High Resolution Multispectral Imagery Research Articles

Delineation of Individual Tree Crowns Using High Spatial Resolution Multispectral WorldView-3 Satellite Imagery

Delineation of individual tree crowns can provide valuable information for sustainable forest management and environmental protection. However, it is hard to find a reliable tree crown delineation method that can continually generate expected results in high spatial resolution multispectral satellite images, because most of the existing methods need user-assigned parameters that greatly affect the quality of the delineation results. In this article, we propose a method based on the marker-controlled watershed segmentation to delineate individual tree crowns using high spatial resolution multispectral WorldView-3 satellite imagery. A gradient binarization process is proposed to accurately locate tree crown borders. The threshold for the binarization is determined by a supervised searching process. Markers used in marker-controlled watershed segmentation are spatial local maxima detected from the information provided by tree crown borders. Moreover, the definition of spatial local maxima from the literature is improved to eliminate false treetops. To validate the performance of the proposed delineation method, delineation results are compared with those obtained from the spectral angle segmentation (SAS) method that has been proposed in literature because the quality of delineation results generated by SAS does not rely on user-assigned parameters. The experiment results in two test images demonstrate that the proposed method outperforms SAS in terms of both delineation accuracy and visual quality of the delineation map. Moreover, it is proved that the modified spatial local maxima are more reliable for detecting treetops.

The use of UAV-based remote sensing to estimate biomass and carbon stock for native desert shrubs

Unmanned Aerial Vehicles (UAVs) have started to receive more attention in the ecological field in the past 15 years, as they provide very high-resolution imagery that ranges from meters to millimeters. Very high-resolution multispectral imagery obtained from UAVs can help in assessing and monitoring native desert vegetation. Thus, this study use UAVs to develop a method to estimate the biomass and carbon stock of native desert shrubs. The method integrates different techniques and software to monitor native plants' coverage, biomass, and carbon stock. The techniques used in this work are also applicable for other native desert shrubs in the region and could support ecosystem managers in assessing and monitoring arid ecosystems and restoration and revegetation programs. A three-stage image and data management are discussed, including: (1) fieldwork and image acquisition using UAVs, (2) image pre-processing, and (3) image processing using different techniques and software.•Determining shrub biomass is not restricted to multispectral data only but could be applicable for RGB data since it mainly depends on the DSM and DTM.•Allometric parameters could help in estimating desert shrub biomass which could be measured easily and rapidly using UAV imagery.•SVM Supervised classification could help in distinguishing between native shrubs and grasses.

Remote sensing of american maple in alluvial forests: a case study in an island complex of the Loire valley (France)

Due to their particular topographic position between land and river, riparian forests are ecosystems rich in biodiversity. In France, along the Middle Loire (from Nevers to Angers), Black poplar (Populus nigra L.) forests are often in mixtures with the American maple (Acer negundo L.), introduced into the country in the 18th century. We tested the detectability of American maple by LiDAR and very high-resolution multispectral imagery on an island complex. We found that coupling the point cloud height standard deviation with a vegetation index in the red, green and blue spectrums discriminated American maple with a success rate of more than 90%.

Automated surface water detection from space: a Canada-wide, open-source, automated, near-real time solution

The goal of this research was to develop a fully automated method to map open water extent that is operationally practical on a national scale. Such a system needs to produce acceptable results in all regions of the country and particularly in the Prairie Potholes Region where understanding water surface dynamics is important for predicting flooding, agriculture/water availability and for evaporation calculations in weather models. A system was developed to automate, ingest, and process Radarsat-2 (RS2) imagery, from which mapping open water body extents in near-real time was carried out using a machine learning classification technique. A Random Forest classification algorithm was trained using the data extracted from the Global Surface Water (GSW) occurrence dataset. The GSW occurrence thresholds used to extract the training data were examined and there was little influence of uncertainty on the classification. The quality of classifications generated from RS2 Fine Wide mode imagery improved with increasing incidence angle. All Fine Quad incident angles produced acceptable results, but Standard and Wide mode imagery produced results below the accuracy thresholds deemed acceptable for this operational solution. Validation was carried out by comparing mapped water extents to temporally coincident high resolution multi-spectral imagery and to the USGS Global Land Cover Characteristics dataset, that is currently used as a land-water mask by ECCC in weather numerical weather modelling. The system that has been developed will allow new image datasets (e.g. Radarsat Constellation Mission) or training data that becomes available to be used to improve models. The open source code will be made available on Github.

Understanding the Effects of Optimal Combination of Spectral Bands on Deep Learning Model Predictions: A Case Study Based on Permafrost Tundra Landform Mapping Using High Resolution Multispectral Satellite Imagery.

Deep learning (DL) convolutional neural networks (CNNs) have been rapidly adapted in very high spatial resolution (VHSR) satellite image analysis. DLCNN-based computer visions (CV) applications primarily aim for everyday object detection from standard red, green, blue (RGB) imagery, while earth science remote sensing applications focus on geo object detection and classification from multispectral (MS) imagery. MS imagery includes RGB and narrow spectral channels from near- and/or middle-infrared regions of reflectance spectra. The central objective of this exploratory study is to understand to what degree MS band statistics govern DLCNN model predictions. We scaffold our analysis on a case study that uses Arctic tundra permafrost landform features called ice-wedge polygons (IWPs) as candidate geo objects. We choose Mask RCNN as the DLCNN architecture to detect IWPs from eight-band Worldview-02 VHSR satellite imagery. A systematic experiment was designed to understand the impact on choosing the optimal three-band combination in model prediction. We tasked five cohorts of three-band combinations coupled with statistical measures to gauge the spectral variability of input MS bands. The candidate scenes produced high model detection accuracies for the F1 score, ranging between 0.89 to 0.95, for two different band combinations (coastal blue, blue, green (1,2,3) and green, yellow, red (3,4,5)). The mapping workflow discerned the IWPs by exhibiting low random and systematic error in the order of 0.17–0.19 and 0.20–0.21, respectively, for band combinations (1,2,3). Results suggest that the prediction accuracy of the Mask-RCNN model is significantly influenced by the input MS bands. Overall, our findings accentuate the importance of considering the image statistics of input MS bands and careful selection of optimal bands for DLCNN predictions when DLCNN architectures are restricted to three spectral channels.

Remotely Sensed Changes in Vegetation Cover Distribution and Groundwater along the Lower Gila River

Introduced as a soil erosion deterrent, salt cedar has become a menace along riverbeds in the desert southwest. Salt cedar replaces native species, permanently altering the structure, composition, function, and natural processes of the landscape. Remote sensing technologies have the potential to monitor the level of invasion and its impacts on ecosystem services. In this research, we developed a species map by segmenting and classifying various species along a stretch of the Lower Gila River. We calculated metrics from high-resolution multispectral imagery and light detection and ranging (LiDAR) data to identify salt cedar, mesquite, and creosote. Analysts derived training and validation information from drone-acquired orthophotos to achieve an overall accuracy of 94%. It is clear from the results that salt cedar completely dominates the study area with small numbers of mesquite and creosote present. We also show that vegetation has declined in the study area over the last 25 years. We discuss how water usage may be influencing the plant health and biodiversity in the region. An examination of ground well, stream gauge, and Gravity Recovery and Climate Experiment (GRACE) groundwater storage data indicates a decline in water levels near the study area over the last 25 years.

Assessment of Olive Tree Canopy Characteristics and Yield Forecast Model Using High Resolution UAV Imagery

Greek agriculture is mainly based on olive tree cultivation. Farmers have always been concerned about annual olive orchard production. The necessity for the improvement of farming practices initiated the development of new technological tools that are useful in agriculture. The main goal of this study is the utilization of new technologies in order to define the geometry of olive tree configuration, while the development of a forecasting model of annual production in a non-linear olive grove, planted on a hilly uneven terrain is the secondary goal. The field’s orthomosaic, its Digital Terrain Model (DTM) and Digital Surface Model (DSM) were created by employing high resolution multispectral imagery. The Normalized Difference Vegetation Index (NDVI) thematic map has also been developed. The trees’ crowns were isolated employing the field’s orthomosaic, rendering individual polygons for each tree through Object Based Image Analysis (OBIA). The measurements were conducted in a Geographic Information System (GIS) environment and were also verified by ground ones. Tree crown height, surface, and volume were calculated, and thematic maps for each variable were created, allowing for the observation of the spatial distribution for each parameter. The compiled data were statistically analyzed revealing important correlations among different variables. These were employed to produce a model, which would enable production forecasting in kilograms per tree. The spatial distribution of the variables gave noteworthy results due to the similar pattern they followed. Future crop yield optimization, even at a tree level, can be based on the results of the present study. Its conclusions may lead to the development and implementation of precision olive tree cultivation practices.

Evaluation of Features Derived from High-Resolution Multispectral Imagery and LiDAR Data for Object-Based Support Vector Machine Classification of Tree Species

Remote sensing can play a key role in understanding the make-up of urban forests. This study analyzes how high-resolution Geoeye-1 multispectral imagery and LiDAR point clouds allow for improved classification of urban tree species using object-based and support vector machine classification (SVM). Five common urban trees are classified: Acer platanoides; Acer platanoides ‘Schwedleri’; Picea pungens; Gleditsia triacanthos; and Tilia cordata. Numerous features are used for classification: index derived from imagery reflectance; texture of imagery; LiDAR height and intensity; and a LiDAR-generated normalized digital surface model. Classification is performed to evaluate the contribution of individual features, groups of features, and the combination of features from both imagery and LiDAR data. Classification results in an overall accuracy of 85.08% when features from both data sources are combined, compared with 77.73% when using only LiDAR features, and 71.85% when using only imagery features.

Tree species classification using UAS-based digital aerial photogrammetry point clouds and multispectral imageries in subtropical natural forests

Tree species composition of forest stand is an important indicator of forest inventory attributes for assessing ecosystem health, understanding successional processes, and digitally displaying forest biodiversity. In this study, we acquired high spatial resolution multispectral and RGB imagery over a subtropical natural forest in southwest China using a fixed-wing UAV system. Digital aerial photogrammetric (DAP) technique was used to generate multi-spectral and RGB derived point clouds, upon which individual tree crown (ITC) delineation algorithms and a machine learning classifier were used to identify dominant tree species. To do so, the structure-from-motion method was used to generate RGB imagery-based DAP point clouds. Then, three ITC delineation algorithms (i.e., point cloud segmentation (PCS), image-based multiresolution segmentation (IMRS), and advanced multiresolution segmentation (AMRS)) were used and assessed for ITC detection. Finally, tree-level metrics (i.e., multispectral, texture and point cloud metrics) were used as metrics in the random forest classifier used to classify eight dominant tree species. Results indicated that the accuracy of the AMRS ITC segmentation was highest (F1-score = 82.5 %), followed by the segmentation using PCS (F1-score = 79.6 %), the IMRS exhibited the lowest accuracy (F1-score = 78.6 %); forest types classification (coniferous and deciduous) had a higher accuracy than the classification of all eight tree species, and the combination of spectral, texture and structural metrics had the highest classification accuracy (overall accuracy = 80.20 %). In the classification of both eight tree species and two forest types, the classification accuracies were lowest when only using spectral metrics, indicated that the texture metrics and point cloud structural metrics had a positive impact on the classification (the overall accuracy and kappa accuracy increased by 1.49–4.46 % and 2.86–6.84 %, respectively).

Study on the Coastline Evolution in Sopot (2008–2018) Based on Landsat Satellite Imagery

The coastline is the boundary between the water surface in a reservoir or watercourse and the land, which is characterised by high instability and functional diversity. For these reasons, research on coastal monitoring has been conducted for several decades. Currently, satellite images performed with synthetic aperture radars (SARs) are used to determine its course and variability together with high-resolution multispectral imagery from satellites such as IKONOS, QuickBird, and WorldView, or moderate-resolution multispectral images from Landsat satellites. This paper analysed the coastline variability in Sopot (2008–2018) based on Landsat satellite imagery. Furthermore, based on multispectral images obtained, it was determined how the beach surface in Sopot changed. Research has shown that the coastline keeps moving away from the land every year. This was particularly noticeable between 2008 and 2018 when the coastline moved on average 19.1 m towards the Baltic Sea. Moreover, it was observed that the area of the sandy beach in Sopot increased by 14 170.6 m2, which translates into an increase of 24.7% compared to 2008. The probable cause of the continuous coastline shift towards the sea and the increase of the beach surface is the oceanographic phenomenon called tombolo, which occurred in this area as a result of the construction of a yacht marina near the coast.

A methodological approach to assess canopy NDVI–based tomato dynamics under irrigation treatments

A tomato field experiment under deficit irrigation treatments was established to assess the feasibility of nonlinear mixed models to predict the change in the canopy Normalized Difference Vegetation Index (NDVI) along the growing season. The irrigation doses were 50, 70, 85, and 100 % of the actual evapotranspiration (ETc). The experiment was performed in the Peruvian Arid Coastal Region (13.10 S, 76.37 W). On-ground destructive plant measurements (7 times) allowed to obtain the leaf area index (LAI) and the canopy equivalent water thickness (EWTcanopy) to validate the experiment. Similarly, volumetric soil moisture was monitored at three hours interval using time-domain reflectometry. High-resolution multispectral imagery was acquired along the growing season (10 times) using an unmanned aerial vehicle. NDVI at plant level was obtained based on oversegmented imagery. The canopy NDVI growth curves were obtained in relation to the irrigation treatments and growing degree days (GDD) using a nonlinear mixed model.Deficit irrigation doses (50, 70 and 85 % ETc) induced lower biomass accumulation and greatly affected LAI and EWTcanopy, attaining lower values (p < 0.05) than the control (100 % ETc). In spite of reaching higher LAI and EWTcanopy values, 70 % ETc was not significantly different from 50 % ETc (p < 0.05) showing the difficulty to accurately discriminate treatments with high deficit irrigation; 85 % ETc produced important differences on EWTcanopy and LAI compared to the other deficit irrigation treatments (p < 0.05) due to a higher leaf area expansion. Deficit irrigation doses led to less vigorous plants resulting in lower NDVI compared to the control. Notwithstanding 50 % ETc obtained lower NDVI values than 70 % ETc, differences were not significant. Volumetric soil moisture was consistent with the irrigation treatments, resulting in a decreased trend with decreasing water application along the growing season. A three-parameter logistic mixed model was found to be well-suited to fit the tomato canopy NDVI growth curves in function of the irrigation treatments (i.e. fixed effects) considering the plant-specific variability (i.e. random effects). At maturity stage, significant differences of the deficit irrigation treatments from the control were obtained (p < 0.05) when cumulative effect of the treatments was well-defined and NDVI reached asymptotic behavior. During the vegetative stage, the inflection point marked the fastest NDVI growth rate for the 100 % ETc treatment whereas the lowest for the 50 % ETc (p < 0.05). The diagnosis plots and the low root mean squared error of the residuals indicated the suitability of using the three-parameter logistic growth curve estimated as a nonlinear mixed model to predict NDVI.

A novel water body extraction neural network (WBE-NN) for optical high-resolution multispectral imagery

Surface water mapping is very important for studying its role in global water cycle, flooding dynamic monitoring, and water resources management. The most famous techniques for water body extraction like those based on water spectral indices (WSI) require rich spectral information. However, the WSI methods are no longer practical in high-resolution multispectral (MS) images due to insufficient spectral information. In addition, surface water mapping faces an utmost overestimation issue because shadows are misclassified as water bodies. To address the above-mentioned problems, in this paper, a novel refined water body extraction neural network (WBE-NN) is proposed. The global spatial-spectral convolution (GSSC) module is developed to enhance surface water body features. A novel multiscale learning module is designed to extract multi-scale contextual information. In addition, the surface water body boundary refinement (SWBBR) module is adopted to enhance surface water body boundaries. The results show that the proposed method achieved good performance with a mean overall accuracy of 98.97%, a mean Kappa coefficient of 94.78%, and a mean boundary overall accuracy of 98.01%. Therefore, WBE-NN can be used for mapping surface water with high accuracy in complex areas.

Mapping Multiple Insect Outbreaks across Large Regions Annually Using Landsat Time Series Data

Forest insect outbreaks have caused and will continue to cause extensive tree mortality worldwide, affecting ecosystem services provided by forests. Remote sensing is an effective tool for detecting and mapping tree mortality caused by forest insect outbreaks. In this study, we map insect-caused tree mortality across three coniferous forests in the Western United States for the years 1984 to 2018. First, we mapped mortality at the tree level using field observations and high-resolution multispectral imagery collected in 2010, 2011, and 2018. Using these high-resolution maps of tree mortality as reference images, we then classified moderate-resolution Landsat imagery as disturbed or undisturbed and for disturbed pixels, predicted percent tree mortality with random forest (RF) models. The classification approach and RF models were then applied to time series of Landsat imagery generated with Google Earth Engine (GEE) to create annual maps of percent tree mortality. We separated disturbed from undisturbed forest with overall accuracies of 74% to 80%. Cross-validated RF models explained 61% to 68% of the variation in percent tree mortality within disturbed 30-m pixels. Landsat-derived maps of tree mortality were comparable to vector aerial survey data for a variety of insect agents, in terms of spatial patterns of mortality and annual estimates of total mortality area. However, low-level tree mortality was not always detected. We conclude that our methodology has the potential to generate reasonable estimates of annual tree mortality across large extents.

Assessing alternative methods for unsupervised segmentation of urban vegetation in very high-resolution multispectral aerial imagery.

To analyze types and patterns of greening trends across a city, this study seeks to identify a method of creating very high-resolution urban vegetation maps that scales over space and time. Vegetation poses unique challenges for image segmentation because it is patchy, has ragged boundaries, and high in-class heterogeneity. Existing and emerging public datasets with the spatial resolution necessary to identify granular urban vegetation lack a depth of affordable and accessible labeled training data, making unsupervised segmentation desirable. This study evaluates three unsupervised methods of segmenting urban vegetation: clustering with k-means using k-means++ seeding; clustering with a Gaussian Mixture Model (GMM); and an unsupervised, backpropagating convolutional neural network (CNN) with simple iterative linear clustering superpixels. When benchmarked against internal validity metrics and hand-coded data, k-means is more accurate than GMM and CNN in segmenting urban vegetation. K-means is not able to differentiate between water and shadows, however, and when this segment is important GMM is best for probabilistically identifying secondary land cover class membership. Though we find the unsupervised CNN shows high degrees of accuracy on built urban landscape features, its accuracy when segmenting vegetation does not justify its complexity. Despite limitations, for segmenting urban vegetation, k-means has the highest performance, is the simplest, and is more efficient than alternatives.

Monthly estimation of the surface water extent in France at a 10-m resolution using Sentinel-2 data

The first national product of Surface Water Dynamics in France (SWDF) is generated on a monthly temporal scale and 10-m spatial scale using an automatic rule-based superpixel (RBSP) approach. The current surface water dynamic products from high resolution (HR) multispectral satellite imagery are typically analyzed to determine the annual trend and related seasonal variability. Annual and seasonal time series analyses may fail to detect the intra-annual variations of water bodies. Sentinel-2 allows us to investigate water resources based on both spatial and temporal high-resolution analyses. We propose a new automatic RBSP approach on the Google Earth Engine platform. The RBSP method employs combined spectral indices and superpixel techniques to delineate the surface water extent; this approach avoids the need for training data and benefits large-scale, dynamic and automatic monitoring. We used the proposed RBSP method to process Sentinel-2 monthly composite images covering a two-year period and generate the monthly surface water extent at the national scale, i.e., over France. Annual occurrence maps were further obtained based on the pixel frequency in monthly water maps. The monthly dynamics provided in SWDF products are evaluated by HR satellite-derived water masks at the national scale (JRC GSW monthly water history) and at local scales (over two lakes, i.e., Lake Der-Chantecoq and Lake Orient, and 200 random sampling points). The monthly trends between SWDF and GSW were similar, with a coefficient of 0.94. The confusion matrix-based metrics based on the sample points were 0.885 (producer's accuracy), 0.963 (user's accuracy), 0.932 (overall accuracy) and 0.865 (Matthews correlation coefficient). The annual surface water extents (i.e., permanent and maximum) are validated by two HR satellite image-based water maps and an official database at the national scale and small water bodies (ponds) at the local scale at Loir-et-Cher. The results show that the SWDF results are closely correlated to the previous annual water extents, with a coefficient >0.950. The SWDF results are further validated for large rivers and lakes, with extraction rates of 0.929 and 0.802, respectively. Also, SWDF exhibits superiority to GSW in small water body extraction (taking 2498 ponds in Loir-et-Cher as example), with an extraction rate improved by approximately 20%. Thus, the SWDF method can be used to study interannual, seasonal and monthly variations in surface water systems. The monthly dynamic maps of SWDF improved the degree of land surface coverage by 25% of France on average compared with GSW, which is the only product that provides monthly dynamics. Further harmonization of Sentinel-2 and Landsat 8 and the introduction of enhanced cloud detection algorithm can fill some gaps of no-data regions.

ANALISIS PERUBAHAN HABITAT DASAR PERAIRAN DANGKAL MENGGUNAKAN CITRA SATELIT RESOLUSI TINGGI DI KARANG LEBAR, KEPULAUAN SERIBU

Keperluan data dan informasi tentang habitat bentik sangat diperlukan untuk menjaga dan melestarikan ekosistem yang ada di perairan. Rusaknya habitat bentik dapat terjadi karena adanya aktifitas antropogenik dan bencana alam yang akan berimbas pada biota dan ekosistem yang ada disekitarnya, oleh karena itu untuk mengetahui dan memantau kondisi perairan dan habitat perairan dangkal perlu dilakukannya pemetaan. Penelitian ini bertujuan untuk menganalisis perubahan habitat dasar perairan laut dangkal di Karang Lebar, Kepulauan Seribu. Penelitian ini memanfaatkan citra multispektral resolusi tinggi QuickBird 2008 dan WordView-2 2018 untuk mendeteksi perubahan geospasial habitat bentik. Klasifikasi citra multispektral dilakukan dengan penerapan algoritma SVM (Support Vector Machine) dan transformasi DII (Depth Invariant Index) pada kedua citra yang digunakan. Jumlah kelas habitat bentik yang dihasilkan adalah sebanyak lima kelas yaitu terumbu karang, karang mati, padang lamun, pasir, dan rubble. Hasil analisis menunjukkan akurasi keseluruhan 58,18% dan 70,9% pada penerapan klasifikasi dengan input band multispektral masing-masing untuk citra 2008 dan 2018, serta 60% dan 80% pada hasil transformasi DII masing masing untuk citra 2008 dan 2018. Perubahan kelas rubble di tahun 2008 menjadi kelas pasir tahun 2018 merupakan yang paling besar luasannya dibandingkan dengan perubahan pada kelas lainnya, yaitu 81,46 ha.

The 2019 Brumadinho tailings dam collapse: Possible cause and impacts of the worst human and environmental disaster in Brazil

On 25th January 2019, the tailings dam of the Brumadinho iron mine operated by Vale S/A failed catastrophically. The death toll stood at 259 and 11 people remained missing as of January 2020. This tragedy occurred three years after Mariana’s tailings dam rupture – the most significant tailing dam disaster in Brazilian history. Thus far, a systematic investigation on the cause and effect of the failure has yet to be conducted. Here, we use satellite-driven soil moisture index, multispectral high-resolution imagery and Interferometric Synthetic Aperture Radar (InSAR) products to assess pre-disaster scenarios and the direct causes of the tailings dam collapse. A decreasing trend in the moisture content at the surface and the full evanescence of pond water through time (2011–2019) suggest that the water was gradually penetrating the fill downwards and caused the seepage erosion, saturating the tailings dam. Large-scale slumping of the dam (extensional failure) upon the rupture indicates that the materials of the fill were already saturated. InSAR measurements reveal a dramatic, up to 30 cm subsidence in the dam (at the rear part) within the past 12 months before the dam collapse, signifying that the sediments had been removed from the fill. Although the information on the resistance level of the tailings dam to infiltrations is not available, these pieces of evidence collectively indicate that the seepage erosion (piping) is the primary cause for the chronic weakening of the structure and, hence, the internal “liquefaction” condition. Upon the collapse, the fully saturated mud tailings flowed down the gentle slope area (3.13 × 106 m2), where 73 % were originally covered by tree, grass or agricultural tracts. The toxic mud eventually reached the Paraopeba River after travelling 10 km, abruptly increasing the suspended particulate matter (SPM) concentration and the toxic chemical elements in the river, immediately affecting the local livelihoods that depend on its water. The Paraopeba River is a major tributary of the San Francisco River, the second-longest river in Brazil reaching the Atlantic Ocean. We anticipate that the environmental repercussions of this toxic seepage will be felt throughout the entire basin, especially riverine communities located downstream.

A New Individual Tree Crown Delineation Method for High Resolution Multispectral Imagery

In current individual tree crown (ITC) delineation methods for high-resolution multispectral imagery, either a spectral band or a brightness component of the multispectral image is employed in delineation with reference to edges or shapes of crowns, whereas spectra of tree crowns are seldom taken into account. Such methods normally perform well in coniferous forests with obvious between-crown shadows, but fail in dense deciduous or mixed forests, in which tree crowns are close to each other, between-crown shadows and boundaries are unobvious, whereas adjacent tree crowns may be of distinguishable spectra. In order to effectively delineate crowns in dense deciduous or mixed forests, a new ITC delineation method using both brightness and spectra of the image is proposed in this study. In this method, a morphological gradient map of the image is first generated, treetops of multi-scale crowns are extracted from the gradient map and refined regarding the spectral differences between neighboring crowns, the gradient map is segmented using a watershed approach with treetops as markers, and the resulting segmentation map is refined to yield a crown map. Evaluated on images of a rainforest and a deciduous forest, the proposed method more accurately delineated adjacent broad-leaved tree crowns with similar brightness but different spectra than the other two typical ITC delineation algorithms, achieving a delineation accuracy of up to 76% in the rainforest and 63% in the deciduous forest.

Cloud and Cloud Shadow Detection Based on Multiscale 3D-CNN for High Resolution Multispectral Imagery

Cloud and cloud shadow detection is one of the most important tasks for optical remote sensing image preprocessing. It is not an easy task due to the variety and complexity of underlying surfaces, such as the low-albedo objects (water and mountain shadow) and the high-albedo objects (snow and ice). In this study, an end-to-end multiscale 3D-CNN method is proposed for cloud and cloud shadow detection in high resolution multispectral imagery. Specifically, a multiscale learning module is designed to extract cloud and cloud shadow contextual information of different levels. In order to make full use of band information, four band-combination images are inputted into the multiscale 3D-CNN. A joint spectral-spatial information of 3D-convolution layer is developed to fully explore the joint spatial-spectral correlations feature in the input data. Overall, in the experiments undertaken in this paper, the proposed method achieved a mean overall accuracy of 97.27% for cloud detection, with a mean precision of 96.02% and a mean recall of 95.86%. For cloud shadow detection, the proposed method achieved a mean precision of 95.92% and a mean recall of 92.86%. Experimental results on two validation datasets (GF-1 WFV validation data and ZY-3 validation data) show that the proposed multiscale-3D-CNN method achieved good performance with limited spectral ranges.

Investigating the Use of Unmanned Aerial Vehicles and High-Resolution Multispectral Imagery to Characterize Grain Sorghum Senescence Patterns

Investigating the Use of Unmanned Aerial Vehicles and High-Resolution Multispectral Imagery to Characterize Grain Sorghum Senescence Patterns