Error and Uncertainty in Earth Observation Value Chains

Earth observation systems are providing a growing amount of high resolution and multi-spectral data including agricultural crop yield predictions, local weather forecasts, wild fire tracking, and water quality monitoring. With the advent of multiple elements and sophisticated processing and modeling, there are also increasing avenues for introduction of errors. It is important to quantify and characterize these errors in the remotely sensed data as it gets increasingly used to inform vital decisions. Here, a data value-chain approach is presented for conceptualizing introduction and propagation of errors in remote sensing data acquisition, processing, and decisions. A detailed case of calibration errors and their propagation to higher level data products is then discussed. An NDVI analysis is used as an example, and it is shown (for the selected region), that a 3% error in reflectance ratio of red and NIR bands translates into a difference of 60 % for category 5 NDVI classification (representing high vegetation). This amplification of errors along the data value chain is caused by nonlinear operators such as the application of quotients to differences of reflectance values as well as the use of classifiers based on fixed color thresholding.