Technical note: A flexible framework for precision reduction of WRF inputs and outputs to balance storage efficiency...

The rapid growth of data volumes from high-resolution regional climate simulations necessitates effective storage reduction strategies that do not compromise scientific integrity. Applying lossy precision reduction prior to lossless compression provides a promising approach. However, the distinct scientific implications of reducing precision in time-varying input forcings versus prognostic model outputs remain insufficiently quantified. Using a one-year Weather Research and Forecasting (WRF) simulation, we systematically evaluate the storage benefits and downstream diagnostic impacts of retaining 5, 4, and 3 significant digits for both inputs and outputs. From a storage perspective, combining precision reduction (retaining 5-3 digits) with bzip2 compression reduces model outputs to 19.2 %-7.5 % of their original uncompressed sizes and model inputs to 52.4 %-18.5 %. Scientifically, precision-reduced inputs interact with nonlinear model dynamics and can induce spatial phase shifts in simulated meteorological systems. Although this process reduces deterministic grid-scale correspondence, the overall spatial morphology of the atmospheric structures remains largely preserved. Consequently, aggregate statistical distributions are weakly affected, especially during dynamically less active periods, rendering input precision reduction suitable for large-scale spatial aggregates and event-averaged statistical analyses. In contrast, output precision reduction acts as a static numerical filter whose impacts depend strongly on the intrinsic characteristics of individual variables. For example, regarding surface wind speed, retaining 3 significant digits still preserves an adequate error buffer. Cumulative variables, such as precipitation, progressively amplify quantization errors during temporal differencing and therefore require 4-5 significant digits to avoid artificial increments. Ultimately, this study constructs a flexible framework for WRF data compression. By dynamically tailoring precision to specific variable typologies and downstream scientific demands, the modeling community can substantially improve storage efficiency while rigorously safeguarding physical integrity.