Search

Deep learning has substantially pushed forward remote sensing image change detection through extracting discriminative hierarchical features. However, as the increasingly high-resolution remote sensing images have abundant spatial details but limited spectral information, the use of conventional backbone networks would give rise to blurry boundaries between different semantics among hierarchical features. This explains why most false alarms in the final predictions distribute around change boundaries. To alleviate the problem, we pay attention to feature refinement and propose deep learning networks that deliver improved separability (ISNet). Our ISNet reaps the advantages from two strategies applied to refining bitemporal feature hierarchies: 1) margin maximization that clarifies the gap between changed and unchanged semantics and 2) targeted arrangement of attention mechanisms that direct the use of channel attention (CA) and spatial attention (SA) for highlighting semantic and positional information, respectively. Specifically, we insert CA modules into share-weighted backbone networks to facilitate semantic-specific feature extraction. The semantic boundaries in the extracted bitemporal hierarchical features are then clarified by margin maximization modules, followed by SA modules to enhance positional change responses. A top–down fusion pathway makes the final refined features cover multiscale representations and have strong separability for remote sensing image change detection. Extensive experimental evaluations demonstrate that our ISNet achieves state-of-the-art performance on the LEVIR-CD, SYSU-CD, and Season-Varying datasets in terms of overall accuracy (OA), Intersection-of-Union (IoU), and F1 score. Code is available at https://github.com/xingronaldo/ISNet. [ABSTRACT FROM AUTHOR]

Forests play an important role in global ecosystems, but natural disasters such as drought and disease and pest-induced damages often affect growth of trees and even lead to tree death. There have been various vegetation indices (VIs) developed to detect the damages of trees using remote sensing technologies. However, developing an effective and accurate VI for detecting forest damages at an earlier stage is still challenging. In this study, a total of 32 target trees and 17 reference trees were selected from one of Chinese fir plantations that account for 24% of the afforested area in China. The barks of the target trees at 20 cm and 70 cm above the ground were respectively peeled off in August of 2016 to let the trees die. The hyperspectral data were collected from the selected branches with leaves using a high-performance spectrometer HR-1024i (Spectra Vista Corporation) with 1024 bands covering the region of wavelengths from 350 nm to 2500 nm. The spectral data collection was conducted from both the target and reference trees by non-imaging monthly from August of 2016 to February of 2017 until the target trees died. Based on statistical analysis, a Green to Red Region Spectral anGle Index (GRRSGI) was proposed and compared with other ten widely used VIs to detect the early dying process of the damaged trees. The results showed that compared with the reference trees, the damaged trees reduced the reflectance peaks of green bands and the absorption valleys of red bands and the frequency distributions of spectral reflectance from the damaged trees turned from statistically normal to non-normal in the region of 550–640 nm. This indicated that the spectral bands in the region of green to red bands were sensitive to the dying process of the target trees. Moreover, the performance of the VIs for detecting the dying process of the damaged trees varied depending on the time period after the trees damaged. The proposed GRRSGI based on the spectral bands of 550–640 nm led to the performance similar to all other VIs after three months of the damages but significantly better than them in the first two months after the damages. Except for GRRSGI, the Green-Red Spectral Area Index (GRSAI) also performed better than other VIs. Compared with GRSAI, however, the proposed GRRSGI performed slight better when the spectral data of 550–640 nm were re-sampled using the spectral intervals of 5–40 nm with an increase of 5 nm and significantly better when the re-sampling of the hyperspectral data was conducted in the region of 550–670 nm. This implied the proposed GRRSGI provided the greater potential of detecting the early dying process of the damage trees. [ABSTRACT FROM AUTHOR]

Remotely sensed data have been widely used for classification of land-use and land-cover (LULC) types. However, classifying different forest types in Northeast China using satellite images is still a great challenge because of the similar spectral reflectances of different tree species. The differences of vegetation phenological characteristics provide the potential of classifying the types using time series of spectral variables derived from images. In this study, time series of the normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI) from Moderate Resolution Imaging Spectroradiometer (MODIS) images obtained in 2012 for Northeast China were used to calculate various phenological metrics and to further derive amplitude and phase information of harmonic components using Fourier transforms. The separability of eight vegetation cover types plus water and built-up areas was then analysed using phenological metrics, and amplitude and phase of harmonic components. Moreover, a phenology-based decision tree classifier was developed to classify the types in this area. Out of 4900 national forest inventory plots, 3700 plots were used to train the decision tree classifier and 1200 plots to assess the accuracy of classification by combining the plots’ observations with the values of a published LULC map that had a higher spatial resolution and accuracy of classification using a window majority rule. In addition, three data sets from different temporal resolution MODIS NDVI and EVI time series and two similarity measures were compared for separability and classification of the types. The results showed that (1) Fourier transforms of NDVI and EVI time series led to the first four harmonic components (including component 0, average annual NDVI, and EVI) that captured the phenological characteristics of the cover types; (2) compared to those using only NDVI, the separability values of the classes using NDVI, amplitude, and phase increased from 1.71 to 1.95, implying the potential improvement of classification; (3) the data set from 10-day NDVI time series had higher classification accuracy than those from 16-day NDVI and EVI time series, although the EVI time series performed slightly better than the NDVI time series at the same temporal resolution; and (4) a classification accuracy of 83.8% with a kappa coefficientof 0.79 was finally obtained. This study implied that this method is applicable for classification of vegetation cover types for large areas. [ABSTRACT FROM AUTHOR]

Current studies on large-scale remotely sensed images are of great national importance for monitoring and evaluating global climate and ecological changes. In particular, real time distributed high-performance visualization and computation have become indispensable research components in facilitating the extraction of remotely sensed image textures to enable mining spatiotemporal patterns and dynamics of landscapes from massive geo-digital information collected from satellites. Remotely sensed images are usually highly correlated with rich landscape features. By exploiting the structures of these images and extracting their textures, fundamental insights of the landscape can be derived. Furthermore, the interdisciplinary collaboration on the remotely sensed image analysis demands multifarious expertise in a wide spectrum of fields including geography, computer science, and engineering. This paper develops a Distributed Computing System for Remotely Sensed Images framework (DCSRI) to support distributed and high performance computing for geospatial images. A new algorithm supporting parallel computing with dynamic workload balance for large images, namely Variogram-based Image Texture Extractor (VITE) for extracting image texture from massive and dynamic digital remotely sensed images is presented. The VITE algorithm is used to represent and transform the original data into image textures. The DCSRI framework has the capacity to perform high performance computing on Linux clusters or supercomputers to address the intensive computing challenges arising from large and multiple images. Advanced web technology is also exploited to enable interactive user experience with prompt visual feedback as well as good interoperability. Users can also dynamically steer the visualization and computation process by adjusting the computing parameters on-the-fly. This system leads to a great reduction of image data and provides useful information for knowledge discovery and digital image classification in a user friendly and computing efficient way. [ABSTRACT FROM AUTHOR]

Forest carbon sinks significantly contribute to mitigation of atmospheric concentrations of carbon dioxide. Thus, estimating forest carbon is becoming important to develop policies for mitigating climate change and trading carbon credits. However, a great challenge is how to quantify uncertainties in estimation of forest carbon. This study investigated uncertainties of mapping aboveground forest carbon due to location errors of sample plots for Lin-An County of China. National forest inventory plot data and Landsat TM images were combined using co-simulation algorithm. The findings show that randomly perturbing plot locations within 10 distance intervals statistically did not result in biased population mean predictions of aboveground forest carbon at a significant level of 0.05, but increased root mean square errors of the maps. The perturbations weakened spatial autocorrelation of aboveground forest carbon and its correlation with spectral variables. The perturbed distances of 800 m or less did not obviously change the spatial distribution of predicted values. However, when the perturbed distances were 1600 m or larger, the correlation coefficients of the predicted values from the perturbed locations with those from the true plot locations statistically did not significantly differ from zero at a level of 0.05 and the spatial distributions became random. [ABSTRACT FROM AUTHOR]

Abstract: A great challenge the US military land managers are often faced with is how to optimize road networks in order to maintain roads (including all roads, trails, and paths) for the purpose of military training and reducing negative impacts on environment. In this study, a methodology was developed to identify superfluous roads for being closed in terms of both sustainable military land carrying capacity and environment for Fort Riley. In this method, Geographic Information Systems (GIS), remote sensing, and landscape analysis technologies were combined to derive various spatial data layers of factors that had significant impacts on both military training and environment. The factors included maintenance cost of roads, road access area, military training intensity, soil erosion, water quality, landscape fragmentation, and noise production. The factors were quantified and normalized. A spatial multicriteria decision was then developed to obtain the weights of the factors, combine the data layers, and derive a priority map of all the roads for being closed. This map summarized the negative and positive impacts of the factors on environment and military land carrying capacity and can provide the US military land managers with useful guidelines and tools for determining superfluous roads in terms of both sustainable military training and environment. It is expected this effort can provide a method to quickly ascertain which roads are most cost-effective for being closed without hindering the mission and at the same time with benefits for environmental protection and thus provide the land managers with a comprehensive analysis and assessment of alternatives at their disposal. [Copyright &y& Elsevier]

This study investigates applications and efficiencies of remotely sensed data and the sensitivity of grid spacing for the sampling and mapping of a ground and vegetation cover factor in a monitoring system of soil erosion dynamics by cokriging with Landsat Thematic Mapper (TM) imagery based on regionalized variable theory. The results show that using image data can greatly reduce the number of ground sample plots and sampling cost required for collection of data. Under the same precision requirement, the efficiency gain is significant as the ratio of ground to image data used varies from 1: 1 to 1: 16. Moreover, we proposed and discussed several modifications to the cokriging procedure with image data for sampling and mapping. First, directly using neighbouring pixels for image data in sampling design and mapping is more efficient at increasing the accuracy of maps than using sampled pixels. Although information among neighbouring pixels might be considered redundant, spatial cross-correlation of spectral variables with the cover factor can provide the basis for an increase in accuracy. Secondly, this procedure can be applied to investigate the appropriate spatial resolution of imagery, which, for sampling and mapping the cover factor, should be 90 m × 90 m - nearly consistent with the line transect size of 100 m used for the ground field survey. In addition, we recommend using the average of cokriging variance to determine the global grid spacing of samples, instead of the maximum cokriging variance. [ABSTRACT FROM AUTHOR]

Abstract: Military training activities disturb ground and vegetation cover of landscapes and increases potential soil erosion. To monitor the dynamics of soil erosion, there is an important need for an optimal sampling design in which determining the optimal spatial resolutions in terms of size of sample plots used for the collection of ground data and the size of pixels for mapping. Given a sample size, an optimal spatial resolution should be cost-efficient in both sampling costs and map accuracy. This study presents a spatial variability-based method for that purpose and compared it with the traditional methods in a study area in which a soil erosion cover factor was sampled and mapped with multiple plot sizes and multi-sensor images. The results showed that the optimal spatial resolutions obtained using the spatial variability-based method were 12 and 20m for years 1999 and 2000, respectively, and were consistent with those using the traditional methods. Moreover, the most appropriate spatial resolutions using the high-resolution images were also consistent with those using ground sample data, which provides a potential to use the high-resolution images instead of ground data to determine the optimal spatial resolutions before sampling. The most appropriate spatial resolutions above were then verified in terms of cost-efficiency which was defined as the product of sampling cost and map error using ordinary kriging without images and sequential Gaussian co-simulation with images to generate maps. [Copyright &y& Elsevier]

In the management of natural resources, multiple variables correlated with each other usually need to be mapped jointly. However, joint mapping and spatial uncertainty analyses are very difficult mainly because of interactions among variables and imperfection of existing methods. There is abundant evidence that considering interactions among variables and spatial information from neighbors can result in improved maps. This study presents a remote sensing-aided method for that purpose. The method is based on the integration of joint sequential co-simulation with Landsat TM image for mapping and polynomial regression for spatial uncertainty analysis. The method was applied to a case study in which ground cover (GC), canopy cover (CC), and vegetation height (VH) were jointly mapped to derive a map of the vegetation cover factor for predicting soil loss. The variance contributions from the variables, their interactions, and the spatial information from neighbors leading to uncertainty of predicted vegetation cover factor were assessed. The results showed that in addition to unbiased maps, this method reproduced the spatial variability of the variables and the spatial correlation among them, and successfully quantified the effect of variation from all the components on the prediction of the vegetation cover factor. [Copyright &y& Elsevier]

The need for timely, spatially, and thematically accurate information regarding forests is increasing because of the key role of forests in the global carbon balance and sustainable social, economic, ecological, and cultural development. While an increasing number of countries in the world already are conducting statistically sound forest inventories, a few for 100 years, inventories in other countries are still lacking, which makes the global information about forest statistics inaccurate. This implies that remote sensing-based approaches are vulnerable to misclassification and inaccurate estimation of forest parameters and changes, such as the recognition of temporarily unstocked forest areas, forest degradation, and species composition, as well as in strictly following the globally adopted definitions. [Extracted from the article]

Tai, X.; Wang, G.; Grecos, C., and Ren, P., 2020. Coastal image classification under noisy labels. In: Jung, H.-S.; Lee, S.; Ryu, J.-H., and Cui, T. (eds.), Advances in Geospatial Research of Coastal Environments. Journal of Coastal Research, Special Issue No. 102, pp. 151-156. Coconut Creek (Florida), ISSN 0749-0208. Remote sensing coastal images commonly suffer from noisy labels, which tend to reduce the benefits of training coastal image classifiers and thus deteriorate the classification performance. This article explores the feasibility of classifying coastal images under noisy labels. The state-of-the-art co-teaching paradigm in computer vision is exploited for processing remote sensing coastal images that are partially incorrectly labeled. The data with noisy labels are hard to learn and induce large loss values for a classifier. In the light of this observation, the co-teaching paradigm trains two identically structured but randomly initialized convolutional neural network (CNN) classifiers with stochastic optimization. Training of the two CNN classifiers is performed on mini-batches in an interactive and iterative manner. Specifically, the data with small loss values of one CNN classifier are used for updating the other CNN classifier in the next training iteration. The two CNN classifiers turn robust to noisy labels through the iterative process of teaching each other. Experimental evaluations on a coastal image dataset, i.e., the Moye Island remote sensing dataset, validate the effectiveness of the proposed coastal image classification strategy. [ABSTRACT FROM AUTHOR]

Yang, Q.; Wang, G.; Zhang, X.; Grecos, C., and Ren, P., 2020. Coastal image captioning. In: Jung, H.-S.; Lee, S.; Ryu, J.-H., and Cui, T. (eds.), Advances in Geospatial Research of Coastal Environments. Journal of Coastal Research, Special Issue No. 102, pp. 145-150. Coconut Creek (Florida), ISSN 0749-0208. Coastal images convey immense semantic information of the corresponding coastal areas. This article presents a preliminary approach to coastal image captioning that describes salient semantic information of coastal images with accurate and meaningful sentences. Specifically, this article exploits a state-of-the-art method referred to as self-critical sequence training for coastal image captioning. Firstly, a convolutional neural network (CNN) produces fixed-length features from the coastal images. Secondly, the fixed-length features are fed into a long short-term memory (LSTM) to generate captions, i.e., the accurate and meaningful sentences. Finally, the LSTM is optimized in the context of reinforcement learning. Experimental evaluation on the Moye island remote sensing dataset validates the effectiveness. [ABSTRACT FROM AUTHOR]

As a cross-disciplinary scientific domain, a great deal of research on landscape geochemistry has been conducted since its establishment. However, the used methods lack the ability of revealing the relationships of landscape geochemistry with other sciences, and the obtained knowledge is thus fragmented and isolated. In this study, the state-of-the-art regarding the applications of relatively new geospatial technologies including fractal theory, geographic information system and remote sensing to landscape geochemistry was reviewed and analyzed to provide deep insights of current research and a roadmap for furthering the development of landscape geochemistry as a cross-disciplinary discipline. The results showed that substantial research on the applications of fractal theory, GIS and RS technologies for analyzing the processes and data of landscape geochemistry has been conducted by using the advantages of the geospatial technologies. However, the great challenges still exist when the geospatial technologies were individually utilized due to the limitations of the technologies themselves and the complexity of landscape geochemistry. In the end, opportunities and challenges for advancing the further studies of landscape geochemistry were discussed in detail and new directions of studying landscape geochemistry using multidisciplinary or integrated approaches to enhance understanding the relationships among the relevant disciplines were suggested. [ABSTRACT FROM AUTHOR]

Quickly and accurately mapping soil properties is critical for agricultural, forestry and environmental management. In this study, a new hyperspectral remote sensing method of soil property prediction was developed and validated in Stipa purpurea dominated alpine grasslands located in Shenzha County of the Qiangtang Plateau, northwestern Qinghai-Tibet Plateau. Hyperspectral data were collected in a total of 67 sample points. At the same time, soil samples were obtained at the locations and soil properties including organic carbon, total nitrogen, total potassium and total phosphorus were measured. The correlations of the soil properties with original bands and enhanced spectral variables derived from both field and satellite hyperspectral data were analyzed. Regression models that explained the relationships were further developed to map the soil properties. The results showed that the stepwise regression models based on the satellite hyperspectral image derived enhanced spectral variables produced reasonable spatial distributions of the soil properties and the relative RMSE values of 68.9, 46.3, 31.4 and 45.5% for soil organic carbon, total nitrogen, total phosphorus and total potassium, respectively. Thus, this study implied that the hyperspectral data based method provided great potential to predict the soil properties. [ABSTRACT FROM AUTHOR]

There is a large amount of remote sensing data available for land use and land cover (LULC) classification and thus optimizing selection of remote sensing variables is a great challenge. Although many methods such as Jeffreys-Matusita (JM) distance and random forests (RF) have been developed for this purpose, the existing methods ignore correlation and information duplication among remote sensing variables. In this study, a novel approach was proposed to improve the measures of potential class separability for the selection of remote sensing variables by taking into account correlations among the variables. The proposed method was examined with a total of thirteen spectral variables from a Gaofen-1 image, three class separability measures including JM distance, transformed divergence and B-distance and three classifiers including Bayesian discriminant (BD), Mahalanobis distance (MD) and RF for classification of six LULC types at the East Dongting Lake of Hunan, China. The results showed that (1) The proposed approach selected the first three spectral variables and resulted in statistically stable classification accuracies for three improved class separability measures. That is, the classification accuracies using three or more spectral variables statistically did not significantly differ from each other at a significant level of 0.05; (2) The statistically stable classification accuracies obtained by integrating MD and BD classifiers with the improved class separability measures were also statistically not significantly different from those by RF; (3) The numbers of the selected spectral variables using the improved class separability measures to create the statistically stable classification accuracies by MD and BD classifiers were much smaller than those from the original class separability measures and RF; and (4) Three original class separability measures and RF led to similar ranks of importance of the spectral variables, while the ranks achieved by the improved class separability measures were different due to the consideration of correlations among the variables. This indicated that the proposed method more effectively and quickly selected the spectral variables to produce the statistically stable classification accuracies compared with the original class separability measures and RF and thus improved the selection of the spectral variables for the classification. [ABSTRACT FROM AUTHOR]

Remote sensing-based methods of aboveground biomass (AGB) estimation in forest ecosystems have gained increased attention, and substantial research has been conducted in the past three decades. This paper provides a survey of current biomass estimation methods using remote sensing data and discusses four critical issues – collection of field-based biomass reference data, extraction and selection of suitable variables from remote sensing data, identification of proper algorithms to develop biomass estimation models, and uncertainty analysis to refine the estimation procedure. Additionally, we discuss the impacts of scales on biomass estimation performance and describe a general biomass estimation procedure. Although optical sensor and radar data have been primary sources for AGB estimation, data saturation is an important factor resulting in estimation uncertainty. LIght Detection and Ranging (lidar) can remove data saturation, but limited availability of lidar data prevents its extensive application. This literature survey has indicated the limitations of using single-sensor data for biomass estimation and the importance of integrating multi-sensor/scale remote sensing data to produce accurate estimates over large areas. More research is needed to extract a vertical vegetation structure (e.g. canopy height) from interferometry synthetic aperture radar (InSAR) or optical stereo images to incorporate it into horizontal structures (e.g. canopy cover) in biomass estimation modeling. [ABSTRACT FROM PUBLISHER]

For the past three decades, ecologists and biogeographers have increasingly incorporated remote sensing and geographical information systems (GISs) to inventory and analyse spatially organized data. Although there are many studies exploring the effects of fine resolution on remote sensing and GIS mapping, there is still a gap on how to identify the most appropriate spatial resolutions for studying landscapes and their structures and dynamics. This study investigated the effects of landscape changes over a 64-year study period at different spatial resolutions using four resampling schemes. The study was conducted on a national wildlife refuge of five land-use and land-cover (LULC) categories using aerial photos recorded in three distant years (1938, 1971, and 2001). This refuge has undergone major landscape changes in the last 64 years. Among the five LULC categories studied, the one that lost the most surface is agriculture; the most gain was made in forest and water. In terms of net change and swapping, agriculture and forest were the most dynamic categories in the National Wildlife Refuge. Our findings showed considerable spatial variability in landscape dynamics at different scales. We specifically observed that hard-classified maps with spatial resolutions of 30 m or finer provided better analysis of landscape dynamics, whereas with soft-classified maps it could go up to 90 m or finer. This implies that there is a range of optimum resolution that could allow for the use of medium-resolution data, such as Landsat, for reliable land change analysis. These findings offer further insight on landscape change analysis at different spatial resolutions and advance our understanding and knowledge on the effect of scale on landscape ecology. [ABSTRACT FROM AUTHOR]

Estimating the net primary production (NPP) of vegetation is essential for eco-environment conservation and carbon cycle research. Remote sensing techniques, combined with algorithm models, have been proven to be promising methods for NPP estimation. High-precision and real-time NPP monitoring in heterogeneous areas requires high spatio-temporal resolution remote sensing data, which are not easy to acquire by single remote sensors, especially in cloudy weather. This study proposes to fuse images of different sensors to provide high spatio-temporal resolution data for NPP estimation in cloud-prone areas. Firstly, the time series Normalized Difference Vegetation Index (NDVI) with a spatial resolution of 30 m and a temporal resolution of 16 days, are obtained by the enhanced spatial and temporal adaptive reflectance fusion model (ESTARFM). Then, the time series NDVI data, combined with meteorological data are input into an improved Carnegie–Ames–Stanford Approach (CASA) model for NPP estimation. This method is validated by a case study of a heavily urbanized area, in the middle reaches of the Yangtze River in China. The results indicate that the NPP estimated by the fused NDVI data has more detailed spatial information than by using the MODIS data. The results show a strong correlation between the actual Landsat8 NDVI and the fused NDVI images, which means that the accuracy of synthetic NDVI images (a 16 day interval and a 30 m resolution) is reliable, and it can provide superior inputs for accurate estimations of a NPP time series. The correlation coefficient (R) and root mean square error between the NPP, based on the fused NDVI and the measured NPP, are 0.66 and 14.280 g C/(m2·yr), respectively, indicating a good consistency. The small discrepancy is caused by the uncertainties of fused NDVI, measurement errors, conversion errors, and other factors in the CASA model. In this study, we achieved NPP with high spatial and temporal resolutions, which can provide higher accuracies of NPP data for analyzing the carbon cycling heavily urbanized areas, compared with similar studies using mono-temporal NPP data. The spatio-temporal fusion technique is an effective way of generating high spatio-temporal resolution images from different sensors, thereby providing enough data for NPP monitoring in urbanized areas. [ABSTRACT FROM AUTHOR]

Land degradation and desertification in arid and semi-arid areas is of great concern. Accurately mapping percentage vegetation cover (PVC) of the areas is critical but challenging because the areas are often remote, sparsely vegetated, and rarely populated, and it is difficult to collect field observations of PVC. Traditional methods such as regression modeling cannot provide accurate predictions of PVC in the areas. Nonparametric constant k-nearest neighbors (Cons_kNN) has been widely used in estimation of forest parameters and is a good alternative because of its flexibility. However, using a globally constant k value in Cons_kNN limits its ability of increasing prediction accuracy because the spatial variability of PVC in the areas leads to spatially variable k values. In this study, a novel method that spatially optimizes determining the spatially variable k values of Cons_kNN, denoted with Opt_kNN, was proposed to map the PVC in both Duolun and Kangbao County located in Inner Mongolia and Hebei Province of China, respectively, using Landsat 8 images and sample plot data. The Opt_kNN was compared with Cons_kNN, a linear stepwise regression (LSR), a geographically weighted regression (GWR), and random forests (RF) to improve the mapping for the study areas. The results showed that (1) most of the red and near infrared band relevant vegetation indices derived from the Landsat 8 images had significant contributions to improving the mapping accuracy; (2) compared with LSR, GWR, RF and Cons_kNN, Opt_kNN resulted in consistently higher prediction accuracies of PVC and decreased relative root mean square errors by 5%, 11%, 5%, and 3%, respectively, for Duolun, and 12%, 1%, 23%, and 9%, respectively, for Kangbao. The Opt_kNN also led to spatially variable and locally optimal k values, which made it possible to automatically and locally optimize k values; and (3) the RF that has become very popular in recent years did not perform the predictions better than the Opt_kNN for the both areas. Thus, the proposed method is very promising to improve mapping the PVC in the arid and semi-arid areas. [ABSTRACT FROM AUTHOR]

Traditional parametric methods for classification of land use and land cover (LULC) types using remote sensing imagery assume a global distribution model and fail to consider local variation of categorical variables. Differently, non-parametric methods do not make any statistical assumptions but are typically sensitive to the sample sizes of training sample data that usually require a high cost to collect in the field. Geostatistical classifiers, such as indicator kriging and simulation, are local variability-based methods that exhibit great potential for image-based classification of LULC types. However, variogram models required are highly sensitive to the spatial configuration of training samples as well as sample size given a study area. Moreover, when a large number of spectral variables are considered into kriging systems, modeling the variograms and cross-variograms would be problematic. To circumvent these issues, this study extended the geostatistical methods from a 2-dimensional geographic space to a m -dimensional image feature space to derive feature-space indicator variograms (FSIVs). Moreover, a novel stochastic simulation classification algorithm, Feature-Space Indicator Simulation (FSIS), was proposed and examined for classification of LULC types in Duolun County located in Inner Mongolia and in Huang-Feng-Qiao (HFQ) forest farm, Hunan of China. In Duolun, six LULC types were involved and in HFQ a complicated forest landscape consisting of nine forest types plus water, built-up area, and agricultural/bare soil, was classified. The classification results of FSIS were compared with another feature-space geostatistical classifier – feature-space indicator kriging (FSIK), a traditional parametric method – maximum likelihood (ML), a widely used nonparametric method – support vector machine (SVM), and a recently popular method – random forest (RF). The results showed that compared with ML, SVM and RF, in both study areas FSIS statistically significantly increased the accuracy of the classifications by 10.0–29.9% for percentage correct and 19.0–47.6% for Kappa statistic. Compared with FSIK, FSIS also improved the classification accuracy but the accuracy increases were relatively smaller with the percentages correct of 3.5% and 7.6% and the Kappa values of 4.6% and 8.6% for Duolun and HFQ, respectively. Moreover, FSIS led to the spatial uncertainties of the classification estimates as the quality measure of the estimates. In addition, the results also demonstrated that FSIVs were sensitive to the within-class heterogeneity but not very much to the size of training samples. Overall, FSIS exhibited the greater potential to improve the classification accuracy of LULC and forest types using remote sensing image. [ABSTRACT FROM AUTHOR]

Estimating individual tree diameters at breast height (DBH) from delineated crowns and tree heights on the basis of airborne light detection and ranging (LiDAR) data provides a good opportunity for large-scale forest inventory. Generally, ground-based measurements are more accurate, but LiDAR data and derived DBH values can be obtained over larger areas for a relatively smaller cost if a right procedure is developed. A nonlinear least squares (NLS) regression is not an appropriate approach to predict the aboveground biomass (AGB) of individual trees from the estimated DBH because both the response variable and the regressor are subject to measurement errors. In this study, a system of compatible individual tree DBH and AGB error-in-variable models was developed using error-in-variable regression techniques based on both airborne LiDAR and field-measured datasets of individual Picea crassifolia Kom. trees, collected in northwestern China. Two parameter estimation algorithms, i.e., the two-stage error-in-variable model (TSEM) and the nonlinear seemingly unrelated regression (NSUR), were proposed for estimating the parameters in the developed system of compatible individual tree DBH and AGB error-in-variable models. Moreover, two model structures were applied to estimate AGB for comparison purposes: NLS with AGB estimation depending on DBH (NLS&DD) and NLS with AGB estimation not depending on DBH (NLS&NDD). The results showed that both TSEM and NSUR led to almost the same parameter estimates for the developed system. Moreover, the developed system effectively accounted for the inherent correlation between DBH and AGB as well as for the effects of measurement errors in the DBH on the predictions of AGB, whereas NLS&DD and NLS&NDD did not. A leave-one-out cross-validation indicated that the prediction accuracy of the developed system of compatible individual tree DBH and AGB error-in-variable models with NSUR was the highest among those estimated by the four methods evaluated, but, statistically, the accuracy improvement was not significantly different from zero. The main reason might be that, except for the measurement errors, other source errors were ignored in the modeling. This study implies that, overall, the proposed method provides the potential to expand the estimations of both DBH and AGB from individual trees to stands by combining the error-in-variable modeling and LiDAR data and improve their estimation accuracies, but its application needs to be further validated by conducting a systematical uncertainty analysis of various source errors in a future study. [ABSTRACT FROM AUTHOR]

Northern China has been long threatened by aeolian desertification. In recent years, all levels of the Chinese government have performed a series of ecological protection and sand control projects. To grasp the implementation effects of these projects and adjust policies in time, it is necessary to understand the process of aeolian desertification quickly and accurately. Remote sensing technologies play an irreplaceable role in aeolian desertification monitoring. In this study, the Zhenglan Banner, which is in the hinterland of the Hunshandake Sandy Land, was considered as the research area. Based on unmanned aerial vehicle (UAV) images, ground survey data, and Landsat images called in Google Earth Engine (GEE), the aeolian desertified land (ADL) in 2000, 2004, 2010, 2015, and 2019 was extracted using spectral mixture analysis. A desertification index (DI) was constructed to evaluate the spatial and temporal dynamics of the ADL in the Zhenglan Banner. Finally, a residual analysis explored the driving forces of aeolian desertification. The results showed that (1) the ADL area in the Zhenglan Banner has been trending downwards over the past 20 years but rebounded from 2004 to 2010; (2) over the past 20 years, the area of slightly, moderately, and severely desertified land has decreased at annual rates of 0.4%, 2.7%, and 3.4%, respectively; (3) human activities had significantly positive and negative impacts on the aeolian desertification trend for 20.0% and 21.0% of the study area, respectively, but not for the rest. This paper explored new techniques for rapid aeolian desertification monitoring and is of great significance for controlling and managing aeolian desertification in this region. [ABSTRACT FROM AUTHOR]

Conventional ground survey data are very accurate, but expensive. Airborne lidar data can reduce the costs and effort required to conduct large-scale forest surveys. It is critical to improve biomass estimation and evaluate carbon stock when we use lidar data. Bayesian methods integrate prior information about unknown parameters, reduce the parameter estimation uncertainty, and improve model performance. This study focused on predicting the independent tree aboveground biomass (AGB) with a hierarchical Bayesian model using airborne LIDAR data and comparing the hierarchical Bayesian model with classical methods (nonlinear mixed effect model, NLME). Firstly, we chose the best diameter at breast height (DBH) model from several widely used models through a hierarchical Bayesian method. Secondly, we used the DBH predictions together with the tree height (LH) and canopy projection area (CPA) derived by airborne lidar as independent variables to develop the AGB model through a hierarchical Bayesian method with parameter priors from the NLME method. We then compared the hierarchical Bayesian method with the NLME method. The results showed that the two methods performed similarly when pooling the data, while for small sample sizes, the Bayesian method was much better than the classical method. The results of this study imply that the Bayesian method has the potential to improve the estimations of both DBH and AGB using LIDAR data, which reduces costs compared with conventional measurements. [ABSTRACT FROM AUTHOR]

Remote sensing–based forest aboveground biomass (AGB) estimation has been extensively explored in the past three decades, but how to effectively combine different sensor data and modeling algorithms is still poorly understood. This research conducted a comparative analysis of different datasets (e.g., Landsat Thematic Mapper (TM), ALOS PALSAR L-band data, and their combinations) and modeling algorithms (e.g., artificial neural network (ANN), support vector regression (SVR), Random Forest (RF), k-nearest neighbor (kNN), and linear regression (LR)) for AGB estimation in a subtropical region under non-stratification and stratification of forest types. The results show the following: (1) Landsat TM imagery provides more accurate AGB estimates (root mean squared error (RMSE) values in 27.7–29.3 Mg/ha) than ALOS PALSAR (RMSE values in 30.3–33.7 Mg/ha). The combination of TM and PALSAR data has similar performance for ANN and SVR, worse performance for RF and KNN, and slightly improved performance for LR. (2) Overestimation for small AGB values and underestimation for large AGB values are major problems when using the optical (e.g., Landsat) or radar (e.g., ALOS PALSAR) data. (3) LR is still an important tool for AGB modeling, especially for the AGB range of 40–120 Mg/ha. Machine learning algorithms have limited effects on improving AGB estimation overall, but ANN can improve AGB modeling when AGB values are greater than 120 Mg/ha. (4) Forest type and AGB range are important factors that influence AGB modeling performance. (5) Stratification based on forest types improved AGB estimation, especially when AGB was greater than 160 Mg/ha, using the LR approach. This research provides new insight for remote sensing-based AGB modeling for the subtropical forest ecosystem through a comprehensive analysis of different source data, modeling algorithms, and forest types. It is critical to develop an optimal AGB modeling procedure, including the collection of a sufficient number of sample plots, extraction of suitable variables and modeling algorithms, and evaluation of the AGB estimates. [ABSTRACT FROM AUTHOR]