High-accuracy Mapping Research Articles

LIDAR ASSESSMENTS AND MAPPING FOR KLANG VALLEY: A CASE STUDY AT JINJANG DISTRICT, SELANGOR

Abstract. Light Detection and Ranging (LiDAR) technology has become a significant factor in producing up-to-date and accurate topographic data in the current world. LiDAR technology has been used for years for many applications, including the efficient creation of digital model for large scale, high accuracy mapping. This technology offers fast, accurate, expedient and cost-effective ways of capturing wide area elevation information to produce highly detailed digital model of the earth. LiDAR is based on airborne laser scanners enables to acquire dense and accurate 3D data of the surveyed area, i.e., the Digital Surface Model (DSM). This paper presents an exploratory study to assess the accuracy of constructed DTM (Digital Terrain Model) and evaluating ground height without surface features using LiDAR Digital Surface Model (DSM). The study area comprised of an undulated area situated at Jinjang in the Klang Valley region, Malaysia covering an area of one kilometre square. LiDAR DSM and DTM constructed and derived from LiDAR were critically assessed with reference to the USGS Map Accuracy Standards. The accuracy of derived DSM and DTM were evaluated using ground control points derived from conventional surveying technique. The constructed models were accessed quantitatively and qualitatively.

Biomac3D: 2D-to-3D Human Pose Analysis Model for Tele-Rehabilitation Based on Pareto Optimized Deep-Learning Architecture

The research introduces a unique deep-learning-based technique for remote rehabilitative analysis of image-captured human movements and postures. We present a ploninomial Pareto-optimized deep-learning architecture for processing inverse kinematics for sorting out and rearranging human skeleton joints generated by RGB-based two-dimensional (2D) skeleton recognition algorithms, with the goal of producing a full 3D model as a final result. The suggested method extracts the entire humanoid character motion curve, which is then connected to a three-dimensional (3D) mesh for real-time preview. Our method maintains high joint mapping accuracy with smooth motion frames while ensuring anthropometric regularity, producing a mean average precision (mAP) of 0.950 for the task of predicting the joint position of a single subject. Furthermore, the suggested system, trained on the MoVi dataset, enables a seamless evaluation of posture in a 3D environment, allowing participants to be examined from numerous perspectives using a single recorded camera feed. The results of evaluation on our own self-collected dataset of human posture videos and cross-validation on the benchmark MPII and KIMORE datasets are presented.

Consistent, high-accuracy mapping of daily and sub-daily wildfire growth with satellite observations

Background Fire research and management applications, such as fire behaviour analysis and emissions modelling, require consistent, highly resolved spatiotemporal information on wildfire growth progression. Aims We developed a new fire mapping method that uses quality-assured sub-daily active fire/thermal anomaly satellite retrievals (2003–2020 MODIS and 2012–2020 VIIRS data) to develop a high-resolution wildfire growth dataset, including growth areas, perimeters, and cross-referenced fire information from agency reports. Methods Satellite fire detections were buffered using a historical pixel-to-fire size relationship, then grouped spatiotemporally into individual fire events. Sub-daily and daily growth areas and perimeters were calculated for each fire event. After assembly, fire event characteristics including location, size, and date, were merged with agency records to create a cross-referenced dataset. Key results Our satellite-based total fire size shows excellent agreement with agency records for MODIS (R2 = 0.95) and VIIRS (R2 = 0.97) in California. VIIRS-based estimates show improvement over MODIS for fires with areas less than 4047 ha (10 000 acres). To our knowledge, this is the finest resolution quality-assured fire growth dataset available. Conclusions and Implications The novel spatiotemporal resolution and methodological consistency of our dataset can enable advances in fire behaviour and fire weather research and model development efforts, smoke modelling, and near real-time fire monitoring.

Visible Band Index Optimation of Unmanned Aerial Vehicle for Estimating NDVI by Sentinel Imagery on Rice Vegetation

Serntinel 2A provide Normalized Difference Vegetation Index to be used as an estimate of soil fertility, plant varieties and productivity. The weakness of satellite data is that the data obtained is often inaccurate due to cloud cover, especially in tropical countries with high rainfall such as Indonesia. The use of unmanned aerial vehicle as an alternative data have limitation as it captured RGB imagery. The research was conducted from July to September 2020 at Pasir Kaliki Village, District of Rawamerta, Karawang Regency, West Java province. The study has discovered that NDVI showed higher number in result of vegetation index compared to NGRDI with correlation coefficient is 0.944625. The regression model resulted as y=4.7722x+0.3845 and MAPE value expresses as 26.74%, where the regression model with Pearson’s correlation coefficient value is 0.877885. A qualitative assessment using statistical data and a spatial assessment using sampled data from the rice vegetation map reveal a high mapping accuracy with the corresponding R2 being as high as 0.7429; however, the mapped rice vegetation accuracy might influenced by other physical factors such as water reflectant, sunlight and the RGB camera limitation itself. Nonetheless, the highest values of NGRDI only reach 0.2 while NDVI can attain at 0.9 at the peak of vegetative phase of rice growth stage. This means that Green Band have limitation in detecting vegetation index. In relation to the different approaches performed, it is noted that the average trend line on both NDVI and NGRDI shown the similarity tendency in all growth stage.

On the Buildings and Location of Armagh Observatory

We consider the buildings of Armagh Observatory and present their location to a few tens of centimetres or approximately 0.01 arcsecond in latitude and longitude, referred to the position of the Ordnance Survey (OS) benchmark (OSBM) inscribed on the historic eighteenth-century building. This has Irish Grid (IG) coordinates (287829.309BM, 345757.815BM) and a height above mean sea level of 61.08 m. It lies approximately 12.1 ± 0.15 m west and 7.9 ± 0.1 m south of the Observatory’s earliest transit telescopes and 1.5 ± 0.05 m east and 1.7 ± 0.1 m south of the 1795 Troughton telescope. Ground level at the Observatory has an orthometric height of 61.9 ± 0.1 m, approximately 2 metres below that given in the Astronomical Almanac. Accurate measurements of height have implications for values of atmospheric pressure reduced to mean sea level, for precise comparisons between weather stations, and for weather forecasting. Our position for the Observatory’s earliest transit telescopes lies within ~0.4 m of the coordinate origin of the 1835 OS 6-inch County Armagh maps, highlighting the link between astronomy and the first high-accuracy maps of Ireland. An 1834 plan and measurements of the surviving north meridian marks suggest that the transit room’s east windows were moved ∼0.8 m east when the sector tower was added around 1840. This resolves an unexplained discrepancy between the present distance (3.0 m) between the centres of the west and east windows of the transit room and the original distance (~2.2 m) inferred by other means. We suggest that we may havefound parts of early transit instruments believed lost and give improved coordinates for the meridian marks. This work has applications to the history of astronomy, geodesy, meteorology, and climate change.

3D Environment Mapping with a Variable Resolution NDT Method

With the continuous development of the 3D LiDAR (Light Detection And Ranging) mapping algorithm and its application in various fields, the size of the point cloud map becomes a bottleneck that limits the 3D LiDAR mapping algorithm from running for a long time. In this paper, a 3D LiDAR mapping method based on scan-to-map and variable resolution NDT (normal-distributions transform) registration is proposed. When updating the global map, variable resolution processing can reduce the size of the global map and improve the accuracy of map construction. In addition, the size of the map created by the mapping algorithm is proportional to the size of the space and does not grow infinitely over time. The mapping experiments using a rotating LiDAR in the room, corridor, and outdoor environments show that the algorithm has higher mapping accuracy and smaller map size than without considering the variable resolution strategy. The experimental results of the map construction for a long time in an appropriate test area illustrate that the map built by the algorithm does not grow infinitely with time in the fixed space. In summary, by adjusting the map resolution adaptively according to the curvature of different areas in the 3D LiDAR mapping process, the proposed variable resolution strategy can maintain the size of the global map almost proportional to the size of the space. Moreover, the mapping accuracy can be improved as well.

Satellite Laser Altimetry Data-Supported High-Accuracy Mapping of GF-7 Stereo Images

GaoFen 7 (GF-7) is China’s first submeter high-resolution stereo mapping satellite with dual-linear-array cameras and a laser altimeter system onboard for high-precision mapping. To further take advantage of the very high elevation accuracy of laser altimetry data and the high relative accuracy with stereo images, an innovative combined adjustment method for GF-7 stereo images with laser altimetry data is presented in this paper. In this method, two flexible and effective schemes were proposed to extract the elevation control point according to the registration of footprint images and stereo images and then utilized as vertical control in the block adjustment to improve the elevation accuracy without ground control points (GCPs). The validation experiments were conducted in Shandong, China, with different terrains. The results demonstrated that, after using the laser altimetry data, the root mean square error (RMSE) of elevation was dramatically improved from the original 2.15 m to 0.75 m, while the maximum elevation error was less than 1.6 m. Moreover, by integrating a few horizontal control points, the planar and elevation accuracy can be simultaneously improved. The results show that the method will be useful for reducing the need for field survey work and improving mapping efficiency.

An Intelligent Tracking System for Moving Objects in Dynamic Environments

Localization of suspicious moving objects in dynamic environments requires high accuracy mapping. A deep learning model is proposed to track crossing moving objects in the opposite direction. Moving objects locus measurements are computed from the space included in the boundaries of the images in the intersecting cameras. Object appearance is designated by the color and textural histograms in the intersecting camera views. The incorrect mapping of moving objects in a dynamic environment through synchronized localization can be considerably increased in complex areas. This is done due to the presence of unfit points that are triggered by moving targets. To face this problem, a robust model using the dynamic province rejection technique (DPR) is presented. We are proposing a novel model that incorporates a combination of the deep learning method and a tracking system that rejects dynamic areas which are not within the environment boundary of interest. The technique detects the dynamic points from sequential video images and partitions the current video image into super blocks and tags the border differences. In the last stage, dynamic areas are computed from dynamic points and superblock boundaries. Static regions are utilized to compute the positions to enhance the path computation precision of the model. Simulation results show that the introduced model has better performance than the state-of-the-art similar models in both the VID and MOVSD4 datasets and is higher than the state-of-the-art tracking systems with better speed performance. The experiments prove that the computed path error in the dynamic setting can be decreased by 81%.

Multi-objective optimization sampling based on Pareto optimality for soil mapping

Precise mapping based on sampling data is meaningful for efficient soil quality management. The distributions of sampling sites in feature space and geographical space, as well as the distribution of point pairs at different distances, all significantly impact mapping precision, and these three aspects should all be considered in a sampling design. Although optimizing the spread of sampling sites in the above three aspects has been realized and addressed in previous studies, their tradeoff relationship and influence on mapping accuracy have not been comprehensively investigated, partly due to the limitations of weighting-based optimization way. In this article, we proposed a sampling strategy based on Pareto optimality to examine the tradeoff relationship among the three aspects and their influence on interpolation precision. Based on soil organic matter data from Yi’an district, we applied this approach to generate sampling schemes and analyze their distributions and prediction errors. Single-objective optimization of these three aspects was also conducted for comparison. The results revealed that: (1) the sampling strategy that simultaneously optimizes the distributions of sampling sites in the three aspects was able to obtain sampling designs with higher mapping accuracy than the method that only optimizes one aspect. (2) There was an apparent synergistic relationship between the distributions of sampling sites in feature space and geographical space. However, these objectives had antagonistic relationships with the distribution of point pairs at different distances. (3) When considering the tradeoff among the three aspects, the more even the distribution in geographical space was, the higher the mapping accuracy they produced. Furthermore, more even distribution in feature space improved the mapping accuracy, but the benefit faded after a certain degree, whereas the mapping accuracy initially increased as the distance distribution of point pairs in different intervals became more even but subsequently declined beyond a certain point. Attention should be given to the tradeoff relationship and impact on interpolation when sampling for mapping. In similar scenarios like the study case, it is recommended that sampling sites should be distributed in feature space with more than 45% evenness of the optimal state and then distributed in geographical space as evenly as possible, while the distribution of point pairs should not exceed 65% evenness of its optimal state when the sample size range is approximately 50 to 150. This study provides helpful references for scientific sampling with the goal of precise mapping.

Multiscale Diagnosis of Mangrove Status in Data-Poor Context Using Very High Spatial Resolution Satellite Images: A Case Study in Pichavaram Mangrove Forest, Tamil Nadu, India

Highlighting spatiotemporal changes occurring within mangrove habitats at the finest possible scale could contribute fundamental knowledge and data for local sustainable management. This study presents the current situation of the Pichavaram mangrove area, a coastal region of Southeast India prone to both cyclones and reduced freshwater inflow. Based on the supervised classification and visual inspection of very high spatial resolution (VHSR) satellite images provided with a pixel size of <4 m, we generated time-series maps to analyze the changes that occurred in both the natural and planted mangroves between 2003 and 2019. We achieved a high mapping accuracy (>85%), which confirmed the potential of classification techniques applied to VHSR images in capturing changes in mangroves on a very fine scale. Our diagnosis reveals variable expansion rates in plantations made by the local authorities. We also report an ongoing mangrove dieback and confirm progressive shoreline erosion along the coastline. Despite a lack of field data, VHSR images allowed for the multiscale diagnosis of the ecosystem situation, thus constituting the first fine-scale assessment of the fragile Pichavaram mangrove area upon which the coastal community is dependent.

3D Distance Filter for the Autonomous Navigation of UAVs in Agricultural Scenarios

In precision agriculture, remote sensing is an essential phase in assessing crop status and variability when considering both the spatial and the temporal dimensions. To this aim, the use of unmanned aerial vehicles (UAVs) is growing in popularity, allowing for the autonomous performance of a variety of in-field tasks which are not limited to scouting or monitoring. To enable autonomous navigation, however, a crucial capability lies in accurately locating the vehicle within the surrounding environment. This task becomes challenging in agricultural scenarios where the crops and/or the adopted trellis systems can negatively affect GPS signal reception and localisation reliability. A viable solution to this problem can be the exploitation of high-accuracy 3D maps, which provide important data regarding crop morphology, as an additional input of the UAVs’ localisation system. However, the management of such big data may be difficult in real-time applications. In this paper, an innovative 3D sensor fusion approach is proposed, which combines the data provided by onboard proprioceptive (i.e., GPS and IMU) and exteroceptive (i.e., ultrasound) sensors with the information provided by a georeferenced 3D low-complexity map. In particular, the parallel-cuts ellipsoid method is used to merge the data from the distance sensors and the 3D map. Then, the improved estimation of the UAV location is fused with the data provided by the GPS and IMU sensors, using a Kalman-based filtering scheme. The simulation results prove the efficacy of the proposed navigation approach when applied to a quadrotor that autonomously navigates between vine rows.

Drone-Based Characterization of Seagrass Habitats in the Tropical Waters of Zanzibar

Unmanned automatic systems (UAS) are increasingly being applied as an alternative to more costly time-consuming traditional methods for mapping and monitoring marine shallow-water ecosystems. Here, we demonstrate the utility of combining aerial drones with in situ imagery to characterize the habitat conditions of nine shallow-water seagrass-dominated areas on Unguja Island, Zanzibar. We applied object-based image analysis and a maximum likelihood algorithm on the drone images to derive habitat cover maps and important seagrass habitat parameters: the habitat composition; the seagrass species; the horizontal- and depth-percent covers, and the seascape fragmentation. We mapped nine sites covering 724 ha, categorized into seagrasses (55%), bare sediment (31%), corals (9%), and macroalgae (5%). An average of six seagrass species were found, and 20% of the nine sites were categorized as “dense cover” (40–70%). We achieved high map accuracy for the habitat types (87%), seagrass (80%), and seagrass species (76%). In all nine sites, we observed clear decreases in the seagrass covers with depths ranging from 30% at 1–2 m, to 1.6% at a 4–5 m depth. The depth dependency varied significantly among the seagrass species. Areas associated with low seagrass cover also had a more fragmented distribution pattern, with scattered seagrass populations. The seagrass cover was correlated negatively (r2 = 0.9, p < 0.01) with sea urchins. A multivariate analysis of the similarity (ANOSIM) of the biotic features, derived from the drone and in situ data, suggested that the nine sites could be organized into three significantly different coastal habitat types. This study demonstrates the high robustness of drones for characterizing complex seagrass habitat conditions in tropical waters. We recommend adopting drones, combined with in situ photos, for establishing a suite of important data relevant for marine ecosystem monitoring in the Western Indian Ocean (WIO).

Phenotyping a diversity panel of quinoa using UAV-retrieved leaf area index, SPAD-based chlorophyll and a random forest approach

Given its high nutritional value and capacity to grow in harsh environments, quinoa has significant potential to address a range of food security concerns. Monitoring the development of phenotypic traits during field trials can provide insights into the varieties best suited to specific environmental conditions and management strategies. Unmanned aerial vehicles (UAVs) provide a promising means for phenotyping and offer the potential for new insights into relative plant performance. During a field trial exploring 141 quinoa accessions, a UAV-based multispectral camera was deployed to retrieve leaf area index (LAI) and SPAD-based chlorophyll across 378 control and 378 saline-irrigated plots using a random forest regression approach based on both individual spectral bands and 25 different vegetation indices (VIs) derived from the multispectral imagery. Results show that most VIs had stronger correlation with the LAI and SPAD-based chlorophyll measurements than individual bands. VIs including the red-edge band had high importance in SPAD-based chlorophyll predictions, while VIs including the near infrared band (but not the red-edge band) improved LAI prediction models. When applied to individual treatments (i.e. control or saline), the models trained using all data (i.e. both control and saline data) achieved high mapping accuracies for LAI (R2 = 0.977–0.980, RMSE = 0.119–0.167) and SPAD-based chlorophyll (R2 = 0.983–0.986, RMSE = 2.535–2.861). Overall, the study demonstrated that UAV-based remote sensing is not only useful for retrieving important phenotypic traits of quinoa, but that machine learning models trained on all available measurements can provide robust predictions for abiotic stress experiments.

LeGO-LOAM-SC: An Improved Simultaneous Localization and Mapping Method Fusing LeGO-LOAM and Scan Context for Underground Coalmine.

Simultaneous localization and mapping (SLAM) is one of the key technologies for coal mine underground operation vehicles to build complex environment maps and positioning and to realize unmanned and autonomous operation. Many domestic and foreign scholars have studied many SLAM algorithms, but the mapping accuracy and real-time performance still need to be further improved. This paper presents a SLAM algorithm integrating scan context and Light weight and Ground-Optimized LiDAR Odometry and Mapping (LeGO-LOAM), LeGO-LOAM-SC. The algorithm uses the global descriptor extracted by scan context for loop detection, adds pose constraints to Georgia Tech Smoothing and Mapping (GTSAM) by Iterative Closest Points (ICP) for graph optimization, and constructs point cloud map and an output estimated pose of the mobile vehicle. The test with KITTI dataset 00 sequence data and the actual test in 2-storey underground parking lots are carried out. The results show that the proposed improved algorithm makes up for the drift of the point cloud map, has a higher mapping accuracy, a better real-time performance, a lower resource occupancy, a higher coincidence between trajectory estimation and real trajectory, smoother loop, and 6% reduction in CPU occupancy, the mean square errors of absolute trajectory error (ATE) and relative pose error (RPE) are reduced by 55.7% and 50.3% respectively; the translation and rotation accuracy are improved by about 5%, and the time consumption is reduced by 2~4%. Accurate map construction and low drift pose estimation can be performed.

Building Instance Mapping From ALS Point Clouds Aided by Polygonal Maps

Building region extraction from ALS point clouds has been widely studied, whereas instance-level building mapping has been overlooked and remains unsolved. In this study, we present a method to extract individual buildings from ALS point clouds with the help of widely accessible polygonal footprints. The key idea is to merge roof segments to a set of building candidates, from which correct instances are selected by finding optimal matches between polygonal footprints and building candidates. The method has three steps: roof segmentation, building candidate generation, and instance-polygon matching. The method is tested on two large-scale scenes of different building types and can generally achieve high instance-level building mapping accuracy (around 90%) when there are large positioning errors (6.0 m) among polygons. Future work will focus on classification errors in preprocessing, shape inconsistency between point clouds and polygons, and building footprint delineation and updating in postprocessing.

Neural network guided interpolation for mapping canopy height of China's forests by integrating GEDI and ICESat-2 data

Spatially continuous estimates of forest canopy height at national to global scales are critical for quantifying forest carbon storage, understanding forest ecosystem processes, and developing forest management and restoration policies to mitigate global climate change. Spaceborne light detection and ranging (lidar) platforms, especially the Global Ecosystem Dynamics Investigation (GEDI) and Ice, Cloud, and land Elevation Satellite-2 (ICESat-2) Advanced Topographic Laser Altimeter System (ATLAS), can measure forest canopy height in discrete footprints globally. Their coverage provides a promising data source for national to global-scale forest canopy height estimates. However, previous studies usually used a regression-based approach to develop spatially continuous forest canopy height distribution through the aid of optical images, which cannot take full advantage of the dense spaceborne lidar footprints and may still suffer from the saturation effect of optical images. In this study, we developed a novel neural network guided interpolation (NNGI) method to map forest canopy height by fusing GEDI, ICESat-2 ATLAS, and Sentinel-2 images. To evaluate the performance of the proposed NNGI method, we generated a 30-m forest canopy height product of China for the year 2019. More than 140 km2 drone-lidar data were collected across the country to train and validate the NNGI method. The results showed that the average forest canopy height of China is 15.90 m with a standard deviation of 5.77 m. We evaluated the interpolated forest canopy height product of China by over 1,100,000 GEDI validation footprints (R2 = 0.55, RMSE = 5.32 m), about 33 km2 drone-lidar validation data (R2 = 0.58, RMSE = 4.93 m), and over 59,000 field plot measurements (R2 = 0.60, RMSE = 4.88 m). Benefiting from the interpolation-based mapping strategy, the resulting product had almost no saturation effect in areas with tall forest canopies. The high mapping accuracy demonstrates the feasibility of the proposed NNGI method for monitoring spatially continuous forest canopy height at national to global scales by integrating multi-platform spaceborne lidar data and optical images, enabling opportunities to provide more accurate quantification of terrestrial carbon storage and better understanding of forest ecosystem processes.

Harnessing Multi Source Point Cloud Technology to Overcome High Resolution Building Information Modeling of Manganti Dam, Indonesia: A Preliminary Result

Building Information Modelling (BIM) is a technology involving the generation and management of digital information of physical and functional characteristics of places (in 2D or 3D), which is developed to obtain built assets, planning, construction maintenance and operation, and construction management system. This case study is BIM development for the Manganti dam, located in Central Java, Indonesia. The dam itself is a vital irrigation system and is located in an earthquake-prone area, which makes significant risk factors for the surrounding environment. Therefore, continuous monitoring for the dam should be taken seriously into action by acquiring accurate information. Supporting that purpose, the spatial information of the dam, in 3D form, should be accurately presented in the mapping process. Although the Terrestrial Laser Scanner (TLS) is well known for providing accurate geometry information, its information detail is still limited to an accessible area. Consequently, it creates gaps in shadow areas which possibly can be compensated by the data from Handheld Laser Scanner (HLS). In this contribution, we are focusing on analyzing the reliability of TLS and HLS combination for high-resolution mapping and, possibly, high accuracy mapping. The geometric accuracy of the 3D model is comparing the dimension of the model to the dimension measurement by the Electronic Total Station (ETS) measurement.

OBIA-Based Extraction of Artificial Terrace Damages in the Loess Plateau of China from UAV Photogrammetry

Terraces, which are typical artificial landforms found around world, are of great importance for agricultural production and soil and water conservation. However, due to the lack of maintenance, terrace damages often occur and affect the local flow process, which will influence soil erosion. Automatic high-accuracy mapping of terrace damages is the basis of monitoring and related studies. Researchers have achieved artificial terrace damage mapping mainly via manual field investigation, but an automatic method is still lacking. In this study, given the success of high-resolution unmanned aerial vehicle (UAV) photogrammetry and object-based image analysis (OBIA) for image processing tasks, an integrated framework based on OBIA and UAV photogrammetry is proposed for terrace damage mapping. The Pujiawa terrace in the Loess Plateau of China was selected as the study area. Firstly, the segmentation process was optimised by considering the spectral features and the terrains and corresponding textures obtained from high-resolution images and digital surface models. The feature selection was implemented via correlation analysis, and the optimised segmentation parameter was achieved using the estimation of scale parameter algorithm. Then, a supervised k-nearest neighbourhood classifier was used to identify the terrace damages in the segmented objects, and additional geometric features at the object level were considered for classification. The comparison with the ground truth, as delineated by the image and field survey, showed that proposed classification can be adequately performed. The F-measures of extraction on three terrace damages were 92.07% (terrace sinkhole), 81.95% (ridge sinkhole), and 85.17% (collapse), and the Kappa coefficient was 85.34%. Finally, the potential application and spatial distribution of the terrace damages in this study were determined. We believe that this work can provide a credible framework for mapping terrace damages in the Loess Plateau of China.

A New Lightweight Convolutional Neural Network for Multi-Scale Land Surface Water Extraction from GaoFen-1D Satellite Images

Mapping land surface water automatically and accurately is closely related to human activity, biological reproduction, and the ecological environment. High spatial resolution remote sensing image (HSRRSI) data provide extensive details for land surface water and gives reliable data support for the accurate extraction of land surface water information. The convolutional neural network (CNN), widely applied in semantic segmentation, provides an automatic extraction method in land surface water information. This paper proposes a new lightweight CNN named Lightweight Multi-Scale Land Surface Water Extraction Network (LMSWENet) to extract the land surface water information based on GaoFen-1D satellite data of Wuhan, Hubei Province, China. To verify the superiority of LMSWENet, we compared the efficiency and water extraction accuracy with four mainstream CNNs (DeeplabV3+, FCN, PSPNet, and UNet) using quantitative comparison and visual comparison. Furthermore, we used LMSWENet to extract land surface water information of Wuhan on a large scale and produced the land surface water map of Wuhan for 2020 (LSWMWH-2020) with 2m spatial resolution. Random and equidistant validation points verified the mapping accuracy of LSWMWH-2020. The results are summarized as follows: (1) Compared with the other four CNNs, LMSWENet has a lightweight structure, significantly reducing the algorithm complexity and training time. (2) LMSWENet has a good performance in extracting various types of water bodies and suppressing noises because it introduces channel and spatial attention mechanisms and combines features from multiple scales. The result of land surface water extraction demonstrates that the performance of LMSWENet exceeds that of the other four CNNs. (3) LMSWENet can meet the requirement of high-precision mapping on a large scale. LSWMWH-2020 can clearly show the significant lakes, river networks, and small ponds in Wuhan with high mapping accuracy.

A comparative mapping of plant species diversity using ensemble learning algorithms combined with high accuracy surface modeling

Plant species diversity (PSD) has always been an essential component of biodiversity and plays an important role in ecosystem functions and services. However, it is still a huge challenge to simulate the spatial distribution of PSD due to the difficulties of data acquisition and unsatisfactory performance of predicting algorithms over large areas. A surge in the number of remote sensing imagery, along with the great success of machine learning, opens new opportunities for the mapping of PSD. Therefore, different machine learning algorithms combined with high-accuracy surface modeling (HASM) were firstly proposed to predict the PSD in the Xinghai, northeastern Qinghai-Tibetan Plateau, China. Spectral reflectance and vegetation indices, generated from Landsat 8 images, and environmental variables were taken as the potential explanatory factors of machine learning models including least absolute shrinkage and selection operator (Lasso), ridge regression (Ridge), eXtreme Gradient Boosting (XGBoost), and Random Forest (RF). The prediction generated from these machine learning methods and in situ observation data were integrated by using HASM for the high-accuracy mapping of PSD including three species diversity indices. The results showed that PSD was closely associated with vegetation indices, followed by spectral reflectance and environmental factors. XGBoost combined with HASM (HASM-XGBoost) showed the best performance with the lowest MAE and RMSE. Our results suggested that the fusion of heterogeneous data and the ensemble of heterogeneous models may revolutionize our ability to predict the PSD over large areas, especially in some places limited by sparse field samples.