While decades of research have constrained the drag coefficients of low Earth orbit satellites below 500 km, little is known about drag coefficients above 500 km altitude. At altitudes below ∼300 km, the adsorption of atomic oxygen to spacecraft surfaces has been well-established. Furthermore, the adsorption of atomic oxygen at altitudes ∼500 km has been well fit with a Langmuir isotherm; however, there is disagreement between the Langmuir fit and fitted drag coefficients above 500 km altitude. Previous work with STELLA and the Calsphere satellites has indicated that gas-surface interactions are primarily diffuse, even at 700 km altitude. This finding is at odds with drag coefficient models based on adsorption models for atomic oxygen which lead to increasingly specular reflection at higher altitudes due to a sharp decline in atomic oxygen adsorption. There are several possible explanations for the disagreement of adsorption theory with previous work. These possibilities include erosion, atmospheric density biases, and deficiencies in the adsorption model. We explored the third option in a previous paper and found it unable to explain the disagreement. The first and second possibilities are explored in this work using an orbital propagator, IMPACT-prop, in conjunction with the empirical atmospheric model, NRLMSISE-00, to calculate fitted drag coefficients throughout the lifetime of a collection of spherical satellites. The satellites are analyzed for signatures of erosion (defined by a decrease in the drag coefficient over several months not due to changes in atmospheric properties) and atmospheric density biases. A few satellites, especially GFZ-1, are found to have an erosion signature; however, atmospheric density biases are far more widespread in the results and often lead to ambiguous conclusions regarding erosion. Atmospheric density biases and updated fitted drag coefficients for the STELLA and the Calsphere satellites help to resolve the disagreement between previous work and adsorption-based drag coefficient models.