Rice yield loss from pre-release in irrigation reservoirs across Japan: A national-scale assessment incorporating precipitation-forecast error

Pre-release from irrigation reservoirs can create flood-storage capacity but may reduce water available for rice production if subsequent rainfall fails to restore storage. We quantified this agricultural response around the heading and flowering stage across Japan and examined its relation to reservoir structure and precipitation-forecast uncertainty. AquaCrop was coupled with a daily reservoir water-balance model using 20 historical d4PDF ensemble members for 2000–2010. Paired simulations compared no release with post-release storage fractions of 0, 0.25, 0.50, and 0.75 across five levels each of the catchment-to-irrigated-area ratio and storage capacity per unit irrigated area. The yield-loss event proportion (the proportion of paired records with relative yield loss >10%) was evaluated for release scenarios not conditioned on precipitation and is not an annual probability of loss under actual operation. Yield-loss events were concentrated under complete drawdown (a stress test); the transition occurs somewhere within the unsampled 0–0.25 storage-fraction range. Under complete drawdown, the event proportion decreased from 5.4% at a catchment ratio of 0.04889 to 0.1% at 492.4, whereas conditional mean relative loss among events remained 13.9–15.1%. Differences among tested storage capacities were within 0.2 percentage points, while the grid-cell event proportion reached approximately 25%. The 90th percentile of relative loss varied more clearly with 8-day post-release precipitation than with target-day precipitation. Thus, yield-loss events were concentrated under complete drawdown and depended mainly on inflow-recovery conditions and regional meteorology rather than storage capacity within the tested range. The framework provides a first-stage national screening of irrigation reservoirs and regions requiring site-specific verification.