Journal
REMOTE SENSING OF ENVIRONMENT
Volume 156, Issue -, Pages 500-509Publisher
ELSEVIER SCIENCE INC
DOI: 10.1016/j.rse.2014.10.016
Keywords
Snow depth retrieval; Integration of multi-sensor data; Support vector machine
Funding
- National Basic Research Program of China (973 Program) [2011CB707103]
- National Science Fund for Excellent Young Scholars [41322009]
- Foundation for the Author of National Excellent Doctoral Dissertation of PR China [3149001]
Ask authors/readers for more resources
The accuracy of snow depth retrieval by remote sensing depends heavily on the characteristics of the snow, and both passive microwave and visible/infrared sensors can contribute to the acquisition of this information. A method integrating these two remotely sensed data sets is presented in this study. Snow depth retrieval is performed using microwave brightness temperature at 19 and 37 GHz from the Special Sensor Microwave/Imager (SSM/I) and the Special Sensor Microwave Image/Sounder (SSMI/S), and visible/infrared surface reflectance from Moderate Resolution Imaging Spectroadiometer (MODIS) products. Microwave brightness temperature provides information about the volume of snow pack, and visible/infrared surface reflectance can indicate snow presence and surface grain size. With these two remote sensing data sets, snow depth is retrieved by a nonlinear data mining technique, the modified sequential minimal optimization (SMO) algorithm for support vector machine (SVM) regression. The proposed method is tested by using 16,329 records of dry snow measured at 54 meteorological stations in Xinjiang, China over an area of 1.6 million km(2) from 2000 to 2009. The root mean square error (RMSE), relative RMSE and the correlation coefficient of our method are 621 cm, 0.64 and 0.87, respectively. These results are better than those obtained using only brightness temperature data (8.80 cm, 0.90 and 0.73), the traditional spectral polarization difference (SPD) algorithm (15.07 cm, 1.54 and 0.58), a modified Chang algorithm in WESTDC (9.80 cm, 1.00 and 0.62), or the multilayer perceptron classifier of artificial neural networks (ANN) (9.23 cm, 0.94 and 0.72). The daily snow water equivalent (SWE) retrieved by this method has an RMSE of 8.05 mm and a correlation of 0.84, which are better than those of NASA NSIDC (32.87 mm and 0.47) or Globsnow (19.07 mm and 0.59). This study demonstrates that the combination of visible/infrared surface reflectance and microwave brightness temperature via an SVM regression can provide a more accurate retrieval of snow depth. (C) 2014 Elsevier Inc. All rights reserved.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available