Errors and uncertainties in ocean colour remote sensing (1)

IOCCG Summer Lecture Series 2012

One of the main questions you will be asked as a remote sensing expert is: how reliable and good is information, which we derive from remotely sensed ocean colour data? Can we trust them? What is the error or uncertainty range of these data? In this section of the IOCCG training course, which consists of 3 lectures and exercises, we will look into this problem. Lectures The first lecture will be dedicated to the sources of uncertainties. We have to consider that our observations are the reflectivity in a number of spectral bands, which are measured at the top of atmosphere (TOA) or, in case of an aircraft platform, in a certain height above the water. We try nothing less than to isolate, retrieve and quantify a small effect on these spectra, which is caused by absorption and scattering of e.g. of phytoplankton, from a large number of other effects, of which in particular the atmosphere dominates the TOA spectrum. Problems of this kind may induce large uncertainties. In some cases it might be even impossible to retrieve reliable information of the ocean from remotely sensed reflectance spectra. Thus, one important area of ocean colour research is to analyze sources of uncertainties, to develop methods to quantify uncertainties and finally to find way to reduce uncertainties. In this lecture we will consider Natural factors, which determine uncertainties, and their variability Uncertainties, which are induced by reducing the manifold of factors to a few dominant wavelength (nm) Radiance (Wm?2 sr?1 ?m?1) air molecules different aerosols thin clouds Sky reflectance Sun glint foam floating material chlorophyll Suspended particles different phytoplankton species dissolved organic matter Vertical distribution Bottom reflection contrails Factors, which determine top of atmosphere reflection spectra, from which try to retrieve e.g. the chlorophyll concentration Errors caused by spaceborne or airborne instruments: calibration, ageing, noise Errors caused by in situ measurements, sampling and procedures Problem of comparing in situ with space borne In the second lecture we look into procedures, how to determine uncertainties: How to quantify uncertainties: scatter, bias, robustness, stability Validation procedures and strategies Testing of algorithms Round robin exercises Sensitivity studies Determination of uncertainties on a pixel by pixel bases flagging The third lecture will finally discuss the results of our exercises and will be dedicated to the question, how to reduce uncertainties. This is a wide field, where a lot of research is still needed, and it offers themes for your future work. Detection of spectra / pixels, which are out of scope of the algorithm Masking of clouds and cloud shadows Use of additional information Pre?classification of water types and use of dedicated algorithms How to produce maps from satellite data, which include information about uncertainties.