|
Abstract:
|
Remote sensing, model and reanalysis approaches provide noisy, biased global estimates of large scale geophysical variables at different spatial and temporal scales. While all three groups of estimates are vital in different applications, their validation still remains a challenge since the error-free true value of the variable of interest is not known. The most common practice in validating these estimates is comparing them to point-based ground measurements assuming them as error-free truth. However, ground-based estimates have their own measurement error that is neglected in many cases; moreover, the different spatial scale of the two types of products (pixel vs. point) adds a 'representativeness error' to the comparison that is not accounted for. In this study, we review and extend a family of methods that use three collocated estimates of the same variable to characterize the random error in each estimate without assuming any of them as error-free. We will show the application of these methods in error characterization of wind speed, precipitation intensity, soil moisture, and freeze/thaw estimates.
|
