The global availability of high spatial resolution images makes mapping tree species distribution possible for better management of forest resources. Previous research mainly focused on mapping ...single tree species, but information about the spatial distribution of all kinds of trees, especially plantations, is often required. This research aims to identify suitable variables and algorithms for classifying land cover, forest, and tree species. Bi-temporal ZiYuan-3 multispectral and stereo images were used. Spectral responses and textures from multispectral imagery, canopy height features from bi-temporal stereo imagery, and slope and elevation from the stereo-derived digital surface model data were examined through comparative analysis of six classification algorithms including maximum likelihood classifier (MLC), k-nearest neighbor (kNN), decision tree (DT), random forest (RF), artificial neural network (ANN), and support vector machine (SVM). The results showed that use of multiple source data—spectral bands, vegetation indices, textures, and topographic factors—considerably improved land-cover and forest classification accuracies compared to spectral bands alone, which the highest overall accuracy of 84.5% for land cover classes was from the SVM, and, of 89.2% for forest classes, was from the MLC. The combination of leaf-on and leaf-off seasonal images further improved classification accuracies by 7.8% to 15.0% for land cover classes and by 6.0% to 11.8% for forest classes compared to single season spectral image. The combination of multiple source data also improved land cover classification by 3.7% to 15.5% and forest classification by 1.0% to 12.7% compared to the spectral image alone. MLC provided better land-cover and forest classification accuracies than machine learning algorithms when spectral data alone were used. However, some machine learning approaches such as RF and SVM provided better performance than MLC when multiple data sources were used. Further addition of canopy height features into multiple source data had no or limited effects in improving land-cover or forest classification, but improved classification accuracies of some tree species such as birch and Mongolia scotch pine. Considering tree species classification, Chinese pine, Mongolia scotch pine, red pine, aspen and elm, and other broadleaf trees as having classification accuracies of over 92%, and larch and birch have relatively low accuracies of 87.3% and 84.5%. However, these high classification accuracies are from different data sources and classification algorithms, and no one classification algorithm provided the best accuracy for all tree species classes. This research implies the same data source and the classification algorithm cannot provide the best classification results for different land cover classes. It is necessary to develop a comprehensive classification procedure using an expert-based approach or hierarchical-based classification approach that can employ specific data variables and algorithm for each tree species class.
Accurately mapping crop area using coarse spatial resolution remote sensing imageries is challenging due to the existence of various spatial heterogeneities. The objective of this study is to analyze ...the accuracy of crop classification and area estimation affected by spatial heterogeneities, especially for sample impurity and landscape heterogeneity. The Normalized Difference Vegetation Index (NDVI) time series calculated from Moderate Resolution Imaging Spectroradiometer (MODIS) MOD09Q1 8-day composites and the derived phenology metrics were used to classify crop areas over Manitoba, Canada. The Classification and Regression Trees (CART) approach was applied in the classification. The Agriculture and Agri-Food Canada (AAFC) Land Cover Dataset with 30m spatial resolution was used as the base map to determine the study regions and training and validation samples. The results allowed to conclude that: (1) the classification accuracy of MODIS imagery is sensitive to both sample impurity and landscape heterogeneity. Purity limitations in samples can have a large impact on the classification accuracy. Regions with more homogenous pixels are more likely to be accurately classified and vice versa; (2) the crop area estimation error is less sensitive to sample impurity. It is not only determined by the purity of training samples but also by the actual purity condition of the crop type. The purest training sample group does not correspond well with the lowest error; (3) the impact of configurational heterogeneity on the area estimation is more significant than that of the compositional heterogeneity. Overall, both the sample impurity and landscape heterogeneities can largely affect the classification accuracy while only configurational heterogeneity has significant influence on crop area estimation.
•ZiYuan-3 multispectral and stereo imagery was used for mapping urban vegetation types.•Hierarchy-based classifier provided better classification results than random forest.•Identification of optimal ...variables for each tree node is needed to improve classification.•Incorporation of canopy height features into spectral data significantly improve vegetation classification.•Selection of proper variables plays more important roles than a classifier in improving classification.
Urban vegetation has important impacts on urban heat island, human living environments and even quality of life. The areal increase of urban vegetation has great contribution in achieving Sustainable Development Goals (SDGs) of United Nations. It is needed to accurately extract different urban vegetation types using high spatial resolution images, but the limitation of remotely sensed data and complexity of urban landscapes make it challenging. This research aims to explore the integration of multispectral and stereo imagery with high spatial resolution for vegetation classification in the urban landscape in East China. A hierarchy-based classifier based on optimization of selected variables in each tree node is developed to conduct urban vegetation classification through incorporation of canopy height features into spectral and textural data. The results show that use of canopy height features improved overall classification accuracy of 4.6% comparing with the dataset without use of canopy height features. The proposed hierarchy-based classifier can further improve the vegetation classification accuracy by 3% comparing with random forest. This research indicates that selection of proper variables from different source data, especially canopy height features, plays important roles in improving urban vegetation classification. This research provides a new insight for accurate urban vegetation classification using a hierarchy-based classification approach based on integration of spectral, spatial and canopy features.
Many studies have investigated the effects of spectral and spatial features of remotely sensed data and topographic characteristics on land-cover and forest classification results, but they are ...mainly based on individual sensor data. How these features from different kinds of remotely sensed data with various spatial resolutions influence classification results is unclear. We conducted a comprehensively comparative analysis of spectral and spatial features from ZiYuan-3 (ZY-3), Sentinel-2, and Landsat and their fused datasets with spatial resolution ranges from 2 m, 6 m, 10 m, 15 m, and to 30 m, and topographic factors in influencing land-cover classification results in a subtropical forest ecosystem using random forest approach. The results indicated that the combined spectral (fused data based on ZY-3 and Sentinel-2), spatial, and topographical data with 2-m spatial resolution provided the highest overall classification accuracy of 83.5% for 11 land-cover classes, as well as the highest accuracies for almost all individual classes. The improvement of spectral bands from 4 to 10 through fusion of ZY-3 and Sentinel-2 data increased overall accuracy by 14.2% at 2-m spatial resolution, and by 11.1% at 6-m spatial resolution. Textures from high spatial resolution imagery play more important roles than textures from medium spatial resolution images. The incorporation of textural images into spectral data in the 2-m spatial resolution imagery improved overall accuracy by 6.0–7.7% compared to 1.1–1.7% in the 10-m to 30-m spatial resolution images. Incorporation of topographic factors into spectral and textural imagery further improved overall accuracy by 1.2–5.5%. The classification accuracies for coniferous forest, eucalyptus, other broadleaf forests, and bamboo forest can be 85.3–91.1%. This research provides new insights for using proper combinations of spectral bands and textures corresponding to specifically spatial resolution images in improving land-cover and forest classifications in subtropical regions.
•A uniform procedure to map Eucalyptus using medium and high spatial images.•The novel procedure incorporates four robust variables and random forest method.•The procedure demonstrates a good ...performance at an experimental site.•Applying this procedure to two test sites shows over 94% of overall accuracy.
Accurate and up-to-date mapping of Eucalyptus plantations is essential for assessing their effects on soil, hydrology, and biodiversity. However, no proper approaches are available for extracting such data on Eucalyptus. This study proposes a uniform procedure for mapping Eucalyptus plantations based on fused medium-high spatial resolution satellite datasets. To develop this procedure, a total of 810 variables were extracted from the fused images of an experimental site, the Gaofeng Forest Farm in Guangxi, China. Six key variables were determined by using the Z-statistic and random forest methods. A total of 123 scenarios were then designed based on the key variables and three decision tree classifiers. Potential robust variables were determined through a comparative analysis of accuracy assessments from all scenarios. All the designed procedures were tested at a second site, Xingye County, Guangxi, using the same data processing methodology. The uniform procedure was finally determined through a comparative analysis of accuracy assessments from all the scenarios at both sites. The proposed procedure was then applied to a third study site, Yunxiao County, Fujian, China, to examine its transferability. Results from the Gaofeng site showed that the first short-wave infrared band (SWIR1) and the normalized difference index with near-infrared and green bands (NDNG) were two independent spectral-based variables for Eucalyptus delineation. Scenarios comprising mean texture with a window size of 15 (MEAN15), homogeneity with a window size of 5 (HOM5), or ratio of number of pixels where homogeneity based on window size 5 is greater than threshold T to the total number of pixels in the target segment polygon (HOM5RT) performed significantly greater than scenarios without these variables (α = 0.05). Scenarios with HOM5RT performed significantly greater than those with HOM5 (α = 0.05), and scenarios with random forest (RF) classifier performed greater than those with the other two classifiers. The results from the Xingye site were highly consistent with those from the Gaofeng site. Thus, the final uniform procedure containing these four optimal variables (SWIR1, NDNG, MEAN15, and HOM5RT) and RF classifier was determined. An additional independent test at the Yunxiao site showed an overall accuracy of 96.01% and a high spatial agreement between the delineated Eucalyptus and observed distributions, indicating that the proposed procedure has good transferability. This research implies that the proposed procedure is robust and can be used in other subtropical regions to map Eucalyptus distributions.
In the past several decades, drought events have occurred frequently around the world. However, research on the propagation of drought events has not been adequately explored. This study investigated ...the drought propagation process from meteorological drought to agricultural drought (PMAD) and from meteorological drought to hydrological drought (PMHD) using a 72-year reanalysis dataset in the tropical Lancang–Mekong River Basin. Firstly, we used a new method—Standardized Drought Analysis Toolbox—to construct drought indices. Then, a linear method (Pearson correlation analysis) and a nonlinear method (mutual information) were used to investigate the drought propagation process. Cross-wavelet analysis and wavelet coherence analysis were employed to explore the statistical relationship among the three drought types. Finally, the random forest method was applied to quantify the major factors in drought response time (DRT). The results revealed the following: (1) both linear and nonlinear methods exhibited strong temporal and spatial consistency for both PMAD and PMHD, with linear relationships being stronger than nonlinear ones. (2) The DRTs of PMAD and PMHD were around 1–2 months and 3–5 months, respectively. Significant differences existed in the DRT between the dry season and the rainy season. (3) A divergent spatial pattern of the proportion of DRT was observed between PMAD and PMHD. (4) Significant statistical correlations between meteorological drought and agricultural drought and between meteorological drought and hydrological drought were observed in specific periods for each sub-region; (5) Hydrometeorological factors contributed the most to DRT, followed by terrain factors and the land cover types. The findings of this study deepened our understanding of the spatial–temporal relationship of multiple drought propagation types in this transboundary river basin.
Tree species distribution mapping using remotely sensed data has long been an important research area. However, previous studies have rarely established a comprehensive and efficient classification ...procedure to obtain an accurate result. This study proposes a hierarchical classification procedure with optimized node variables and thresholds to classify tree species based on high spatial resolution satellite imagery. A classification tree structure consisting of parent and leaf nodes was designed based on user experience and visual interpretation. Spectral, textural, and topographic variables were extracted based on pre-segmented images. The random forest algorithm was used to select variables by ranking the impact of all variables. An iterating approach was used to optimize variables and thresholds in each loop by comprehensively considering the test accuracy and selected variables. The threshold range for each selected variable was determined by a statistical method considering the mean and standard deviation for two subnode types at each parent node. Classification of tree species was implemented using the optimized variables and thresholds. The results show that (1) the proposed procedure can accurately map the tree species distribution, with an overall accuracy of over 86% for both training and test stages; (2) critical variables for each class can be identified using this proposed procedure, and optimal variables of most tree plantation nodes are spectra related; (3) the overall forest classification accuracy using the proposed method is more accurate than that using the random forest (RF) and classification and regression tree (CART). The proposed approach provides results with 3.21% and 7.56% higher overall land cover classification accuracy and 4.68% and 10.28% higher overall forest classification accuracy than RF and CART, respectively.
Evapotranspiration (ET) plays a crucial role in water balance within the global hydrological cycle. Timely assessment of ET products can provide the scientific basis for quantitative analysis of ...hydrological cycle processes and water resources assessment. In this paper, four high spatial resolution remote sensing ET products—the Moderate-resolution Imaging Spectroradiometer global terrestrial evapotranspiration product (MOD16), the ET product based on Penman–Monteith–Leuning equation version 2 (PML-V2), the ET product based on the Breathing Earth System Simulator (BESS) and the ET product of the Global LAnd Surface Satellite (GLASS)—were firstly assessed using the eddy covariance (EC) of different vegetation types in the Lancang–Mekong River Basin (LMRB). To fully assess the performances of these four products, spatiotemporal inter-comparisons and literature comparisons were also conducted across different climatic zones. The results are summarized as follows: (1) MOD16 does not perform well as compared to the other three products, with its Root Mean Square Error (RMSE) being higher than GLASS, PML-V2 and BESS, which are approximately 0.47 mm/8-day, 0.66 mm/8-day, and 0.90 mm/8-day, respectively; (2) the performance of each product varies across different vegetation types, and even within the same climate zone. PML-V2 performs best in evergreen broadleaf forests, BESS performs best in deciduous broadleaf forests and croplands, and GLASS performs best in shrubs, grasslands and mixed vegetation; (3) each product can well reflect the spatial difference brought by topography, climate and vegetation over the entire basin but all four ET products do not show either a consistent temporal trend or a uniform spatial distribution; (4) ET ranges of these four products over LMRB are consistent with previous literature in evergreen broadleaf forests, deciduous broadleaf forests, needleleaf forests and mixed forests in other regions with the same climate zones, but they show great differences in croplands, grasslands and shrubs. This study will contribute to improving our understanding of these four ET products in the different climatic zones and vegetation types over LMRB.
Tree species distribution is valuable for forest resource management. However, it is a challenge to classify tree species in subtropical regions due to complex landscapes and limitations of remote ...sensing data. The objective of this study was to propose a modified hierarchy-based classifier (MHBC) by optimizing the classification tree structures and variable selection method. Major steps to create an MHBC include automatic determination of classification tree structures based on the
Z
-score algorithm, selection and optimization of variables for each node, and classification using the optimized model. Experiments based on the fusion of Gaofen-1/Ziyuan-3 panchromatic (GF-1/ZY-3 PAN) and Sentinel-2 multispectral (MS) data indicated that (1) the MHBC provided overall classification accuracies of 85.19% for Gaofeng Forest Farm in China’s southern subtropical region and 94.4% for Huashi Township in China’s northern subtropical region, which had higher accuracies than random forest (RF) and classification and regression tree (CART); (2) critical variables for each class can be identified using the MHBC, and optimal variables of most nodes are spectral bands and vegetation indices; (3) compared to results from RF and CART, MHBC mainly improved the accuracies of the lower levels of classification tree structures (difficult classes to separate). The novelty in using MHBC is its simple and practical operation, easy-to-understand, and visualized variables that were selected in each node of the automatically constructed hierarchical trees. The robust performance of MHBC implies the potential to apply this approach to other sites for accurate classification of forest types.
Forest canopy height is one of the important forest parameters for accurately assessing forest biomass or carbon sequestration. ICESat-2 ATLAS provides the potential for retrieval of forest canopy ...height at global or regional scale, but the current canopy height product (ATL08) has coarse resolution and high uncertainty compared to airborne LiDAR-derived canopy height (hereafter ALCH) in mountainous regions, and is not ready for such applications as biomass modeling at finer scale. The objective of this research was to explore the approach to accurately retrieve canopy height from ATLAS data by incorporating an airborne-derived digital terrain model (DTM) and a data-filtering strategy. By linking ATLAS ATL03 with ATL08 products, the geospatial locations, types, and (absolute) heights of photons were obtained, and canopy heights at different lengths (from 20 to 200 m at 20-m intervals) of segments along a track were computed with the aid of airborne LiDAR DTM. Based on the relationship between the numbers of canopy photons within the segments and accuracy of ATLAS mean canopy height compared to ALCH, a filtering method for excluding a certain portion of unreliable segments was proposed. This method was further applied to different ATLAS ground tracks for retrieval of canopy heights and the results were evaluated using corresponding ALCH. The results show that the incorporation of high-precision DTM and ATLAS products can considerably improve the retrieval accuracy of forest canopy height in mountainous regions. Using the proposed filtering approach, the correlation coefficients (r) between ATLAS canopy height and corresponding ALCH were 0.61–0.91, 0.65–0.92, 0.68–0.94 for segment lengths of 20, 60, and 100 m, respectively; RMSE were 1.90–4.35, 1.55–3.63, and 1.34–3.23 m for the same segment lengths. The results indicate the necessity of using high-precision DTM and using the proposed filtering method to retrieve accurate canopy height from ICESat-2 ATLAS in mountainous regions with dense forest cover and complex terrain conditions.
●ICESat-2 ATLAS ATL08 product has high uncertainty in mountainous regions;●Combination of precise DTM and ATL08 is need to accurately retrieve forest canopy height;●Screening out unreliable segments is needed to ensure the quality of retrieved forest canopy height;●Use of retention rate is an effective filtering method to remove poor quality of canopy photons
PDF
