[论文解读] Three-Dimensional Variational Data Assimilation with Rapid Update Cycling for Short-Range Precipitation Forecasting: A Case Study of Heavy Rainfall in Bali, Indonesia

Nurjanna Joko Trilaksono, Sandy Herho+4 more|arXiv (Cornell University)|Mar 20, 2026

Meteorological Phenomena and Simulations被引用 0

一句话总结

该研究在 Bali 的 WRFDA 中评估带有快速更新循环（RUC）的 3D-Var 数据同化；与较少频率的循环和非数据同化相比，逐小时循环显著改善温度和降水预报。

ABSTRACT

This study evaluates the effectiveness of three-dimensional variational (3D-Var) data assimilation coupled with a Rapid Update Cycle (RUC) framework for improving short-range precipitation forecasts over the Indonesian Maritime Continent (IMC). We employ the Weather Research and Forecasting (WRF) model and its data assimilation component (WRFDA) to assimilate surface observations from Automatic Weather Stations (AWS) at cycling intervals of 1, 3, 6, and 12 hours. Our test case is a heavy rainfall event on 7 July 2023 in Bali Province, during which accumulated precipitation exceeded 193 mm.day$^{-1}$. The 1-hour cycling interval yields the lowest root-mean-square error (RMSE) for both 2-meter temperature (0.0-0.3$\,^\circ$C) and hourly precipitation (1.295 mm.h$^{-1}$), corresponding to reductions of roughly 75% and 57%, respectively, relative to non-assimilated forecasts. Frequent cycling constrains initial-condition errors and captures mesoscale convective evolution, as confirmed by improved spatial agreement with radar reflectivity observations. These results demonstrate that high-frequency assimilation cycling offers clear advantages for nowcasting in tropical maritime environments.

研究动机与目标

• Evaluate the effectiveness of 3D-Var data assimilation integrated with a Rapid Update Cycle for short-range rain forecasts over the Indonesian Maritime Continent.
• Quantify how different cycling intervals (1, 3, 6, 12 hours) affect forecast accuracy during a heavy rainfall event in Bali.
• Assess the impact of high-frequency assimilation on initial-condition error growth and mesoscale convective evolution.

提出的方法

• Implement 3D-Var data assimilation within WRFDA using CV3 control variables and domain-specific background error covariances via the NMC method.
• Use a two-domain WRF setup (9 km outer, 3 km inner) over Bali with AWS surface observations at 10-minute intervals.
• Test cycling intervals of 12, 6, 3, and 1 hour(s) starting 08:00 WITA on 6 July 2023 for 48-hour forecasts.
• Evaluate forecasts with RMSE and Bias for 2 m temperature and hourly precipitation against AWS observations and radar/ BMKG data.
• Compare against a non-DA (cold-start) run to isolate the impact of data assimilation.
• Include an observational preprocessing workflow (quality control and interpolation) and standard observation error specifications.

实验结果

研究问题

• RQ1 Does more frequent data assimilation cycling improve short-range precipitation forecasts in a tropical maritime setting like Bali?
• RQ2 How does cycling interval affect the spatial distribution of rainfall and convective evolution compared with radar observations?
• RQ3 What are the quantitative gains in forecast RMSE and bias for temperature and precipitation when using hourly versus 6-, 3-, and 12-hour cycling?
• RQ4 What are the practical considerations and limitations of implementing hourly cycling in terms of computation and observation networks?

主要发现

• 1 小时循环在 2 m 温度的 RMSE（0.0–0.3 °C）和逐小时降水 RMSE（1.295 mm h^-1）方面达到最低。
• 与非 DA 相比，1 小时循环将温度 RMSE 降低约 75%，降水 RMSE 降低约 57%。
• 在 Tabanan 的逐小时降水 RMSE 随循环变得更粗而下降（Cycle 3、Cycle 6、Cycle 12）。
• 在 Tabanan，Cycle 1：1.295 mm h^-1；Cycle 3：1.432 mm h^-1；Cycle 6：1.745 mm h^-1；Cycle 12：2.572 mm h^-1；DA：2.482 mm h^-1；Non-DA：3.023 mm h^-1。
• 在 Denpasar，Cycle 1：1.487 mm h^-1；Cycle 3：1.853 mm h^-1；Cycle 6：1.855 mm h^-1；Cycle 12：3.224 mm h^-1；DA：3.877 mm h^-1；Non-DA：3.954 mm h^-1。
• Cycle 1 最能再现 GPM 降雨模式和雷达反射率，相对于更高循环间隔和非 DA。