Search

This study aimed to propose a new convolutional neural network (CNN)-based model for spatial resolution enhancement of Landsat images without finer resolution bands for training the CNN-based model. Its performance was assessed using Landsat 8 images from eight different regions that were dominated by various topographic features, and land use and land cover types by comparison with bicubic (BIC), nearest neighbor (NN) and Lanczos (LZ) resampling. It was found that for spatial resolution enhancement of 30 m resolution images to 15 m, 10 m and 7.5 m, the proposed model increased the average peak-signal-to-noise-ratio (PSNR) values by 1.7% to 11.5% compared with the compared methods. The PSNR increases were statistically significantly different from zero at the significant level of.05, but the improvement decreased as the spatial resolution of the input images became finer. Moreover, the deeper the CNN model, the better the performance, but after nine layers, the gain of performance slowed down. This indicates that the proposed algorithm is promising for spatial resolution enhancement of optical images without input of finer spatial resolution images. [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]

Spectral reflectance distortions caused by terrain and solar illumination seriously reduce the accuracy of mapping forest carbon density, especially in mountainous regions. Many models have been developed for mitigating or eliminating the terrain effects on the quality of remote sensing images in hilly and mountainous areas. However, these models usually use global parameters, which may lead to overcorrections for regions with poor illumination and steep slopes. In this study, we present a local parameter estimation (LPE) method based on a pixel-moving window for topographic correction (TC), which can be considered as a general optimization framework for most semiempirical TC models. We set seven kernel sizes for the presented framework, which are 15 pixels, 25 pixels, 50 pixels, 100 pixels, 250 pixels, 500 pixels, and 1000 pixels, respectively. The proposed method was then applied to four traditional TC models, Minnaert (MIN), C Correction (CC), Sun Canopy Sensor + C (SCSC) and Statistical Empirical Correction (SEC), to form four new TC models. These new models were used to estimate forest carbon density of a mountainous area in Southern China using field plot data and a Landsat 8 image. Four evaluation methods, including correlation analysis, the stability of land covers, comparison of reflectance between sunlit and shaded slopes, and accuracy assessment of forest carbon density, were employed to evaluate the contributions of moving window sizes, and assess the performance of the TC models for forest carbon density estimation. The results show that the four TC models with LPE perform much better than the traditional TC models in reducing the topographic effects and improving the estimation accuracy of forest carbon density for the study area. Among the traditional TC models, SEC performs slightly better than SCSC, CC, and MIN. Therefore, the SEC-based model with LPE, that is, LPE-SEC, gets greater R2 and smaller relative RMSE values in estimating forest carbon density than other models. Moreover, all the means of the predicted forest carbon density values fall in the confidence interval of the validation data at a significant level of 0.05. Overall, this study implies that the proposed method with LPE provides great potential to improve the performance of TC and forest carbon density estimation for the study area. It is expected that the improved TC method can be applied to other mountainous areas to improve the quality of remotely sensed images. [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]

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]