Observational data obtained specifically to improve model initial conditions for a numerical forecast of a selected weather feature, or to optimize a measure of forecast outcome (e.g., error). Guidance for selecting adaptive observations can be obtained from model-based products, such as singular vectors from adjoint models or ensemble forecasts, which are used to estimate where initial condition error has the greatest impact on the forecast measure. Adaptive observations can be obtained from in-situ platforms such as dropsonde aircraft, or by direction of remote sensors including satellite or radar instruments. See adjoint model, adjoint sensitivity.