Earticle

초록

영어

Previous studies of aerosol optical thickness (AOT) estimations were generally based on observations from a single satellite sensor. Due to the limited observations from one instrument, the observations yielded AOT estimations with a system bias. In this paper, we combined two heterogeneous data sources, Moderate Resolution Imaging Spectroradiometer (MODIS) and Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP), together and proposed collaborative regression models to achieve more accurate AOT estimations than a single sensor does. These two independent remote sensors in the A-train satellite constellation both provide global AOT retrievals and they scan the same location on the Earth surface within a two-minute interval. However, the two remote sensors have different design principles respectively and their heterogeneous observation data streams pose challenges for information fusion. In our study, we proposed two types of heterogeneous collaborative regression approaches. One type of collaborative regression approach fuses information in a feature level. The other type of collaborative approach combines information in a model level. In our study, in each level, we apply a linear regression collaboration model and a neural network collaboration model. The proposed approaches are evaluated based on global observation data from MODIS and CALIOP during April 2, 2009 and April 1, 2011. The encouraging experimental results show that the regression approach collaborating in a model level achieves significantly more accurate AOT estimations than the results from the collaborative regression approach in a feature level. It also obtains significantly superior results to the deterministic AOT retrievals from any single satellite sensor.

목차

Abstract
 1. Introduction
 2. Data Sets and Accuracy Measures
  2.1. AERONET Data
  2.2. MODIS Data
  2.3. CALIOP Data
  2.4. Spatial-Temporal Synchronization Data
  2.5. Accuracy Measures
 3. Heterogeneous Collaborative Regression Models
  3.1. Collaborative Linear Regression in a Feature Level (CLRFL)
  3.2. Collaborative Neural Network Regression in a Feature Level(CNNRFL)
  3.3. Collaborative Linear Regression in a Model Level (CLRML)
  3.4. Collaborative Neural Network Regression in a Model Level (CNNRML)
 4. Experimental Results
  4.1. Experimental Settings and Optimization
  4.2. Experimental Results
 5. Conclusions
 References

키워드

저자

• Bo Han [ International School of Software, Wuhan University, Wuhan, China ]

참고문헌

발행기관

간행물

표지보기 관심저널 등록

• 간행물명

  International Journal of Future Generation Communication and Networking

• 간기

  격월간

• pISSN

  2233-7857

• 수록기간

  2008~2016

• 십진분류

  KDC 505 DDC 605