The impact of improved signal-to-noise ratios on algorithm performance: Case studies for Landsat class instruments

The Landsat Operational Land Imager (OLI) has 5 to 10 times better signal-to-noise ratios (SNRs) in all spectral bands than previous Landsat instruments. SNR performance has long been recognized as a value in instrument design, however, the impact on algorithm performance for earth science applications is poorly documented. Since SNR performance may drive design/cost tradeoffs on future missions, a set of experiments were designed to evaluate the impact of various SNR levels on algorithms applied to different science applications. The application areas studied spanned a wide range including water quality, land cover and forestry. The experiments involved producing data sets with a range of signal-dependent SNR values ranging from Landsat 7 ETM + levels to OLI levels. Algorithms were then run on these otherwise identical data sets and evaluation metrics applied to evaluate the relative performance versus SNR. In all cases, performance was shown to be a strong function of SNR with substantial increase in performance as SNR increased (e.g. constituent retrieval errors reduced by a factor of three). However, in some cases, the rate of increase slowed at higher SNR levels. Regrettably, the point of diminishing returns was not the same for all applications leaving significant burden on design teams to decide which application's needs could be fully met in terms of SNR requirements. (C) 2016 Elsevier Inc. All rights reserved.