5.827, RPD= 1.850. This study provides a new tool for rapid analysis of Cu, As, and Zn in sediments of Jiaozhou Bay and other regions using visible-near infrared spectroscopy.

Mitigating saturation effects in rice nitrogen estimation using Dualex measurements and machine learning

Nitrogen is essential for rice growth and yield formation, but traditional methods for assessing nitrogen status are often labor-intensive and unreliable at high nitrogen levels due to saturation effects. This study evaluates the effectiveness of flavonoid content (Flav) and the Nitrogen Balance Index (NBI), measured using a Dualex sensor and combined with machine learning models, for precise nitrogen status estimation in rice. Field experiments involving 15 rice varieties under varying nitrogen application levels collected Dualex measurements of chlorophyll (Chl), Flav, and NBI from the top five leaves at key growth stages. Incremental analysis was performed to quantify saturation effects, revealing that chlorophyll measurements saturated at high nitrogen levels, limiting their reliability. In contrast, Flav and NBI remained sensitive across all nitrogen levels, accurately reflecting nitrogen status. Machine learning models, particularly random forest and extreme gradient boosting, achieved high prediction accuracy for leaf and plant nitrogen concentrations (R2 > 0.82), with SHAP analysis identifying NBI and Flav from the top two leaves as the most influential predictors. By combining Flav and NBI measurements with machine learning, this approach effectively overcomes chlorophyll-based saturation limitations, enabling precise nitrogen estimation across diverse conditions and offering practical solutions for improved nitrogen management in rice cultivation.

Long-term annual estimation of forest above ground biomass, canopy cover, and height from airborne and spaceborne sensors synergies in the Iberian Peninsula

The Mediterranean Basin has experienced substantial land use changes as traditional agriculture decreased and population migrated from rural to urban areas, which have resulted in a large forest cover increase. The combination of Landsat time series, providing spectral information, with lidar, offering three-dimensional insights, has emerged as a viable option for the large-scale cartography of forest structural attributes across large time spans. Here we develop and test a comprehensive framework to map forest above ground biomass, canopy cover and forest height in two regions spanning the most representative biomes in the peninsular Spain, Mediterranean (Madrid region) and temperate (Basque Country). As reference, we used lidar-based direct estimates of stand height and forest canopy cover. The reference biomass and volume were predicted from lidar metrics. Landsat time series predictors included annual temporal profiles of band reflectance and vegetation indices for the 1985-2023 period. Additional predictor variables including synthetic aperture radar, disturbance history, topography and forest type were also evaluated to optimize forest structural attributes retrieval. The estimates were independently validated at two temporal scales, i) the year of model calibration and ii) the year of the second lidar survey. The final models used as predictor variables only Landsat based metrics and topographic information, as the available SAR time-series were relatively short (1991-2011) and disturbance information did not decrease the estimation error. Model accuracies were higher in the Mediterranean forests when compared to the temperate forests (R2 = 0.6-0.8 vs. 0.4-0.5). Between the first (1985-1989) and the last (2020-2023) decades of the monitoring period the average forest cover increased from 21 ± 2% to 32 ± 1%, mean height increased from 6.6 ± 0.43 m to 7.9 ± 0.18 m and the mean biomass from 31.9 ± 3.6 t ha-1 to 50.4 ± 1 t ha-1 for the Mediterranean forests. In temperate forests, the average canopy cover increased from 55 ± 4% to 59 ± 3%, mean height increased from 15.8 ± 0.77 m to 17.3 ± 0.21m, while the growing stock volume increased from 137.8 ± 8.2 to 151.5 ± 3.8 m3 ha-1. Our results suggest that multispectral data can be successfully linked with lidar to provide continuous information on forest height, cover, and biomass trends.

Mapping soil organic carbon distribution across South Africa's major biomes using remote sensing-topo-climatic covariates and Concrete Autoencoder-Deep neural networks

The management of soil organic carbon (SOC) stocks remains at the forefront of greenhouse gas mitigation. However, unprecedented anthropogenic disturbances emanating from continued land-use change have significantly altered SOC distribution across global biomes leading to considerable carbon losses. Consequently, understanding the spatial distribution of SOC across different biomes, particularly at larger scales, is critical for climate change policy formulation and planning. Advancements in remote sensing, availability of big data, and deep learning architecture offer great potential in large-scale SOC mapping. In this regard, this study mapped SOC distribution across South Africa's major biomes using remotely sensed-topo-climatic data and Concrete Autoencoder-Deep neural networks (CAE-DNN). From the different deep neural frameworks tested, the CAE-DNN model (developed from 26 selected covariates) achieved the best accuracy with an RMSE value of 7.91 t/ha (about 20 % of the mean). Results further showed that SOC stock correlated with general biome coverage, as the Grassland and Savanna biomes contributed the most (32.38 % and 31.28 %) to the overall SOC pool in South Africa. However, despite their smaller footprint, Forests (44.12 t/h) and the Indian Ocean Coastal Belt (43.05 t/h) biomes demonstrated the highest SOC sequestration capacity. The restoration of degraded biomes is advocated for, in order to boost SOC storage; but a balance between carbon sequestration capacity, biodiversity health, and the adequate provision of ecosystem services must be maintained. To this end, these findings provide a guideline to facilitate sustainable SOC stock management within South Africa's major biomes and indeed other regions of the world.

Modelling land use/land cover (LULC) change dynamics, future prospects, and its environmental impacts based on geospatial data models and remote sensing data

The dynamic change in land use/land cover (LULC) caused by rapid urbanization has become a major concern in Lahore, causing a variety of socioeconomic and environmental issues relating to air quality. As a result, it is important to monitor existing LULC change detection, future projection, and the increase in atmospheric pollutants in Lahore. This research work makes use of Landsat, GIOVANNI, SRTM DEM, and vector data. The four key steps of the research approach are as follows: (i) LULC classification from 2000 to 2020 using Lansat data and semi-automatic classification plugin (SCP); (ii) simulation of future prediction using Modules of Land Use Change Evaluation (MOLUSCE) prediction model; (iii) assessment of effects of land use change on air quality using GIOVANNI-NASA product; and (iv) monitoring, change detection, and result interpretation. According to the research findings, there was an increase in metropolitan areas and a decrease in vegetation, barren land, and water bodies for both historical and future projections. The findings also indicated that from 2000 to 2020, about 27.41% of the urban area expanded, with a decline of 42.13% in vegetation, 2.3% in the barren land, and 6.51% in water bodies, respectively. Between 2020 and 2040, the urban area is predicted to increase by 23.15%, while vegetation, barren land, and water bodies are expected to decrease by 28.05%, 1.8%, and 12.31%, respectively. Also, the atmospheric pollutants have been increased including NO2 (1.60%), SO2 (1.02%), CO2 (0.71%), CO (1.56%), O3 (0.15%), and CH4 (0.20%), respectively. And it is projected that by 2040, the average annual atmospheric concentration of NO2, SO2, CO2, CO, O3, and CH4 will be increased by 28.80%, 18.36%, 12.78%, 28.08%, 2.70%, and 3.60%, respectively. In addition, it was also observed that the majority of the city's urban area expansion was found in the city's eastern and southern regions. Therefore, government should focus on natural resource conservation especially vegetation cover to reduce air pollutants concentration and the LULC effect.

Multi-Dataset Hyper-CNN for Hyperspectral Image Segmentation of Remote Sensing Images

This research paper presents novel condensed CNN architecture for the recognition of multispectral images, which has been developed to address the lack of attention paid to neural network designs for multispectral and hyperspectral photography in comparison to RGB photographs. The proposed architecture is able to recognize 10-band multispectral images and has fewer parameters than popular deep designs, such as ResNet and DenseNet, thanks to recent advancements in more efficient smaller CNNs. The proposed architecture is trained from scratch, and it outperforms a comparable network that was trained on RGB images in terms of accuracy and efficiency. The study also demonstrates the use of a Bayesian variant of CNN architecture to show that a network able to process multispectral information greatly reduces the uncertainty associated with class predictions in comparison to standard RGB images. The results of the study are demonstrated by comparing the accuracy of the network’s predictions to the images.

Within and combined season prediction models for perennial ryegrass biomass yield using ground- and air-based sensor data

Across-season biomass assessment is crucial in the cultivar selection process to accurately evaluate the yield performance of lines under different growing conditions. However, it has been difficult to have an accurate, reliable, and repeated fresh biomass (FM) estimation of large populations of plants in the field without destructive harvesting, which incurs significant labor and operation costs. Sensor-based phenotyping platforms have advanced in the data collection of structural and vegetative information of plants, but the developed prediction models are still limited by low correlations at different growth stages and seasons. In this study, our objective was to develop and validate the global prediction models for across-season harvested fresh biomass (FM) yield based on the ground- and air-based sensor data including ground-based LiDAR, ground-based ultrasonic, and air-based multispectral camera to extract LiDAR plant volume (LV), LiDAR point density (LV_Den), height, and Normalized Difference Vegetative Index (NDVI). The study was conducted in a row-plot field trial with 480 rows (3 rows in a plot per cultivar) throughout the whole 2020 growing season up to the reproductive stage. We evaluated the performance of each plant parameter, their relationship, and the best subset prediction models using statistical stepwise selection at the row and plot levels through the seasonal and combined seasonal datasets. The best performing model: F M ~ L V ∗ L V _ D e n ∗ N D V I had a determination of coefficient R ² of at least 0.9 in vegetative stages and 0.8 in the reproductive stage. Similar results can be achieved in a simpler model with just two LiDAR variables- F M ~ L V ∗ L V _ D e n . In addition, LV and LV_Den showed a robust correlation with FM on their own over seasons and growth stages, while NDVI only performed well in some seasons. The simpler model based on only LiDAR data can be widely applied over season without the need of additional sensor data and may thus make the in-field across-season biomass assessment more feasible and practical for fast and cost-effective development of higher biomass yield cultivars.

Spatial Scale Effect and Correction of Forest Aboveground Biomass Estimation Using Remote Sensing

Forest biomass is critically important for forest dynamics in the carbon cycle. However, large-scale AGB mapping applications from remote sensing data still carry large uncertainty. In this study, an AGB estimation model was first established with three different remote sensing datasets of GF-2, Sentinel-2 and Landsat-8. Next, the optimal scale estimation result was considered as a reference AGB to obtain the relative true AGB distribution at different scales based on the law of conservation of mass, and the error of the scale effect of AGB estimation at various spatial resolutions was analyzed. Then, the information entropy of land use type was calculated to identify the heterogeneity of pixels. Finally, a scale conversion method for the entropy-weighted index was developed to correct the scale error of the estimated AGB results from coarse-resolution remote sensing images. The results showed that the random forest model had better prediction accuracy for GF-2 (4 m), Sentinel-2 (10 m) and Landsat-8 (30 m) AGB mapping. The determination coefficient between predicted and measured AGB was 0.5711, 0.4819 and 0.4321, respectively. Compared to uncorrected AGB, R2 between scale-corrected results and relative true AGB increased from 0.6226 to 0.6725 for Sentinel-2, and increased from 0.5910 to 0.6704 for Landsat-8. The scale error was effectively corrected. This study can provide a reference for forest AGB estimation and scale error reduction for AGB production upscaling with consideration of the spatial heterogeneity of the forest surface.

Integrating Multi-Source Remote Sensing to Assess Forest Aboveground Biomass in the Khingan Mountains of North-Eastern China Using Machine-Learning Algorithms

Forest aboveground biomass (AGB) is of great significance since it represents large carbon storage and may reduce global climate change. However, there are still considerable uncertainties in forest AGB estimates, especially in rugged regions, due to the lack of effective algorithms to remove the effects of topography and the lack of comprehensive comparisons of methods used for estimation. Here, we systematically compare the performance of three sources of remote sensing data used in forest AGB estimation, along with three machine-learning algorithms using extensive field measurements (N = 1058) made in the Khingan Mountains of north-eastern China in 2008. The datasets used were obtained from the LiDAR-based Geoscience Laser Altimeter System onboard the Ice, Cloud, and land Elevation satellite (ICESat/GLAS), the optical-based Moderate Resolution Imaging Spectroradiometer (MODIS), and the SAR-based Advanced Land Observing Satellite (ALOS) Phased Array type L-band Synthetic Aperture Radar (PALSAR). We show that terrain correction is effective for this mountainous study region and that the combination of terrain-corrected GLAS and PALSAR features with Random Forest regression produces the best results at the plot scale. Including further MODIS-based features added little power for prediction. Based upon the parsimonious data source combination, we created a map of AGB circa 2008 and its uncertainty, which yields a coefficient of determination (R2) of 0.82 and a root mean squared error of 16.84 Mg ha−1 when validated with field data. Forest AGB values in our study area were within the range 79.81 ± 16.00 Mg ha−1, ~25% larger than a previous, SAR-based, analysis. Our result provides a historic benchmark for regional carbon budget estimation.

Estimating Forest Aboveground Biomass Using Gaofen-1 Images, Sentinel-1 Images, and Machine Learning Algorithms: A Case Study of the Dabie Mountain Region, China

Quantitatively mapping forest aboveground biomass (AGB) is of great significance for the study of terrestrial carbon storage and global carbon cycles, and remote sensing-based data are a valuable source of estimating forest AGB. In this study, we evaluated the potential of machine learning algorithms (MLAs) by integrating Gaofen-1 (GF1) images, Sentinel-1 (S1) images, and topographic data for AGB estimation in the Dabie Mountain region, China. Variables extracted from GF1 and S1 images and digital elevation model data from sample plots were used to explain the field AGB value variations. The prediction capability of stepwise multiple regression and three MLAs, i.e., support vector machine (SVM), random forest (RF), and backpropagation neural network were compared. The results showed that the RF model achieved the highest prediction accuracy (R2 = 0.70, RMSE = 16.26 t/ha), followed by the SVM model (R2 = 0.66, RMSE = 18.03 t/ha) for the testing datasets. Some variables extracted from the GF1 images (e.g., normalized differential vegetation index, band 1-blue, the mean texture feature of band 3-red with windows of 3 × 3), S1 images (e.g., vertical transmit-horizontal receive and vertical transmit-vertical receive backscatter coefficient), and altitude had strong correlations with field AGB values (p < 0.01). Among the explanatory variables in MLAs, variables extracted from GF1 made a greater contribution to estimating forest AGB than those derived from S1 images. These results indicate the potential of the RF model for evaluating forest AGB by combining GF1 and S1, and that it could provide a reference for biomass estimation using multi-source images.

The Development and Application of Machine Learning in Atmospheric Environment Studies

Machine learning (ML) plays an important role in atmospheric environment prediction, having been widely applied in atmospheric science with significant progress in algorithms and hardware. In this paper, we present a brief overview of the development of ML models as well as their application to atmospheric environment studies. ML model performance is then compared based on the main air pollutants (i.e., PM2.5, O3, and NO2) and model type. Moreover, we identify the key driving variables for ML models in predicting particulate matter (PM) pollutants by quantitative statistics. Additionally, a case study for wet nitrogen deposition estimation is carried out based on ML models. Finally, the prospects of ML for atmospheric prediction are discussed.

The Effect of Synergistic Approaches of Features and Ensemble Learning Algorith on Aboveground Biomass Estimation of Natural Secondary Forests Based on ALS and Landsat 8

Although the combination of Airborne Laser Scanning (ALS) data and optical imagery and machine learning algorithms were proved to improve the estimation of aboveground biomass (AGB), the synergistic approaches of different data and ensemble learning algorithms have not been fully investigated, especially for natural secondary forests (NSFs) with complex structures. This study aimed to explore the effects of the two factors on AGB estimation of NSFs based on ALS data and Landsat 8 imagery. The synergistic method of extracting novel features (i.e., COLI1 and COLI2) using optimal Landsat 8 features and the best-performing ALS feature (i.e., elevation mean) yielded higher accuracy of AGB estimation than either optical-only or ALS-only features. However, both of them failed to improve the accuracy compared to the simple combination of the untransformed features that generated them. The convolutional neural networks (CNN) model was much superior to other classic machine learning algorithms no matter of features. The stacked generalization (SG) algorithms, a kind of ensemble learning algorithms, greatly improved the accuracies compared to the corresponding base model, and the SG with the CNN meta-model performed best. This study provides technical support for a wall-to-wall AGB mapping of NSFs of northeastern China using efficient features and algorithms.

Valuable Clues for DCNN-Based Landslide Detection from a Comparative Assessment in the Wenchuan Earthquake Area

Landslide inventories could provide fundamental data for analyzing the causative factors and deformation mechanisms of landslide events. Considering that it is still hard to detect landslides automatically from remote sensing images, endeavors have been carried out to explore the potential of DCNNs on landslide detection, and obtained better performance than shallow machine learning methods. However, there is often confusion as to which structure, layer number, and sample size are better for a project. To fill this gap, this study conducted a comparative test on typical models for landside detection in the Wenchuan earthquake area, where about 200,000 secondary landslides were available. Multiple structures and layer numbers, including VGG16, VGG19, ResNet50, ResNet101, DenseNet120, DenseNet201, UNet−, UNet+, and ResUNet were investigated with different sample numbers (100, 1000, and 10,000). Results indicate that VGG models have the highest precision (about 0.9) but the lowest recall (below 0.76); ResNet models display the lowest precision (below 0.86) and a high recall (about 0.85); DenseNet models obtain moderate precision (below 0.88) and recall (about 0.8); while UNet+ also achieves moderate precision (0.8) and recall (0.84). Generally, a larger sample set can lead to better performance for VGG, ResNet, and DenseNet, and deeper layers could improve the detection results for ResNet and DenseNet. This study provides valuable clues for designing models’ type, layers, and sample set, based on tests with a large number of samples.

Deep Neural Networks with Transfer Learning for Forest Variable Estimation Using Sentinel-2 Imagery in Boreal Forest

Estimation of forest structural variables is essential to provide relevant insights for public and private stakeholders in forestry and environmental sectors. Airborne light detection and ranging (LiDAR) enables accurate forest inventory, but it is expensive for large area analyses. Continuously increasing volume of open Earth Observation (EO) imagery from high-resolution (<30 m) satellites together with modern machine learning algorithms provide new prospects for spaceborne large area forest inventory. In this study, we investigated the capability of Sentinel-2 (S2) image and metadata, topography data, and canopy height model (CHM), as well as their combinations, to predict growing stock volume with deep neural networks (DNN) in four forestry districts in Central Finland. We focused on investigating the relevance of different input features, the effect of DNN depth, the amount of training data, and the size of image data sampling window to model prediction performance. We also studied model transfer between different silvicultural districts in Finland, with the objective to minimize the amount of new field data needed. We used forest inventory data provided by the Finnish Forest Centre for model training and performance evaluation. Leaving out CHM features, the model using RGB and NIR bands, the imaging and sun angles, and topography features as additional predictive variables obtained the best plot level accuracy (RMSE% = 42.6%, |BIAS%| = 0.8%). We found 3×3 pixels to be the optimal size for the sampling window, and two to three hidden layer DNNs to produce the best results with relatively small improvement to single hidden layer networks. Including CHM features with S2 data and additional features led to reduced relative RMSE (RMSE% = 28.6–30.7%) but increased the absolute value of relative bias (|BIAS%| = 0.9–4.0%). Transfer learning was found to be beneficial mainly with training data sets containing less than 250 field plots. The performance differences of DNN and random forest models were marginal. Our results contribute to improved structural variable estimation performance in boreal forests with the proposed image sampling and input feature concept.

Optimizing the Empirical Parameters of the Data-Driven Algorithm for SIF Retrieval for SIFIS Onboard TECIS-1 Satellite

Space-based solar-induced chlorophyll fluorescence (SIF) has been widely demonstrated as a great proxy for monitoring terrestrial photosynthesis and has been successfully retrieved from satellite-based hyperspectral observations using a data-driven algorithm. As a semi-empirical algorithm, the data-driven algorithm is strongly affected by the empirical parameters in the model. Here, the influence of the data-driven algorithm's empirical parameters, including the polynomial order (np), the number of feature vectors (nSV), the fluorescence emission spectrum function, and the fitting window used in the retrieval model, were quantitatively investigated based on the simulations of the SIF Imaging Spectrometer (SIFIS) onboard the First Terrestrial Ecosystem Carbon Inventory Satellite (TECIS-1). The results showed that the fitting window, np, and nSV were the three main factors that influenced the accuracy of retrieval. The retrieval accuracy was relatively higher for a wider fitting window; the root mean square error (RMSE) was lower than 0.7 mW m-2 sr-1 nm-1 with fitting windows wider than 735-758 nm and 682-691 nm for the far-red band and the red band, respectively. The RMSE decreased first and then increased with increases in np range from 1 to 5 and increased in nSV range from 2 to 20. According to the specifications of SIFIS onboard TECIS-1, a fitting window of 735-758 nm, a second-order polynomial, and four feature vectors are the optimal parameters for far-red SIF retrieval, resulting in an RMSE of 0.63 mW m-2 sr-1 nm-1. As for red SIF retrieval, using second-order polynomial and seven feature vectors in the fitting window of 682-697 nm was the optimal choice and resulted in an RMSE of 0.53 mW m-2 sr-1 nm-1. The optimized parameters of the data-driven algorithm can guide the retrieval of satellite-based SIF and are valuable for generating an accurate SIF product of the TECIS-1 satellite after its launch.

Applications of UAS in Crop Biomass Monitoring: A Review

Biomass is an important indicator for evaluating crops. The rapid, accurate and nondestructive monitoring of biomass is the key to smart agriculture and precision agriculture. Traditional detection methods are based on destructive measurements. Although satellite remote sensing, manned airborne equipment, and vehicle-mounted equipment can nondestructively collect measurements, they are limited by low accuracy, poor flexibility, and high cost. As nondestructive remote sensing equipment with high precision, high flexibility, and low-cost, unmanned aerial systems (UAS) have been widely used to monitor crop biomass. In this review, UAS platforms and sensors, biomass indices, and data analysis methods are presented. The improvements of UAS in monitoring crop biomass in recent years are introduced, and multisensor fusion, multi-index fusion, the consideration of features not directly related to monitoring biomass, the adoption of advanced algorithms and the use of low-cost sensors are reviewed to highlight the potential for monitoring crop biomass with UAS. Considering the progress made to solve this type of problem, we also suggest some directions for future research. Furthermore, it is expected that the challenge of UAS promotion will be overcome in the future, which is conducive to the realization of smart agriculture and precision agriculture.

Estimating Pasture Biomass Using Sentinel-2 Imagery and Machine Learning

Effective dairy farm management requires the regular estimation and prediction of pasture biomass. This study explored the suitability of high spatio-temporal resolution Sentinel-2 imagery and the applicability of advanced machine learning techniques for estimating aboveground biomass at the paddock level in five dairy farms across northern Tasmania, Australia. A sequential neural network model was developed by integrating Sentinel-2 time-series data, weekly field biomass observations and daily climate variables from 2017 to 2018. Linear least-squares regression was employed for evaluating the results for model calibration and validation. Optimal model performance was realised with an R2 of ≈0.6, a root-mean-square error (RMSE) of ≈356 kg dry matter (DM)/ha and a mean absolute error (MAE) of 262 kg DM/ha. These performance markers indicated the results were within the variability of the pasture biomass measured in the field, and therefore represent a relatively high prediction accuracy. Sensitivity analysis further revealed what impact each farm’s in situ measurement, pasture management and grazing practices have on the model’s predictions. The study demonstrated the potential benefits and feasibility of improving biomass estimation in a cheap and rapid manner over traditional field measurement and commonly used remote-sensing methods. The proposed approach will help farmers and policymakers to estimate the amount of pasture present for optimising grazing management and improving decision-making regarding dairy farming.

Near-real time aboveground carbon emissions in Peru

Monitoring aboveground carbon stocks and fluxes from tropical deforestation and forest degradation is important for mitigating climate change and improving forest management. However, high temporal and spatial resolution analyses are rare. This study presents the most detailed tracking of aboveground carbon over time, with yearly, quarterly and monthly estimations of emissions using the stock-difference approach and masked by the forest loss layer of Global Forest Watch. We generated high spatial resolution (1-ha) monitoring of aboveground carbon density (ACD) and emissions (ACE) in Peru by incorporating hundreds of thousands of Planet Dove satellite images, Sentinel-1 radar, topography and airborne LiDAR, embedded into a deep learning regression workflow using high-performance computing. Consistent ACD results were obtained for all quarters and months analyzed, with R² values of 0.75–0.78, and root mean square errors (RMSE) between 20.6 and 22.0 Mg C ha^-1. A total of 7.138 Pg C was estimated for Peru with annual ACE of 20.08 Tg C between the third quarters of 2017 and 2018, respectively, or 23.4% higher than estimates from the FAO Global Forest Resources Assessment. Analyzed quarterly, the spatial evolution of ACE revealed 11.5 Tg C, 6.6 Tg C, 8.6 Tg C, and 10.1 Tg C lost between the third quarters of 2017 and 2018. Moreover, our monthly analysis for the dry season reveals the evolution of ACE at unprecedented temporal detail. We discuss environmental controls over ACE and provide a spatially explicit tool for enhanced forest carbon management at scale.

Point2SpatialCapsule: Aggregating Features and Spatial Relationships of Local Regions on Point Clouds using Spatial-aware Capsules

Learning discriminative shape representation directly on point clouds is still challenging in 3D shape analysis and understanding. Recent studies usually involve three steps: first splitting a point cloud into some local regions, then extracting the corresponding feature of each local region, and finally aggregating all individual local region features into a global feature as shape representation using simple max-pooling. However, such pooling-based feature aggregation methods do not adequately take the spatial relationships (e.g. the relative locations to other regions) between local regions into account, which greatly limits the ability to learn discriminative shape representation. To address this issue, we propose a novel deep learning network, named Point2SpatialCapsule, for aggregating features and spatial relationships of local regions on point clouds, which aims to learn more discriminative shape representation. Compared with the traditional max-pooling based feature aggregation networks, Point2SpatialCapsule can explicitly learn not only geometric features of local regions but also the spatial relationships among them. Point2SpatialCapsule consists of two main modules. To resolve the disorder problem of local regions, the first module, named geometric feature aggregation, is designed to aggregate the local region features into the learnable cluster centers, which explicitly encodes the spatial locations from the original 3D space. The second module, named spatial relationship aggregation, is proposed for further aggregating the clustered features and the spatial relationships among them in the feature space using the spatial-aware capsules developed in this paper. Compared to the previous capsule network based methods, the feature routing on the spatial-aware capsules can learn more discriminative spatial relationships among local regions for point clouds, which establishes a direct mapping between log priors and the spatial locations through feature clusters. Experimental results demonstrate that Point2SpatialCapsule outperforms the state-of-the-art methods in the 3D shape classification, retrieval and segmentation tasks under the well-known ModelNet and ShapeNet datasets.

Estimation of Canopy Biomass Components in Paddy Rice from Combined Optical and SAR Data Using Multi-Target Gaussian Regressor Stacking

Crop biomass is a critical variable to make sound decisions about field crop monitoring activities (fertilizers and irrigation) and crop productivity forecasts. More importantly, crop biomass estimations by components are essential for crop growth monitoring as the yield formation of crops results from the accumulation and transportation of substances between different organs. Retrieval of crop biomass from synthetic aperture radar SAR or optical imagery is of paramount importance for in-season monitoring of crop growth. A combination of optical and SAR imagery can compensate for their limitations and has exhibited comparative advantages in biomass estimation. Notably, the joint estimations of biophysical parameters might be more accurate than that of an individual parameter. Previous studies have attempted to use satellite imagery to estimate aboveground biomass, but the estimation of biomass for individual organs remains a challenge. Multi-target Gaussian process regressor stacking (MGPRS), as a new machine learning method, can be suitably utilized to estimate biomass components jointly from satellite imagery data, as the model does not require a large amount of data for training and can be adjusted to the required degrees of relationship exhibited by the given data. Thus, the aim of this study was to estimate the biomass of individual organs by using MGPRS in conjunction with optical (Sentinel-2A) and SAR (Sentinel-1A) imagery. Two hybrid indices, SAR and optical multiplication vegetation index (SOMVI) and SAR and optical difference vegetation index (SODVI), have been constructed to examine their estimation performance. The hybrid vegetation indices were used as input for the MGPRS and single-target Gaussian process regression (SGPR). The accuracy of the estimation methods was analyzed by in situ measurements of aboveground biomass (AGB) and organ biomass conducted in 2018 and 2019 over the paddy rice fields of Xinghua in Jiangsu Province, China. The results showed that the combined indices (SOMVI and SODVI) performed better than those derived from either the optical or SAR data only. The best predictive accuracy was achieved by the MGPRS using SODVI as input (r2 = 0.84, RMSE = 0.4 kg/m2 for stem biomass; r2 = 0.87, RMSE = 0.16 kg/m2 for AGB). This was higher than using SOMVI as input for the MGPRS (r2 = 0.71, RMSE = 1.12 kg/m2 for stem biomass; r2 = 0.71, RMSE = 0.56 kg/m2 for AGB) or SGPR (r2 = 0.63, RMSE = 1.08 kg/m2 for stem biomass; r2 = 0.67, RMSE = 1.08 kg/m2 for AGB). Relatively, higher accuracy for leaf biomass was achieved using SOMVI (r2 = 0.83) than using SODVI (r2 = 0.73) as input for MGPRS. Our results demonstrate that the combined indices are effective by integrating SAR and optical imagery and MGPRS outperformed SGPR with the same input variable for estimating rice crop biomass. The presented workflow will improve the estimation of crops biomass components from satellite data for effective crop growth monitoring.

Crop Monitoring Using Satellite/UAV Data Fusion and Machine Learning

Non-destructive crop monitoring over large areas with high efficiency is of great significance in precision agriculture and plant phenotyping, as well as decision making with regards to grain policy and food security. The goal of this research was to assess the potential of combining canopy spectral information with canopy structure features for crop monitoring using satellite/unmanned aerial vehicle (UAV) data fusion and machine learning. Worldview-2/3 satellite data were tasked synchronized with high-resolution RGB image collection using an inexpensive unmanned aerial vehicle (UAV) at a heterogeneous soybean (Glycine max (L.) Merr.) field. Canopy spectral information (i.e., vegetation indices) was extracted from Worldview-2/3 data, and canopy structure information (i.e., canopy height and canopy cover) was derived from UAV RGB imagery. Canopy spectral and structure information and their combination were used to predict soybean leaf area index (LAI), aboveground biomass (AGB), and leaf nitrogen concentration (N) using partial least squares regression (PLSR), random forest regression (RFR), support vector regression (SVR), and extreme learning regression (ELR) with a newly proposed activation function. The results revealed that: (1) UAV imagery-derived high-resolution and detailed canopy structure features, canopy height, and canopy coverage were significant indicators for crop growth monitoring, (2) integration of satellite imagery-based rich canopy spectral information with UAV-derived canopy structural features using machine learning improved soybean AGB, LAI, and leaf N estimation on using satellite or UAV data alone, (3) adding canopy structure information to spectral features reduced background soil effect and asymptotic saturation issue to some extent and led to better model performance, (4) the ELR model with the newly proposed activated function slightly outperformed PLSR, RFR, and SVR in the prediction of AGB and LAI, while RFR provided the best result for N estimation. This study introduced opportunities and limitations of satellite/UAV data fusion using machine learning in the context of crop monitoring.

Application of Convolutional Neural Network on Lei Bamboo Above-Ground-Biomass (AGB) Estimation Using Worldview-2

Above-ground biomass (AGB) directly relates to the productivity of forests. Precisely, AGB mapping for regional forests based on very high resolution (VHR) imagery is widely needed for evaluation of productivity. However, the diversity of variables and algorithms and the difficulties inherent in high resolution optical imagery make it complex. In this paper, we explored the potentials of the state-of-art algorithm convolutional neural networks (CNNs), which are widely used for its high-level representation, but rarely applied for AGB estimation. Four experiments were carried out to compare the performance of CNNs and other state-of-art Machine Learning (ML) algorithms: (1) performance of CNN using bands, (2) performance of Random Forest (RF), support vector regression (SVR), artificial neural network (ANN) on bands, and vegetation indices (VIs). (3) Performance of RF, SVR, and ANN on gray-level co-occurrence matrices (GLCM), and exploratory spatial data analysis (ESDA), and (4) performance of RF, SVR, and ANN based on all combined data and ESDA+VIs. CNNs reached satisfactory results (with R2 = 0.943) even with limited input variables (i.e., only bands). In comparison, RF and SVR with elaborately designed data obtained slightly better accuracy than CNN. For examples, RF based on GLCM textures reached an R2 of 0.979 and RF based on all combined data reached a close R2 of 0.974. However, the results of ANN were much worse (with the best R2 of 0.885).

Inland Lakes Mapping for Monitoring Water Quality Using a Detail/Smoothing-Balanced Conditional Random Field Based on Landsat-8/Levels Data

The sustainable development of water resources is always emphasized in China, and a set of perfect standards for the division of inland water environment quality have been established to monitor water quality. However, most of the 24 indicators that determine the water quality level in the standards are non-optically active parameters. The weak optical characteristics make it difficult to find significant correlations between the single parameters and the remote sensing imagery. In addition, traditional on-site testing methods have been unable to meet the increasingly extensive water-quality monitoring requirements. Based on the above questions, it’s meaningful that the supervised classification process of a detail-preserving smoothing classifier based on conditional random field (CRF) and Landsat-8 data was proposed in the two study areas around Wuhan and Huangshi in Hubei Province. The random forest classifier was selected to model the association potential of the CRF. The results (the first study area: OA = 89.50%, Kappa = 0.841; the second study area: OA = 90.35%, Kappa = 0.868) showed that the water-quality monitoring based on CRF model is feasible, and this approach can provide a reference for water-quality mapping of inland lakes. In the future, it may only require a small amount of on-site sampling to achieve the identification of the water quality levels of inland lakes across a large area of China.

Analysis and prediction of vegetation dynamics under the background of climate change in Xinjiang, China

Background

Vegetation dynamics is defined as a significant indictor in regulating terrestrial carbon balance and climate change, and this issue is important for the evaluation of climate change. Though much work has been done concerning the correlations among vegetation dynamics, precipitation and temperature, the related questions about relationships between vegetation dynamics and other climatic factors (e.g., specific humidity, net radiation, soil moisture) have not been thoroughly considered. Understanding these questions is of primary importance in developing policies to address climate change.

Methods

In this study, the least squares regression analysis method was used to simulate the trend of vegetation dynamics based on the normalized difference vegetation index (NDVI) from 1981 to 2018. A partial correlation analysis method was used to explore the relationship between vegetation dynamics and climate change; and further,the revised greyscale model was applied to predict the future growth trend of natural vegetation.

Results

The Mann-Kendall test results showed that th e air temperature rose sharply in 1997 and had been in a state of high fluctuations since then. Strong changes in hydrothermal conditions had major impact on vegetation dynamics in the area. Specifically, the NDVI value of natural vegetation showed an increasing trend from 1981 to 2018, and the same changes occurred in the precipitation. From 1981 to 1997, the values of natural vegetation increased at a rate of 0.0016 per year. From 1999 to 2009, the NDVI value decreased by an average rate of 0.0025 per year. From 2010 to 2018, the values began an increasing trend and reached a peak in 2017, with an average annual rate of 0.0033. The high vegetation dynamics areas were mainly concentrated in the north and south slopes of the Tianshan Mountains, the Ili River Valley and the Altay area. The greyscale prediction results showed that the annual average NDVI values of natural vegetation may present a fluctuating increasing trend. The NDVI value in 2030 is 0.0196 higher than that in 2018, with an increase of 6.18%.

Conclusions

Our results indicate that: (i) the variations of climatic factors have caused a huge change in the hydrothermal conditions in Xinjiang; (ii) the vegetation dynamics in Xinjiang showed obvious volatility, and then in the end stage of the study were higher than the initial stage the vegetation dynamics in Xinjiang showed a staged increasing trend; (iii) the vegetation dynamics were affected by many factors,of which precipitation was the main reason; (iv) in the next decade, the vegetation dynamics in Xinjiang will show an increasing trend.

Estimating Forest Stock Volume in Hunan Province, China, by Integrating In Situ Plot Data, Sentinel-2 Images, and Linear and Machine Learning Regression Models

The forest stock volume (FSV) is one of the key indicators in forestry resource assessments on local, regional, and national scales. To date, scaling up in situ plot-scale measurements across landscapes is still a great challenge in the estimation of FSVs. In this study, Sentinel-2 imagery, the Google Earth Engine (GEE) cloud computing platform, three base station joint differential positioning technology (TBSJDPT), and three algorithms were used to build an FSV model for forests located in Hunan Province, southern China. The GEE cloud computing platform was used to extract the imagery variables from the Sentinel-2 imagery pixels. The TBSJDPT was put forward and used to provide high-precision positions of the sample plot data. The random forests (RF), support vector regression (SVR), and multiple linear regression (MLR) algorithms were used to estimate the FSV. For each pixel, 24 variables were extracted from the Sentinel-2 images taken in 2017 and 2018. The RF model performed the best in both the training phase (i.e., R2 = 0.91, RMSE = 35.13 m3 ha−1, n = 321) and in the test phase (i.e., R2 = 0.58, RMSE = 65.03 m3 ha−1, and n = 138). This model was followed by the SVR model (R2 = 0.54, RMSE = 65.60 m3 ha−1, n = 321 in training; R2 = 0.54, RMSE = 66.00 m3 ha−1, n = 138 in testing), which was slightly better than the MLR model (R2 = 0.38, RMSE = 75.74 m3 ha−1, and n = 321 in training; R2 = 0.49, RMSE = 70.22 m3 ha−1, and n = 138 in testing) in both the training phase and test phase. The best predictive band was Red-Edge 1 (B5), which performed well both in the machine learning methods and in the MLR method. The Blue band (B2), Green band (B3), Red band (B4), SWIR2 band (B12), and vegetation indices (TCW, NDVI_B5, and TCB) were used in the machine learning models, and only one vegetation index (MSI) was used in the MLR model. We mapped the FSV distribution in Hunan Province (3.50 × 108 m3) based on the RF model; it reached a total accuracy of 63.87% compared with the official forest report in 2017 (5.48 × 108 m3). The results from this study will help develop and improve satellite-based methods to estimate FSVs on local, regional and national scales.

Finer Resolution Estimation and Mapping of Mangrove Biomass Using UAV LiDAR and WorldView-2 Data

To estimate mangrove biomass at finer resolution, such as at an individual tree or clump level, there is a crucial need for elaborate management of mangrove forest in a local area. However, there are few studies estimating mangrove biomass at finer resolution partly due to the limitation of remote sensing data. Using WorldView-2 imagery, unmanned aerial vehicle (UAV) light detection and ranging (LiDAR) data, and field survey datasets, we proposed a novel method for the estimation of mangrove aboveground biomass (AGB) at individual tree level, i.e., individual tree-based inference method. The performance of the individual tree-based inference method was compared with the grid-based random forest model method, which directly links the field samples with the UAV LiDAR metrics. We discussed the feasibility of the individual tree-based inference method and the influence of diameter at breast height (DBH) on individual segmentation accuracy. The results indicated that (1) The overall classification accuracy of six mangrove species at individual tree level was 86.08%. (2) The position and number matching accuracies of individual tree segmentation were 87.43% and 51.11%, respectively. The number matching accuracy of individual tree segmentation was relatively satisfying within 8 cm ≤ DBH ≤ 30 cm. (3) The individual tree-based inference method produced lower accuracy than the grid-based RF model method with R2 of 0.49 vs. 0.67 and RMSE of 48.42 Mg ha–1 vs. 38.95 Mg ha–1. However, the individual tree-based inference method can show more detail of spatial distribution of mangrove AGB. The resultant AGB maps of this method are more beneficial to the fine and differentiated management of mangrove forests.

Mapping Height and Aboveground Biomass of Mangrove Forests on Hainan Island Using UAV-LiDAR Sampling

Hainan Island is the second-largest island in China and has the most species-diverse mangrove forests in the country. To date, the height and aboveground ground biomass (AGB) of the mangrove forests on Hainan Island are unknown, partly as a result of the challenges faced during extensive field sampling in mangrove habitats (intertidal mudflats inundated by periodic seawater). Therefore, this study used a low-cost UAV-LiDAR (light detection and ranging sensor mounted on an unmanned aerial vehicle) system as a sampling tool and Sentinel-2 imagery as auxiliary data to estimate and map the mangrove height and AGB on Hainan Island. Hainan Island has 3697.02 hectares of mangrove forests with an average patch area of approximately 1 ha. The results show that the mangroves on whole Hainan Island have an average height of 6.99 m, a total AGB of 474,199.31 Mg and an AGB density of 128.27 Mg ha−1. The AGB hot spots are located in Qinglan Harbor and the south of Dongzhai Harbor. The proposed height model LiDAR-S2 performed well with an R2 of 0.67 and an RMSE (root mean square error) of 1.90 m; the proposed AGB model G~LiDAR~S2 performed better (an R2 of 0.62 and an RMSE of 50.36 Mg ha−1) than the traditional AGB model G~S2 that directly related ground plots and Sentinel-2 data. The results also indicate that the LiDAR metrics describing the canopy’s thickness and its top and bottom characteristics are the most important variables for mangrove AGB estimation. For the Sentinel-2 indices, the red-edge and shortwave infrared features, especially the red-edge 1 and shortwave infrared Band 11 features, play the most important roles in estimating mangrove AGB and height. In conclusion, this paper presents the first mangrove height and AGB maps of Hainan Island and demonstrates the feasibility of using UAV-LiDAR as a sampling tool for mangrove forests.

Remote estimation of rice LAI based on Fourier spectrum texture from UAV image

BACKGROUND

The accurate estimation of rice LAI is particularly important to monitor rice growth status. Remote sensing, as a non-destructive measurement technology, has been proved to be useful for estimating vegetation growth parameters, especially at large scale. With the development of unmanned aerial vehicles (UAVs), this novel remote sensing platform has been widely used to provide remote sensing images which have much higher spatial resolution. Previous reports have shown that the spectral feature of remote sensing images could be an effective indicator to estimate vegetation growth parameters. However, the texture feature of high-resolution remote sensing images is rarely employed for this purpose. Besides, the physical mechanism between the texture feature and vegetation growth parameters is still unclear.

RESULTS

In this study, a Fourier spectrum texture based on the UAV Image was developed to estimate rice LAI. And the relationship between Fourier spectrum texture and rice LAI was also analyzed. The results showed that Fourier spectrum texture could improve the accuracy of rice LAI estimation.

CONCLUSIONS

In conclusion, the texture feature of high-resolution remote sensing images may be more effective in rice LAI estimation than the spectral feature.