Introduction

Precipitation plays a key role in the regional and global hydrological cycle. Also, precipitation is a difficult environmental variable to model and forecast in numerical weather prediction (NWP) systems. The use of high-resolution forecasting for heavy rains can play an important role in accurately predicting floods and regulating water resources behind dams. Numerical weather prediction (NWP) is of great importance. (WRF) model is the latest generation of mesoscale numerical weather forecasting models. Recent studies have shown that the WRF model has great potential in recording precipitation characteristics such as precipitation time, location, and evolution. However, the results are not ideal due to the low quality of the initial conditions for generating accurate rainfall values, which data assimilation methods can improve.

Materials and methods

In this study, CFSv2 global data with a horizontal separation of half a degree is used for the boundary conditions of the model for atmospheric variables. The time steps of the used data are six hours.The model was implemented for torrential rains from March 24 to 31, 2019 in the southwestern region of Iran. To be more sure of the output of the model, model runs from 00:00 on the 11th to 18:00 on the 13th of March were considered for total rainfall. Using CFSv2 model data, the model was run once without data assimilation (WRF) and once with data assimilation (WRFDA), and finally the model forecast for total precipitation was compared.The data used include data from ATMS, MHS, GPSRO satellites, and (prepbufr) data.

Results and discussion

In this study, CFSv2 climate model data were used as boundary and initial conditions in the WRF model (execution without data assimilation) to predict the flood of April 2019 according to March 24-31, 2019. Then, using satellite data and observational data of the earth's surface and upper atmosphere, the data assimilation process was done by the WRFDA model. Considering the changes of the model output in each model run, especially in a longer time interval, the said model was built in the time interval from March 11 to 13 and four times, namely 00:00, 06:00, 12:00 and 18:00 GMT, and a total of 24 runs. The precipitation output for two cases with data assimilation and without data assimilation was compared with the CFSv2 model output in the same period. From the comparison of the output images of the models, it is clear that the increase in the amount of precipitation of the WRF model compared to the CFSv2 model is clear compared to the zoning of the cumulative amount of actual precipitation. Also, the data assimilation model shows a slight difference in improving the amount of precipitation compared to the model without data assimilation. 10 points with observed rainfall of more than 150 mm were extracted from observation data. The average output of the models without data mining (WRF), with data assimilation (WRFDA), and the CFSv2 model for 12 implementations of each desired point were compared with the amount of actual precipitation in these points. As it is known, the output of the WRF model is closer to the actual amount of precipitation compared to the original CFSv2 model, and the output of the WRFDA data assimilation model is also more accurate than the WRF model without data assimilation.

The calculated continuous validation measures, including a mean absolute error (MAE) of 121.25 and 119.7 and a root mean square error (RMSE) of 136.32 and 135 for the model without data assimilation and the data assimilation model, respectively, indicate a relative increase in the efficiency of the data assimilation model.

Conclusion

The results demonstrate the effectiveness of the data assimilation method when applied to the combined WRF-CFS model for predicting heavy rainfall occurrences from March 24 to 31, 2019 in the southwestern regions of Iran. The model accurately predicted extreme rainfalls in this region, showcasing its ability to forecast heavy rains over a longer period compared to short-term models like the global GFS model.

The slight improvement in the output of the data assimilation model compared to the model without data assimilation has several reasons, some of which are mentioned below:

1- Likely, some of these data (including ground surface and upper atmosphere station data) have been used in CFS global model data assimilation, as a result, the effect of these data has been included in the prediction of the mentioned model, and the reuse of these data will likely bring much improvement. Do not have Of course, this does not mean that the reuse of these data is not meaningful and cannot be investigated because the use of regional models with higher resolution and the use of different data assimilation methods can produce different results.

2- As the forecasting time increases, the quality of the forecasting data decreases, as a result, according to the effect of data quality on the effect of data assimilation, it can be concluded that the decrease in data quality in a longer period leads to a decrease in the effect of data assimilation.

3- Errors of numerical models are divided into two types: random and systematic. The data assimilation method reduces the random error. In areas such as southwestern Iran with the complex Zagros mountain range, where a major part of the model error is related to systematic error, data assimilation will probably not have much effect.

4- There is a possibility that the use of other data assimilation methods and systems will be more efficient than the WRFDA system