Yongming Xu

Study on the estimation of near-surface air temperature from MODIS data by statistical methods

Spatially distributed air temperature is desired for various scientific studies, including climatalogical, hydrological, agricultural, environmental and ecological studies. In this study, empirical models with regard to land cover and spatial scale were introduced and compared to estimate air temperature from satellite-derived land surface temperature and other environmental parameters. Aqua MODIS (Moderate Resolution Imaging Spectroradiometer) data and meteorological data obtained throughout 2005 in the Yangtze River Delta were adopted to develop statistical algorithms of air temperature. Four empirical regression models with different forms and different independent variables resulted in errors ranging from 2.20°C to 2.34°C. Considering the different relationships between air temperature and land surface temperature for different land types, these four models were evaluated and the most proper equation for each land-cover type was determined. The model containing these selected equations gave a slightly improved mean absolute error (MAE) of 2.18°C. Then the spatial scale effect of this empirical model was analysed with observed air temperature and spatially averaged land surface characteristics. The result shows that the estimation error of air temperature tends to be lower as spatial window size increases, suggesting that the model performances are improved by spatially averaging land surface characteristics. Comprehensively considering the accuracy and computational demand, 5 × 5 pixel size is the most favourable window size for estimating air temperature. The validation of the empirical model at 5 × 5 pixel size shows that it achieves an MAE of 1.98°C and an R 2 of 0.9215. This satisfactory result indicates that this approach is proper for estimating air temperature, and spatial window size is an important factor that should be considered when calculating air temperature. It is expected that better accuracy will be achieved if the different weights of pixels at different distances can be set according to high-density micro-meteorological data.

Estimates of carbon fluxes from Poyang Lake wetlands vegetation in the growing season

Poyang Lake is the largest shallow lake wetlands in China, and which vegetation succession is rapid under high changeable hydrological regimes. This study measured the fluxes of carbon dioxide and methane simultaneously by opaque static chamber-gas chromatography technique for typical wetland vegetation ecosystems in the growing season. In view of the advantages both in temporal and spatial, HJ-1 satellite images were chosen as the data source for vegetation cover classification and area estimates. And based on the areas in different vegetation, carbon flux for the entire study area was estimated during the growing season. Results indicated that carbon dioxide flux has closer relationship with vegetation change than methane flux does.

The prediction of nitrogen concentration in soil by VNIR reflectance spectrum

In this paper, the study on predicting nitrogen concentration in soil by VNIR (visible near infrared) spectrum is introduced. First, we analyzed the relationships between absorption features and nitrogen concentration to select the absorption features significantly correlated with nitrogen. Then several parameters of spectra in the selected absorption features were calculated, including first derivative reflectance spectra (FDR), inverse-log spectra (log (1/R)) and Band Depth. All of the soil samples were split into a calibration dataset and a validation dataset. Using stepwise multiple linear regression method, we established the statistical relationships between these parameters and nitrogen concentration. The regression models were calibrated using the calibration dataset, and validated using the validation dataset. The good results indicate that soil spectrum in the VNIR range has the potential for the rapid simultaneous prediction of nitrogen concentration. Keywords-nitrogen; soil; VNIR spectrum; SMLR

Evaluation of various classifiers on regional land cover classification using MODIS data

Five classification methods which are MLC(Maximum Likelihood Classifier)、CART decision tree、BP neural network and Fuzzy ARTMAP neural network and Parzen window which is a new method based on non-parameter statistical theory introduced from pattern recognition are selected to map land cover of Huabei Area in China using MODIS 250m data. The results show that Parzen window performs best in five classifiers . And CART and BP has satisfactory results whereas Fuzzy ARTMAP has unexpected bad accuracy in comparison with MLC . CART decision tree has better flexibility and robustness, however, it pursues high accuracy at the cost of sample size. BP neural network has high accuracy but requires high-quality samples and it is hard to define its net structure parameters. The results also show the classification effect caused by the size of training samples on MLC、Parzen window and Fuzzy ARTMAP 、CART and BP are below 5%、5%-10% and above 10%, respectively. I. INTRODUCE AND BACKGROUND Land cover , an important factor in global change ,plays a major role in global-scale patters of the climate and biogeochemistry of the earth system and land cover type is a key input parameter of energy transferring model and terrestrial ecosystem process mechanism. Classification method of land cover types in remote sensing application is all along the very important aspect. In regional land cover mapping, MLC(Maximum Likelihood Classifier ) and unsupervised methods are classical and simple. The first global land cover product was finished using MLC by DeFries and Townshend(1994)[1]. And global land cover map at 1KM resolution was finished by Loveland(2000) using an unsupervised approach[2]. For the limit of these methods, decision tree and neural net , which have nonlinear and non-parameter characteristics, are beginning to blossom in global and regional land cover remote sensing classification methods in recent years. Hansen et al.(1996) used NOAA/AVHRR data to map global land cover at 1o×1 o resolution using decision tree and MLC, which showed the accuracy of decision tree was superior to MLC[3]. DeFries et al. (1998) derived an 8KM global map using decision tree method[4]. And Hansen et al.(2000) also applied the same method to global UMd product [5]. BP neural net, another method other than decision tree and representative of neural net, has been widely used remote sensing imagery classification. In wheat crops recognition by C.S.Murthy et al. (2003), BP neural net’s performance was superior to MLC’s[6].To overcome the difficult in convergence of BP structure, Fuzzy ARTMAP was developed because it could integrated fuzzy logic and adaptive resonance theory. Sucharita Gopal et al.(1994)finished global land cover map at 1-degree using Fuzzy ARTMAP approach[7]. Borak et al.(1999) compared MLC, decision tree and Fuzzy ARTMAP and the resulat showed that decision tree could reduce dimensions and retain most of information of imagery and was insensitive to training sample, whereas Fuzzy ARTMAP had higher accuracy than decision tree and MLC[8]. However, for the reason of the complexity of remote sensing image and land cover types, many classification methods perform differently and perhaps have different characteristics under different conditions. Currently, MODIS global land cover products, which are made using decision tree and Fuzzy ARTMAP and new technology, however, have bad performance in China[9].The methods above, have been succeeded in Chinese land cover/use mapping[10-13]. But most of them are centered in small area and few of them are applied and compared with each other in regional land cover mapping and performance of various methods are difficult to define . In this research, multi-spectral reflectance and multi-temporal EVI(Enhanced Vegetation Index) data derived from MODIS 250m image and DEM(Digital Elevation Map) data are selected to map land cover of Huabei area in China .And we choose five methods popular in remote sensing classification such as MLC(Maximum Likelihood Classifier)、 CART decision tree、BP neural network 、Fuzzy ARTMAP neural network and Parzen window which is a new non-parameter classifier based on statistical theory. By comparison of classification result using various methods, we will explore the potential of methods in regional land cover mapping. II. EXPERIMENTAL AREA The dimensions of the study area which is named Huabei area are 4725×5543 250M MODIS pixels, bounded by 30 to 43 north latitude and 109 to 124 east longitude. Huabei area comprises Beijing, Tianjin, Hebei province ,Shandong province and Shanxi province and the altitude is from 100m to 3000m. It includes a diverse array of natural and human-modified landscapes including deciduous broadleaf forest, evergreen needleleaf forest, shrublands and short Supported by Key and Important Project of CAS( KZCX1-SW-01) and 863 Plan(2003AA131170) 1281 0-7803-9050-4/05/

Fog Detection Using MODIS Data in the Yangtze River Delta

20.00 ©2005 IEEE. 1281 woody, grasslands, croplands, marsh and meadows, water, and urban and built-up.

Research on regional land cover mapping of the Yangtze River Delta using MODIS 250m data

MODIS data provides good data resources in fog detection for its abundant bands and high resolution. The heavy fogs in the Yangtze River Delta during Dec. 20th and 22th in 2007 are detected by multi-spectral threshold method. According to the differences of the radiation features of the cloud, the fog and the land in the visible, middle infrared and far infrared bands, the multi-spectral method is developed to distinguish the fog from the cloud. The classification results, compared with the surface observation, indicate that this method has a good effect on fog detection.

The characteristics of spatial and temporal variations of land surface temperature in the Yangtze River Delta

During the last two decades, the Yangtze River Delta, one of the most economically developed areas of China, experienced rapid urban expansion, and accordingly, masses of cropland have been converted into human buildings. To analyze the influence of landscape change, it is important to provide up-to-date land cover information of this area. This paper describes the development of Land cover map of the Yangtze River Delta using 250m MODIS data, and the main satellite data used in this study were MODIS EVI data, MODIS reflectance data and DEM. A filter method based on time series was applied to eliminate EVI noise, and a PCA analysis was performed to reduce the volume of data. Besides, homogeneity was calculated to present spatial texture information. Therefore, a compositive classification matrix was generated. Considering the natural and artificial conditions of the study area, a 9-type classification scheme was defined. ROIs (Region of Interest) were selected from Landsat ETM+ images by human interpretation consulting the Vegetation Atlas of China. Then the land cover map was generated using MLC method. After correction by buffering analysis, we got the final land cover classification of the Yangtze River Delta. The classification accuracy was assessed using fineresolution Landsat images, with an overall accuracy of 95.98%. In addition, our classification result was compared with the MODIS-IGBP land cover production and showed better accuracy. The good result indicated the good behavior of the synthetic classification features and technical processing used in our research, and also suggested the advantage of 250m MODIS data in regional land cover mapping.

Temporal and spatial characteristics of atmospheric methane in the Yangtze River basin and the analysis of the main environmental impact factors

Land surface temperature (LST) is one of the key parameters in the atmosphere-land energy and water transfers. An understanding of the spatial and temporal variations of land surface temperature is important to broad research fields, including climate, vegetation, hydrology, etc. In this paper, the cloud contamination of MODIS LST product was analyzed first, and showed that there are numerous data gaps in MODIS 8-day composite LST product, indicating the necessity of data interpolation. Then the Harmonic Analysis of Time-Series (HANTS) algorithm was applied to the LST time-series to rebuild cloud-free images and to distill harmonic components. According to the harmonic characters and reconstruct LST, the spatial and temporal variations of land surface temperature in the Yangtze River Delta were studied.

Possibilities of reflectance spectra data for the assessment of soil potassium concentration

Methane (CH4), a significant atmospheric trace-gas, controls numerous chemical processes and species in the troposphere and stratosphere and is also a strong greenhouse gas with significantly adverse environmental impacts. Since the SCIAMACHY on the Envisat was in orbit since 2002, CH4 measurements at a regional scale became available. This study (1) firstly improved the spatial resolution of 0.5°×0.5° lat/lon grid data provided by University of Bremen IUP/IFE SCIAMACHY near-infrared nadir measurements using the scientific retrieval algorithm WFM-DOAS to 0.1°×0.1° lat/lon with the ordinary Kriging method, (2) then analyzed the spatial-temporal characteristics of atmospheric CH4 concentration in the Yangtze River basin (YRB), China from 2003 to 2005, (3) finally analyzed the relations with the main environmental factors: the precipitation from GSMaP MVK+ 0.1x 0.1 lat/lon degree grid data and the temperature from 147 meteorological stations in the YRB. The analysis shows that atmospheric methane concentration has significant and obvious characteristics of the spatial distribution of the inter-annual cycle fluctuations and seasonal characteristics during the year, and points out that the temperature is the main impact factor.

Monitoring vegetation dynamics with SPOT-VEGETATION NDVI time-series data in Tarim Basin, Xinjiang, China

This paper presents the possibilities of extracting total potassium concentration in topsoil from Visible-near-infrared (VNIR) spectra and reflectance of image data. Stepwise multiple linear regression (SMLR) and partial least-square regression (PLSR) were used to select wavelengths which were highly correlated with the concentration of potassium. For spectral measurements (from 400nm to 2480nm, at 2 nm increments) and chemical analyses, 70 topsoil (0~20 cm) samples were collected in Tianjin City, North of China. Three methodologies of the reflectance spectra of topsoil samples were employed: derivative reflectance spectra (FDR), inverse-log spectra (log (1/R)) and band depth (Depth). According to the root mean square error of prediction (RMSEP), the best model was picked up. The optimal experiential model (R=0.73, RMSEP=1.33) was achieved by PLSR method with parameter- log (1/R). Based on these credible results, space distribution map of soil potassium concentration of Tianjin was drawn by ETM+ image. The coefficient showed that the first and second bands of ETM were important for soil potassium concentration prediction. The potassium concentration of seaboard is higher than that of inland area. Good prediction performance indicates that VNIR spectra are potentially useful for rapid estimation of potassium concentration in topsoil, and inverse-log spectra (log (1/R)) are the best parameter for prediction. Even the image data can be used for soil potassium concentration extraction and the influences of the atmosphere and proper pre-processing are very important to prediction precision.