Xiaoguang Jiang

Comparison of artificial neural networks and support vector machine classifiers for land cover classification in Northern China using a SPOT-5 HRG image

This article presents a sufficient comparison of two types of advanced non-parametric classifiers implemented in remote sensing for land cover classification. A SPOT-5 HRG image of Yanqing County, Beijing, China, was used, in which agriculture and forest dominate land use. Artificial neural networks (ANNs), including the adaptive backpropagation (ABP) algorithm, Levenberg–Marquardt (LM) algorithm, Quasi-Newton (QN) algorithm and radial basis function (RBF) were carefully tested. The LM–ANN and RBF–ANN, which outperform the other two, were selected to make a detailed comparison with support vector machines (SVMs). The experiments show that those well-trained ANNs and SVMs have no significant difference in classification accuracy, but the SVM usually performs slightly better. Analysis of the effect of the training set size highlights that the SVM classifier has great tolerance on a small training set and avoids the problem of insufficient training of ANN classifiers. The testing also illustrates that the ANNs and SVMs can vary greatly with regard to training time. The LM–ANN can converge very quickly but not in a stable manner. By contrast, the training of RBF–ANN and SVM classifiers is fast and can be repeatable.

Modeling of Day-to-Day Temporal Progression of Clear-Sky Land Surface Temperature

This letter presents a method to calculate the width ω over the half-period of the cosine term in a diurnal temperature cycle (DTC) model. ω deduced from the thermal diffusion equation (TDE) is compared with ω obtained from solar geometry. The results demonstrate that ω deduced from the TDE describes the shape of the DTC model more adequately around sunrise and the time of maximum temperature than ω obtained from solar geometry. Additionally, taking into account the physical continuity of land surface temperature (LST) variation, a day-to-day temporal progression (DDTP) model of LST is developed to model several days of DTCs. The results indicate that the DDTP model fits in situ [or Spinning Enhanced Visible and Infrared Imager (SEVIRI)] LST well with a root-mean-square error (RMSE) less than 1 K. Compared with the DTC model, the DDTP model slightly increases the quality of LST fits around sunrise. Assuming that only six LST measurements corresponding to the NOAA/AVHRR and MODIS overpass times for each day are available, several days of DTCs can be predicted by the DDTP model with an RMSE less than 1.5 K.

Spectral characteristics and feature selection of hyperspectral remote sensing data

Hyperspectral remote sensing data with bandwidth of nanometre (nm) level have tens or even several hundreds of channels and contain abundant spectral information. Different channels have their own properties and show the spectral characteristics of various objects in image. Rational feature selection from the varieties of channels is very important for effective analysis and information extraction of hyperspectral data. This paper, taking Shunyi region of Beijing as a study area, comprehensively analysed the spectral characteristics of hyperspectral data. On the basis of analysing the information quantity of bands, correlation between different bands, spectral absorption characteristics of objects and object separability in bands, a fundamental method of optimum band selection and feature extraction from hyperspectral remote sensing data was proposed.

Comparison of land surface temperatures from MSG-2/SEVIRI and Terra/MODIS

Abstract. We evaluate the land surface temperature (LST) generated from the spinning enhanced visible and infrared imager (SEVIRI) onboard the MSG-2 satellite, which was retrieved using the split-window method where the land surface emissivity (LSE) was estimated from the day/night temperature-independent spectral indices-based method. The SEVIRI-derived LST was compared with the MODIS-derived LST extracted from the MOD11B1 V5 product during 7 clear-sky days. The results show that (1) discrepancies exist between the two LST products, with a maximum average difference of 4.9 K; (2) these differences are considered to be time-dependent, since higher discrepancies are observed during the daytime; (3) these differences are land-cover dependent, e.g., bare areas generally present larger differences than vegetated areas; and (4) these differences are inversely proportional to view zenith angle differences. Finally, the main sources of LST differences are investigated and identified in terms of LSE, instrumental noise equivalent temperature difference ( NE Δ T ), and misregistration of the two LST products. The LST differences arising from NE Δ T and misregistration are within 0.4 K. Therefore, these discrepancies may mainly result from errors in LSE, which are caused primarily by the atmospheric correction error for the SEVIRI-derived LST.

A generalized split-window algorithm for land surface temperature estimation from MSG-2/SEVIRI data

This paper aims to determine land surface temperature (LST) using data from a spinning enhanced visible and infrared imager (SEVIRI) on board Meteosat Second Generation 2 (MSG-2) by using the generalized split-window (GSW) algorithm. Coefficients in the GSW algorithm are pre-determined for several overlapping sub-ranges of the LST, land surface emissivity (LSE), and atmospheric water vapour content (WVC) using the data simulated with the atmospheric radiative transfer model MODTRAN 4.0 under various surface and atmospheric conditions for 11 view zenith angles (VZAs) ranging from 0° to 67°. The results show that the root mean square error (RMSE) varies with VZA and atmospheric WVC and that the RMSEs are within 1.0 K for the sub-ranges in which the VZA is less than 30° and the atmospheric WVC is less than 4.25 g cm−2. A sensitivity analysis of LSE uncertainty, atmospheric WVC uncertainty, and instrumental noise (NEΔT) is also performed, and the results demonstrate that LSE uncertainty can result in a larger LST error than other uncertainties and that the total error for the LST is approximately 1.21 and 1.45 K for dry atmosphere and 0.86 and 2.91 K for wet atmosphere at VZA = 0° and at VZA = 67°, respectively, if the uncertainty in the LSE is 1% and that in the WVC is 20%. The GSW algorithm is then applied to the MSG-2 – SEVIRI data with the LSE determined using the temperature-independent spectral indices method and the WVC either determined using the measurements in two split-window channels or interpolated temporally and spatially using European Centre for Medium Range Weather Forecasting (ECMWF) data. Finally, the SEVIRI LST derived in this paper (SEVIRI LST1) is evaluated through comparisons with the SEVIRI LST provided by the land surface analysis satellite applications facility (LSA SAF) (SEVIRI LST2) and the Moderate Resolution Imaging Spectroradiometer (MODIS) LST product (MOD11B1 LST product). The results show that more than 80% of the differences between SEVIRI LST1 and SEVIRI LST2 are within 2 K, and approximately 70% of the differences between SEVIRI LST1 and MODIS LST are within 4 K. Furthermore, compared to MODIS LST, for four specific areas with different land surfaces, our GSW algorithm overestimates the LST by up to 1.0 K for vegetated surfaces and by 1.3 K for bare soil.

Detection of Coal Fire Dynamics and Propagation Direction from Multi-Temporal Nighttime Landsat SWIR and TIR Data: A Case Study on the Rujigou Coalfield, Northwest (NW) China

Coal fires are common and serious phenomena in most coal-producing countries in the world. Coal fires not only burn valuable non-renewable coal reserves but also severely affect the local and global environment. The Rujigou coalfield in Shizuishan City, Ningxia, NW China, is well known for being a storehouse of anthracite coal. This coalfield is also known for having more coal fires than most other coalfields in China. In this study, an attempt was made to study the dynamics of coal fires in the Rujigou coalfield, from 2001 to 2007, using multi-temporal nighttime Landsat data. The multi-temporal nighttime short wave infrared (SWIR) data sets based on a fixed thresholding technique were used to detect and monitor the surface coal fires and the nighttime enhanced thematic mapper (ETM+) thermal infrared (TIR) data sets, based on a dynamic thresholding technique, were used to identify the thermal anomalies related to subsurface coal fires. By validating the coal fires identified in the nighttime satellite data and the coal fires extracted from daytime satellite data with the coal fire map (CFM) manufactured by field survey, we found that the results from the daytime satellite data had higher omission and commission errors than the results from the nighttime satellite data. Then, two aspects of coal fire dynamics were analyzed: first, a quantitative analysis of the spatial changes in the extent of coal fires was conducted and the results showed that, from 2001 to 2007, the spatial extent of coal fires increased greatly to an annual average area of 0.167 km2; second, the spreading direction and propagation of coal fires was analyzed and predicted from 2001 to 2007, and these results showed that the coal fires generally spread towards the north or northeast, but also spread in some places toward the east.

The cross-calibration of CBERS-02B/CCD visible-near infrared channels with Terra/MODIS channels

Radiometric calibration is a prerequisite to quantitative remote sensing, and its accuracy has a direct impact on the reliability and accuracy of the quantitative application of remote-sensing data. This article attempts to cross-calibrate the visible and near-infrared channels 1 (450–520 nm), 2 (520–590 nm), 3 (630–690 nm), and 4 (770–890 nm) of the China–Brazil Earth Resources Satellite (CBERS)-02B/charge-coupled device (CCD) camera with channels 3 (459–479 nm), 4 (545–565 nm), 1 (620–670 nm), and 2 (841–876 nm) of the Terra/Moderate Resolution Imaging Spectroradiometer (MODIS), respectively. The radiative transfer modelling (RTM) method is implemented using the CBERS-02B/CCD and Terra/MODIS measurements at the Dunhuang radiometric calibration site in October 2008. The differences between the cross-calibration and vicarious calibration techniques for channels 1, 2, 3, and 4 are 11.15%, –0.89%, 2.88%, and 11.28%, respectively. Furthermore, the accuracies of the cross-calibration coefficients of CBERS-02B/CCD are analysed, wherein the results indicate that the calibration uncertainty is within 5%.

An analysis of the knife-edge method for on-orbit MTF estimation of optical sensors

Modulation Transfer Function (MTF) is a widely used parameter to assess the on-orbit performance of satellite optical sensors. As one of the most widely used on-orbit MTF estimation methods, the target of knife-edge method is relatively easy to deploy or select, and its data processing is straightforward and can achieve reasonable estimation results. Unfortunately, its estimation accuracy is greatly influenced by many factors and not enough emphasis has been placed on the quantitative analysis of their influences, which limits the creditability and application of estimation result. Therefore, this paper focuses on the implementation and accuracy analysis of the knife-edge method. It begins with a discussion on the necessity and importance of carrying out on-orbit MTF estimation, and the advantages and unresolved problems of the knife-edge method are also analysed. Secondly, the principle, requirements of target and basic data processing steps of the knife-edge method are presented. Thirdly, the influences of major factors affecting the accuracy of MTF estimation result based on the knife-edge method is emphasized using theory analysis and simulation experiments. The analysed factors include atmosphere, the incline angle of target, the contrast and random noise of target, the edge detection accuracy, and the Edge Response Function (ERF) curve-fitting model. The analysis results can provide pivotal guidelines to implement on-orbit MTF estimation based on the knife-edge method. The overall accuracy of on-orbit MTF estimation based on the knife-edge method is also explored using simulation experiments, which provide statistics about the accuracy of MTF estimation results. Finally, an experiment using three knife-edge targets in SPOT-5/Pan image of Dalian airport with different contrasts, signal-to-noise ratios and incline angles is carried out, and its relatively consistent results strongly support the validity of the knife-edge method.

An Improved Algorithm for Retrieving Land Surface Emissivity and Temperature From MSG-2/SEVIRI Data

This paper presents an improved algorithm for simultaneously retrieving both land surface emissivity (LSE) and land surface temperature (LST) using data from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) on board the MSG-2 satellite. First, the temperature-independent spectral index-based method for LSE retrieval is reviewed and improved in terms of three aspects: atmospheric correction, fitting of the bidirectional reflectivity model, and retrieval of the LSE in SEVIRI channel 10. Then, the generalized split-window method with seven unknown coefficients is used to derive the LST. Finally, this improved algorithm is applied to several MSG-2/SEVIRI data sets over a study area with geospatial coverage of latitude 30 ° N-45 ° N and longitude 15 ° W-15 ° E, and using detailed cases, the modifications to the original LSE/LST retrieval methods are shown to be effective and reasonable. In addition, the SEVIRI-derived LSTs are cross-validated primarily using the Moderate Resolution Imaging Spectroradiometer-derived validated LST data extracted from the MOD11B1 product on two clear-sky days (August 22, 2009 and July 3, 2008). The validation results indicate that more than 70% of the differences are within 2.5 K and that the LST differences tend to be lower at night than in the day, which may result from the homogeneous thermal conditions at night.

Analysing the vegetation cover variation of China from AVHRR-NDVI data

In this paper, the characteristics of vegetation cover and variation in China have been studied by using the AVHRR NDVI time‐series data from 1981 to 2001. The Harmonic Analysis of Time Series (HANTS) method was successfully applied to eliminate the clouds on remote sensing data and reconstruct cloud‐free time series images. Then, the Fourier components of NDVI time series data were calculated. Finally, the physical meaning of Fourier components was analysed, and the relationship between Fourier components and land vegetation cover variation was investigated. The mean NDVI, or zeroth‐order harmonic, indicates overall vegetation cover level. The first harmonics of the HANTS summarizes the amplitude and phase of annual values of NDVI data, and the second harmonics of the HANTS summarizes those of biannual values of NDVI data. The amplitude of the first harmonic indicates the variability of vegetation productivity over the year. The phase of the first harmonic summarizes the timing of vegetation green‐up, while the second harmonic indicates the strength and timing of biannual vegetation cover variation. The Fourier components calculated by HANTS algorithm reveal the vegetation distribution and growing cycle characteristics. The physical meaning of Fourier components are significant to the land‐surface vegetation variation study of China. The methodology proposed in this paper is an effective method for the processing, analysis and application of long‐time‐series remote sensing data.