Study on arctic melt pond fraction retrieval algorithm using modis data

J. Su, P. Yu, Y. Qin, G. Zhang, M. Wang

Research output: Contribution to journalConference articlepeer-review


During spring and summer, melt ponds appear on the sea ice surface in the Arctic and play an important role in sea ice-albedo feedback effect. The melt pond fraction (MPF) can be retrieved using multi-band linear equations, but the calculation is complicated by the ill-conditioned reflectance matrix. In this paper, we calculated the condition numbers which represent the degree of the ill-conditioned reflectance matrix in the results of the MPF from a MODIS-based unmixing algorithm. The condition number is introduced here as a criterion for the sensitivity of the solution in the system to the error in the input value. By combining 3 bands among 5 visible and near-infrared bands of MODIS data, the results show that the three-band combination with the lowest sensitivity to the error of input is B245. To improve the algorithm, we introduce pre-processing to remove open water from the four surface types and then remove one reflectance equation from the original set. The best two-band combination algorithm is B15. Compared with the discrimination results from Landsat5-TM, the RMS is 0.14. This algorithm is applied in pan-Arctic scale, the MPF results are larger than that from University of Hamburg, especially in the Pacific sector.

Original languageEnglish (US)
Pages (from-to)893-898
Number of pages6
JournalInternational Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives
Issue numberB3
StatePublished - Aug 6 2020
Externally publishedYes
Event2020 24th ISPRS Congress - Technical Commission III - Nice, Virtual, France
Duration: Aug 31 2020Sep 2 2020


  • Arctic
  • melt pond fraction
  • retrieval algorithm

ASJC Scopus subject areas

  • Information Systems
  • Geography, Planning and Development


Dive into the research topics of 'Study on arctic melt pond fraction retrieval algorithm using modis data'. Together they form a unique fingerprint.

Cite this