有別於三維變分法，四維變分資料同化（Four Dimensional Variational Data Assimilation，4DVAR)方法具有1.擁有與流場相關的背景誤差統計特性，2.可同化不同時間之觀測資料，以及3.可利用適當的約束條件來增加分析場動力與熱力的平衡關係。然而四維變分資料同化需要大量計算資源，包括電腦記憶體的使用和計算效率。為此，本研究在WRF4DVAR的架構下，發展在內迴圈使用多重解析度（Multi-Resolution Incremental 4DVAR，MRI-4DVAR)解決方案，以增進同化的運算效能。同時透過同化傳統觀測資料，以評估此一方案的同化效能。結果顯示，MRI-4DVAR方法透過改變模式內層迭代網格解析度的條件，可以比傳統 Full resolution incremental 4DVAR (FRI-4DVAR)節省至少約13倍的計算時間。在同化單點溫度測試實驗和同化所有傳統觀測資料的實驗中，可看出MRI-4DVAR和FRI-4DVAR的結果差異不大。在同化後進行雨量預報的結果中也可看出MRI-4DVAR與FRI-4DVAR不論是在定性或是定量的比較上皆很相近。午後雷雨個案結果顯示，MRI-4DVAR法，有機會掌握局部發生的極短時強降雨現象，但對於局部複雜地形上的降雨極值預報仍有待進一步改進。

Different with three dimensional variational (3DVAR) method, four dimensional variational data assimilation (4DVAR) method possess the following advantages: 1) flow dependent background error covariance characteristic, 2) be able to assimilate multi-time levels observation data, 3) through model constraints to make the dynamical and physical balance in analysis field. However, 4DVAR needs huge computer resources including memories and CPU times. Therefore, we develop Multi-Resolution Incremental (MRI) method in inner loop of WRF 4DVAR to improve its computing efficiency in this research. And through assimilating all traditional observation data to evaluate the assimilation performance of this method. The result shows that, MRI-4DVAR can save 13 times of computational time comparing to Full resolution 4DVAR (FRI-4DVAR). In the experiments of assimilating single point temperature and all traditional observation data, the difference between MRI and FRI method is not obvious. The precipitation forecast of MRI and FRI is much close in both quantitative and qualitative analysis. In the summer thunderstorm case, the MRI-4DVAR could capture the very short term heavy rainfall phenomenon, but it still remains further improvement at local rainfall extreme value on the complex terrain area.