1 / 23

Optimizing Moderate Resolution Optical Remote Sensing of Snow Fraction April 11, 2018

Optimizing Moderate Resolution Optical Remote Sensing of Snow Fraction April 11, 2018. Igor Appel TAG LLC, Washington, USA. Current Status.

timb
Download Presentation

Optimizing Moderate Resolution Optical Remote Sensing of Snow Fraction April 11, 2018

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Optimizing Moderate Resolution Optical Remote Sensing of Snow FractionApril 11, 2018 Igor Appel TAG LLC, Washington, USA

  2. Current Status The fractional snow cover was a product supplied from the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard the Terra and Aqua satellites from launch until the end of 2016, and became a standard requirement for many users. However, from 1 January 2017 the program stopped providing the fractional snow product, although no other global products of snow fraction exist. This is a matter of significant concern creating problems for users of numerous applications.

  3. Existing Algorithms A large scale international snow products intercomparison and validation project initiated by the European Space Agency (ESA) completed at the Environmental Earth Observation company in Innsbruck, Austria. The validation results show that the snow fraction algorithm of Salomonson and Appel has advantages in comparison with other methods, both for open land and in forested areas for two reference data sets created using different techniques.

  4. Two Types of Reflective Properties Variability The algorithm takes into account that the snow and non-snow reflective properties, even within a single satellite image, are very variable. However, snow and background endmembers could be characterized with a high accuracy by the Normalized Difference Snow Index (NDSI). The enhanced NDSI-based algorithm explicitly accounts for changes in the endmembers from one image to another using scene-specific snow and non-snow endmembers calculated on the fly.

  5. NDSI-Based Snow Fraction: Physical Basis (1) NDSI=(R1-R3)/(R1+R3) I1 I3 The information from bands I1 and I3 is very useful to distinguish snow from non-snow

  6. NDSI-Based Snow Fraction: Physical Basis (2) Snow is a unique land cover type - among the brightest of natural substances in the visible part of the spectrum, but it is also often the darkest in the short wave infrared High visible and low short-wave reflectances correspond to snow; low visible and high short-wave reflectances correspond to non-snow The reflectance spectra shown in the previous figure are representative of one specific case. Any given surface type and individual scene will exhibit spectral variability

  7. Bands Used in VIIRS (and MODIS) Algorithms NDSI-based algorithm NDSI = (I1 - I3) / (I1 + I3) Current Input for Snow Fraction Expected Added Input

  8. Snow & non-snow reflectances are characterized by high variability within a scene NDSI-Based Snow Fraction: Physical Basis (3) Shortwave IR reflectance Shortwave IR reflectance Non-snow Non-snow S n o w S n o w Visible reflectance Visible reflectance High resolution observations Moderate resolution

  9. NDSI-Based Snow Fraction: Physical Basis (4) The lines going through the most probable locations of snow and non-snow in the spectral space are considered as corresponding to 100% and 0% of snow fraction. Pixels between the lines correspond to intermediate snow fraction 0% Shortwave IR reflectance 100% Visible reflectance

  10. NDSI-Based Snow Fraction: Mathematical Description(1) The Normalized Snow Difference Index (NDSI) characterizes the distinction between the reflectance in the visible wavelengths and the reflectance in the near infrared wavelengths NDSI is widely considered as an indicator of the presence of snow on the ground NDSI is sensitive enough to provide the snow fraction NDSI presenting relative ratio of reflectances to a large degree suppresses the influence of varying illumination conditions

  11. NDSI-Based Snow Fraction: Mathematical Description(2) “Universal” approach to NDSI-based snow fraction retrieval The Salomonson-Appel (MODIS) algorithm linearly relates snow fraction to the observed NDSI SFNDSI = a + b * NDSI Parameters a =-0.01 andb=1.45 characterize the optimal “universal” linear function for MODIS data The NDSI snow fraction formula is equivalent to SFNDSI= (NDSI - NDSInon-snow) / (NDSIsnow - NDSInon-snow)

  12. Role of changes in endmembers NDSI-Based Snow Fraction: Mathematical Description(3) Changes in pixel reflectances should not be ascribed exclusively to variable fraction, because they depend also on the variability in spectral signatures of the endmembers The quality of snow retrieval could be improved if the variability of reflective properties characterizing snow and underlying non-snow states is taken into account Allowing for the variability in spectral signatures of endmembers is a key requirement to snow algorithms

  13. Variability of snow & non-snow reflectances (between scenes) Shortwave IR reflectance Visible reflectance The most probable snow (asterisks) and non-snow (squares) VIIRS reflectances for 16 scenes under detailed analysis

  14. NDSI probability density function identifiespredominant magnitudesof snow andnon-snow NDSI Scene-specific Estimate of Snow and Non-snow NDSI Non-snow Snow -1 -0.5 0 0.5 1 NDSI Local snow and non-snow endmembers are estimated from the analysis of NDSI frequency (probability density function)

  15. NDSI-Based Snow Fraction: Mathematical Description(4) Scene-specific approach to NDSI-based snow fraction retrieval The NDSI snow fraction formula is equivalent to SFNDSI= (NDSI - NDSInon-snow) / (NDSIsnow - NDSInon-snow) The quality of snow retrieval could be improved if the variability of reflective properties characterizing snow and underlying non-snow states is taken into account The adjustment of the parameters in snow algorithms to specific local conditions is a promising improvement leading to better quality of the VIIRS snow products

  16. Summary of Quantitative Assessment of Algorithm Performance Average correlation coefficient is 94% despite a couple of low magnitudes Typical intercept of linear regression line is on the order of 1% Average slope of linear regression line is more than 0.9 Average bias of data is 2 % Average standard deviation is 10%

  17. Stratified Assessment of Algorithm Performance NDSI-based fraction Comparison of ground truth with NDSI algorithm results (thick lines) and trends (thin lines) for intermediate fractions demonstrates stratified performance for individual scenes Landsat fraction

  18. Conclusions Estimation of fractional snow cover is not a low priority task due to user requirements, and because this property of the snow that covers up to 40% of land in the Northern Hemisphere is critical for retrieval of many atmosphere and land products. The optimal way to derive moderate resolution optical remote sensing information on fractional snow cover should allow for the local variability of snow and non-snow reflective properties within a scene-specific algorithm. Creation of unbiased and consistent information on fractional snow cover is required for global studies as well as for numerous regional and local scale applications.

  19. Recommendations It is possible to assume that the collaboration between several parties tackling the same problem can be very promising and successful. Collaboration between researchers with experiences supplementing each other helps solve the problem of optimizing fractional snow algorithm development. Such collaboration can be advised to ESA to consider as a key recommendation to develop snow product algorithms and other related activities for the Sentinel-3 mission. 

  20. Published Summary Article - Optical remote sensing of snow fraction—status and future prospects. 2 page OPINION EDITORIAL in Advances in Polar Science (December 2017) http://www.aps-polar.org/ paper /2017/28/04/A180411000001 iappel@earthlink.net h

  21. Comparing Performance of Alternative Algorithms The study case includes the comparison of the following approaches under completely comparable conditions for 16 images: (a) tie-point algorithm for a visible band assuming a linear relationship between visible reflectance and snow fraction. (b) the Normalized Difference Snow Index (NDSI) method based on the concept of a linear relationship between NDSI and snow fraction.

  22. Tie-Point vs NDSI Algorithms The linear relationships of snow fraction with visible reflectance and the Normalized Difference Snow Index were estimated using the regressions of ground truth snow fraction on both visible reflectance and NDSI. The study examines the performance of the algorithms under consideration on the basis of constructing the best linear regressions for each individual scene under consideration, which allows direct comparison of optimal relationships for the algorithms.

  23. Comparison of Results Such a nontraditional approach provides the opportunity to compare potential optimal versions of the alternative algorithms simply using by the correlation tools. The linear regressions on NDSI (correlation coefficients of 0.95) have obvious advantages when compared to the linear regressions on the visible reflectance (correlation coefficients of 0.85). There is 60% increase in the standard deviation (100% in variance) for the regressions on visible reflectance in comparison with the standard deviation for the regressions on NDSI.

More Related