Earth Observation through satellites enables decision makers to assess situations near real-time with unprecedented spatial coverage. The data-value added products from radiometric satellite images often use indices derived from the unique spectral properties of materials and are sensitive to the relative gains of the different bands of the satellite sensor. However, satellite sensors are susceptible to degradation from the space environment, leading to drift in band response. For well-calibrated satellites such as Landsat 8, these drifts are well characterized and can be corrected for during processing--however, for satellites lacking on-board calibration (such as CubeSats), these trends can be difficult to detect and require novel methods combining cross calibration with machine learning. Given that satellite data often undergoes several levels of processing prior to use, there is a need to quantify the relationship between calibration errors and the errors of the final data-valued added product. This study investigates two applications of Earth Observation data: crop classification and Harmful Algal Bloom detection, and quantifies the impact of induced radiometric error on the final data product.