This study proposes another approach to develop multiyear single-column model (SCM) and cloud system--resolving model (CSRM) forcing data from numerical weather prediction (NWP) model analyses constrained with the observed surface and top of the atmosphere measurements by using a variational analysis approach. In the approach the atmospheric state variables from NWP analyses are adjusted to balance the observed column budgets of mass, heat, moisture, and momentum rather than the NWP model-produced budgets. The derived constrained NWP forcing data from the National Oceanic and Atmospheric Administration rapid update cycle (RUC) analyses are evaluated by the observed forcing data derived from radiosonde and wind profiler data collected at the Atmospheric Radiation Measurement (ARM) Program Southern Great Plains site under three selected cases: A strong convective case in the ARM summer 1997 intensive operational period (IOP), a moderate synoptic-scale process-dominated precipitation period in the spring 2000 IOP, and a nonprecipitation period in the late fall 2000 IOP. We show that the forcing data derived from the RUC analyses using ARM column constraints agree with the observed forcing reasonably well. The largest improvements are seen during precipitation periods since precipitation is a strong constraint used in the proposed approach. During the nonprecipitation period the improvements are moderate because the constraints are weak in the absence of precipitation. The constrained NWP forcing and the observed forcing, however, show better agreement during the moderate precipitation period and the nonprecipitation period than during the strong convective period. In SCM tests we show that most model errors revealed by the SCM driven by the observed forcing can be seen in the SCM driven by the constrained NWP forcing. These results suggest the feasibility of using the derived constrained NWP forcing data from RUC analyses for statistical studies of SCM/CSRM results over long time periods.