Accurate classification of individual tree species is essential for inventorying, managing, and protecting forest resources. Individual tree species classification in subtropical forests remains ...challenging as existing individual tree segmentation algorithms typically result in over-segmentation in subtropical broadleaf forests, in which tree crowns often have multiple peaks. In this study, we proposed a watershed-spectral-texture-controlled normalized cut (WST-Ncut) algorithm, and applied it to delineate individual trees in a subtropical broadleaf forest situated in Shenzhen City of southern China (114°23′28″E, 22°43′50″N). Using this algorithm, we first obtained accurate crown boundary of individual broadleaf trees. We then extracted different suites of vertical structural, spectral, and textural features from UAV-based LiDAR, hyperspectral, and ultrahigh-resolution RGB data, and used these features as inputs to a random forest classifier to classify 18 tree species. The results showed that the proposed WST-Ncut algorithm could reduce the over-segmentation of the watershed segmentation algorithm, and thereby was effective for delineating individual trees in subtropical broadleaf forests (Recall = 0.95, Precision = 0.86, and F-score = 0.91). Combining the structural, spectral, and textural features of individual trees provided the best tree species classification results, with overall accuracy reaching 91.8%, which was 10.2%, 13.6%, and 19.0% higher than that of using spectral, structural, and textural features alone, respectively. In addition, results showed that better individual tree segmentation would lead to higher accuracy of tree species classification, but the increase of the number of tree species would result in the decline of classification accuracy.
•A watershed-spectral-texture-controlled normalized cut (WST-Ncut) is proposed.•The WST-Ncut method reduces over-segmentation in subtropical broadleaf forests.•Fusing structure, spectrum, and texture get the best tree species classification.•Better individual tree segmentation leads to better tree species classification.•The increase of tree species leads to the decrease of classification accuracy.
In this article, we propose an efficient and effective framework to fuse hyperspectral and light detection and ranging (LiDAR) data using two coupled convolutional neural networks (CNNs). One CNN is ...designed to learn spectral-spatial features from hyperspectral data, and the other one is used to capture the elevation information from LiDAR data. Both of them consist of three convolutional layers, and the last two convolutional layers are coupled together via a parameter-sharing strategy. In the fusion phase, feature-level and decision-level fusion methods are simultaneously used to integrate these heterogeneous features sufficiently. For the feature-level fusion, three different fusion strategies are evaluated, including the concatenation strategy, the maximization strategy, and the summation strategy. For the decision-level fusion, a weighted summation strategy is adopted, where the weights are determined by the classification accuracy of each output. The proposed model is evaluated on an urban data set acquired over Houston, USA, and a rural one captured over Trento, Italy. On the Houston data, our model can achieve a new record overall accuracy (OA) of 96.03%. On the Trento data, it achieves an OA of 99.12%. These results sufficiently certify the effectiveness of our proposed model.
With the development of the sensor technology, complementary data of different sources can be easily obtained for various applications. Despite the availability of adequate multisource observation ...data, for example, hyperspectral image (HSI) and light detection and ranging (LiDAR) data, existing methods may lack effective processing on structural information transmission and physical properties alignment, weakening the complementary ability of multiple sources in the collaborative classification task. The complementary information collaboration manner and the redundancy exclusion operator need to be redesigned for strengthening the semantic relatedness of multisources. As a remedy, we propose a structural optimization transmission framework, namely, structural optimization transmission network (SOT-Net), for collaborative land-cover classification of HSI and LiDAR data. Specifically, the SOT-Net is developed with three key modules: 1) cross-attention module; 2) dual-modes propagation module; and 3) dynamic structure optimization module. Based on above designs, SOT-Net can take full advantage of the reflectance-specific information of HSI and the detailed edge (structure) representations of multisource data. The inferred transmission plan, which integrates a self-alignment regularizer into the classification task, enhances the robustness of the feature extraction and classification process. Experiments show consistent outperformance of SOT-Net over baselines across three benchmark remote sensing datasets, and the results also demonstrate that the proposed framework can yield satisfying classification result even with small-size training samples.
Joint use of multisensor information has attracted considerable attention in the remote sensing community. While applications in land-cover observation benefit from information diversity, multisensor ...integration technique is confronted with many challenges, including inconsistent size of data, different data structures, uncorrelated physical properties, and scarcity of training data. In this article, an information fusion network, named interleaving perception convolutional neural network (IP-CNN), is proposed for integrating heterogeneous information and improving joint classification performance of hyperspectral image (HSI) and light detection and ranging (LiDAR) data. Specifically, a bidirectional autoencoder is designed to reconstruct hyperspectral and LiDAR data together, and the reconstruction process is trained with no dependence upon annotated information. Both HSI-perception constraint and LiDAR-perception constraint are imposed on multisource structural information integration. Accordingly, fused data are fed into a two-branch CNN for final classification. To validate the effectiveness of the model, the experiments were conducted using three datasets (i.e., Muufl Gulfport data, Trento data, and Houston data). The final results demonstrate that the proposed framework can significantly outperform state-of-the-art methods even with small-size training samples.
A three-dimensional variational (3DVAR) lidar data assimilation method is developed based on the Community Radiative Transfer Model (CRTM) and Weather Research and Forecasting model coupled to ...Chemistry (WRF-Chem) model. A 3DVAR data assimilation (DA) system using lidar extinction coefficient observation data is established, and variables from the Model for Simulating Aerosol Interactions and Chemistry (MOSAIC) mechanism of the WRF-Chem model are employed. Hourly lidar extinction coefficient data from 12:00 to 18:00 UTC on March 13, 2018 at four stations in Beijing are assimilated into the initial field of the WRF-Chem model; subsequently, a 24 h PM2.5 concentration forecast is made. Results indicate that assimilating lidar data can effectively improve the subsequent forecast. PM2.5 forecasts without using lidar DA are remarkably underestimated, particularly during heavy haze periods; in contrast, forecasts of PM2.5 concentrations with lidar DA are closer to observations, the model low bias is evidently reduced, and the vertical distribution of the PM2.5 concentration in Beijing is distinctly improved from the surface to 1200 m. Of the five aerosol species, improvements of NO3− are the most significant. The correlation coefficient between PM2.5 concentration forecasts with lidar DA and observations at 12 stations in Beijing is increased by 0.45, and the corresponding average RMSE is decreased by 25 μg·m−3, which respectively compared to those without DA.
Display omitted
•The 3DVAR assimilation system of lidar extinction coefficient data developed based on CRTM and WRF-Chem;•PM2.5 forecasts with lidar DA close to observations, while those without DA remarkably underestimated;•The vertical distribution of PM2.5 distinctly improved with lidar DA and significant improvement for nitrate.
•A new method for large-scale photovoltaic arrays placement optimisation is proposed.•Buildings’ topographic information is obtained from Light Detection And Ranging data.•Optimisation considers ...direct, anisotropic diffuse and reflected irradiances.•Shadowing from the surrounding environment is fully captured.•The method is highly suitable for urban environments with high obstruction.
The availability of high-resolution LiDAR (Light Detection And Ranging) geospatial data has increased immensely, providing new opportunities to solve challenges in the field of spatial energy planning. This paper presents a new method for large-scale placement of photovoltaic arrays over buildings’ rooftops in an optimal manner by using the global optimisation approach. The position, aspect and slope are the ey geometrical parameters being optimised for each photovoltaic array. The predicted energy generation (i.e. photovoltaic potential) is simulated by using state-of-the-art hourly shadowing estimation from the surroundings, anisotropic diffuse, reflected, and direct irradiances that are based on a Typical Meteorological Year, and non-linear efficiency characteristics of a considered photovoltaic system configuration. The optimisation performs multiple simulation scenarios throughout an entire year for each photovoltaic array, in order to maximise its photovoltaic potential. The method was tested over three LiDAR datasets with different landscape topographies and urban densities. In comparison to the methods for photovoltaic arrays’ fixed optimal slope estimation, the proposed method is substantially more suitable for application in urban environments.
•Deep neural network (DNN) allows the fine-mapping of complex wetland areas.•PCA-based feature selection was effective in the optimization of DNN performance.•WorldView-3 data are suitable to ...preserve shape and extent of small wetlands.•DNN classification achieved relatively similar accuracies to other ML methods.
Wetland inventory maps are essential information for the conservation and management of natural wetland areas. The classification framework is crucial for successful mapping of complex wetlands, including the model selection, input variables and training procedures. In this context, deep neural network (DNN) is a powerful technique for remote sensing image classification, but this model application for wetland mapping has not been discussed in the previous literature, especially using commercial WorldView-3 data. This study developed a new framework for wetland mapping using DNN algorithm and WorldView-3 image in the Millrace Flats Wildlife Management Area, Iowa, USA. The study area has several wetlands with a variety of shapes and sizes, and the minimum mapping unit was defined as 20 m2 (0.002 ha). A set of potential variables was derived from WorldView-3 and auxiliary LiDAR data, and a feature selection procedure using principal components analysis (PCA) was used to identify the most important variables for wetland classification. Furthermore, traditional machine learning methods (support vector machine, random forest and k-nearest neighbor) were also implemented for the comparison of results. In general, the results show that DNN achieved satisfactory results in the study area (overall accuracy = 93.33 %), and we observed a high spatial overlap between reference and classified wetland polygons (Jaccard index ∼0.8). Our results confirm that PCA-based feature selection was effective in the optimization of DNN performance, and vegetation and textural indices were the most informative variables. In addition, the comparison of results indicated that DNN classification achieved relatively similar accuracies to other methods. The total classification errors vary from 0.104 to 0.111 among the methods, and the overlapped areas between reference and classified polygons range between 87.93 and 93.33 %. Finally, the findings of this study have three main implications. First, the integration of DNN model and WorldView-3 image is useful for wetland mapping at 1.2-m, but DNN results did not outperform other methods in this study area. Second, the feature selection was important for model performance, and the combination of most relevant input parameters contributes to the success of all tested models. Third, the spatial resolution of WorldView-3 is appropriate to preserve the shape and extent of small wetlands, while the application of medium resolution image (30-m) has a negative impact on the accurate delineation of these areas. Since commercial satellite data are becoming more affordable for remote sensing users, this study provides a framework that can be utilized to integrate very high-resolution imagery and deep learning in the classification of complex wetland areas.
Biodiversity is considered to be an essential element of the Earth system, driving important ecosystem services. However, the conservation of biodiversity in a quickly changing world is a challenging ...task which requires cost-efficient and precise monitoring systems. In the present study, the suitability of airborne discrete-return LiDAR data for the mapping of vascular plant species richness within a Sub-Mediterranean second growth native forest ecosystem was examined. The vascular plant richness of four different layers (total, tree, shrub and herb richness) was modeled using twelve LiDAR-derived variables. As species richness values are typically count data, the corresponding asymmetry and heteroscedasticity in the error distribution has to be considered. In this context, we compared the suitability of random forest (RF) and a Generalized Linear Model (GLM) with a negative binomial error distribution. Both models were coupled with a feature selection approach to identify the most relevant LiDAR predictors and keep the models parsimonious. The results of RF and GLM agreed that the three most important predictors for all four layers were altitude above sea level, standard deviation of slope and mean canopy height. This was consistent with the preconception of LiDAR's suitability for estimating species richness, which is its capacity to capture three types of information: micro-topographical, macro-topographical and canopy structural. Generalized Linear Models showed higher performances (r2: 0.66, 0.50, 0.52, 0.50; nRMSE: 16.29%, 19.08%, 17.89%, 21.31% for total, tree, shrub and herb richness respectively) than RF (r2: 0.55, 0.33, 0.45, 0.46; nRMSE: 18.30%, 21.90%, 18.95%, 21.00% for total, tree, shrub and herb richness, respectively). Furthermore, the results of the best GLM were more parsimonious (three predictors) and less biased than the best RF models (twelve predictors). We think that this is due to the mentioned non-symmetric error distribution of the species richness values, which RF is unable to properly capture.
From an ecological perspective, the predicted patterns agreed well with the known vegetation composition of the area. We found especially high species numbers at low elevations and along riversides. In these areas, overlapping distributions of thermopile sclerophyllos species, water demanding Valdivian evergreen species and species growing in Nothofagus obliqua forests occur.
The three main conclusions of the study are: 1) appropriate model selection is crucial when working with biodiversity count data; 2) the application of RF for data with non-symmetric error distributions is questionable; and 3) structural and topographic information derived from LiDAR data is useful for predicting local plant species richness.
•Vascular plant richness for total, tree, shrub and herb richness were estimated.•Generalized Linear Models (GLM) and random forest (RF) were compared.•GLM assuming a negative binomial error distribution outperformed random forest.•RF dealt less efficiently with the asymmetric error distribution of count data.•Total species richness was estimated with good accuracies (r2=0.66, nRMSE=16.29%).
Domain adaption (DA) is a challenging task that integrates knowledge from source domain (SD) to perform data analysis for target domain. Most of the existing DA approaches only focus on ...single-source-single-target setting. In contrast, multisource (MS) data collaborative utilization has been extensively used in various applications, while how to integrate DA with MS collaboration still faces great challenges. In this article, we propose a multilevel DA network (MDA-NET) for promoting information collaboration and cross-scene (CS) classification based on hyperspectral image (HSI) and light detection and ranging (LiDAR) data. In this framework, modality-related adapters are built, and then a mutual-aid classifier is used to aggregate all the discriminative information captured from different modalities for boosting CS classification performance. Experimental results on two cross-domain datasets show that the proposed method consistently provides better performance than other state-of-the-art DA approaches.
In this study we fused high-spatial resolution (3.7m) hyperspectral imagery with 22pulse/m2 lidar data at the individual crown object scale to map 29 common tree species in Santa Barbara, California, ...USA. We first adapted and parallelized a watershed segmentation algorithm to delineate individual crowns from a gridded canopy maxima model. From each segment, we extracted all spectra exceeding a Normalized Difference Vegetation Index (NDVI) threshold and a suite of crown structural metrics computed directly from the three-dimensional lidar point cloud. The variables were fused and crowns were classified using canonical discriminant analysis. The full complement of spectral bands along with 7 lidar-derived structural metrics were reduced to 28 canonical variates and classified. Species-level and leaf-type level maps were produced with respective overall accuracies of 83.4% (kappa=82.6) and 93.5%. The addition of lidar data resulted in an increase in classification accuracy of 4.2 percentage points over spectral data alone. The value of the lidar structural metrics for urban species discrimination became particularly evident when mapping crowns that were either small or morphologically unique. For instance, the accuracy with which we mapped the tall palm species Washingtonia robusta increased from 29% using spectral bands to 71% with the fused dataset. Additionally, we evaluated the role that automated segmentation plays in classification error and the prospects for mapping urban forest species not included in a training sample. The ability to accurately map urban forest species is an important step towards spatially explicit urban forest ecosystem assessment.
•We map 29 urban tree species using hyperspectral and lidar data fusion.•Crown-objects are delineated using watershed segmentation on a canopy maxima model.•Species classified with 83.4% accuracy using canonical discriminant analysis•Lidar structural metrics critical for classifying species with small crowns•Segmentation errors only minimally impacted classification accuracy.