|
Evaluation of WRF Model's Performance to Predict Late-Spring Frost (Case Study: Ardabil and Zanjan Provinces)
|
Seyedeh Mahboubeh Ebnehejazi   , Hojjatollah Yazdanpanah * 1 |
|
|
|
Abstract: (3 Views) |
Introduction
Mesoscale and regional forecasts of temperature play an important role in decreasing the damage caused by frost, especially in the agricultural sector. Among atmospheric hazards, frost is an important cause of serious damages to agricultural products. Due to the important influence of frostbite on production of agricultural products, a solution which can help decrease the potential damage of frostbite is prediction of minimum temperature (Sabziparvar and Khoshhal Jahromi, 2018). Since climate changes can affect the extent and intensity of late-spring frostbite, prediction of such events is of utmost importance (Chamberlain et al., 2019). One of the methods of prediction is numerical weather prediction (NWP), an example of which is WRF model (Weather Research & Forecasting Model) which has drawn much attention to itself nowadays. Assessment of models to determine the capacity of numerical weather prediction systems to generate accurate data is essential (Dzebre et al., 2021). Indeed, much research has been conducted on this basis. In these studies, performance of WRF model in predicting various meteorological variables like rainfall and 2m temperature in various regions has been assessed by using statistical indicators (Bhimala et al., 2021, Nooni et al., 2022, Lin et al., 2023, and Castorina et al., 2023). Furthermore, in some other studies the accuracy of WRF model's predictions in various predicting periods of e.g. 24, 48, and 72 hours was evaluated. In such studies, for each prediction period, obtained data from model were compared with observed data from statistical indicators and, ultimately, through comparison of indicators, the best period for prediction of variables was determined (Zhang et al., 2019, Naveena et al., 2021, Tedla et al., 2022, Ghassabi et al., 2023). In this study, we also aim to assess the accuracy of 24 and 48-hour simulations of temperature through WRF model in predicting the incidence of late-spring frost in Ardabil and Zanjan provinces.
Materials and Methods
In order to evaluate the 24 and 48-hour simulations of minimum temperature through WRF model in Ardabil and Zanjan provinces, with regards to the end of frost time in these two provinces, 1 and 2-day, 2m temperature simulations were conducted for 19 days from March 28th to April 16th, 2012 at 3 UTC. The computational network for simulation included three nested domains with grid resolutions of 3, 9, and 27 kilometers. Horizontal resolutions of land roughness data and land use data equaled 30 seconds. Initial and boundary conditions from Global Forecast System (GFS) data in time periods of 3 hours with horizontal resolutions of 0.5 degree were obtained from National Center for Environmental Information (NCEI). Temperature data of at least 8 synoptic stations from Zanjan province and 11 synoptic stations from Ardabil province were obtained from Iran Meteorological Organization (IRIMO). For a more precise determination of data match rate of minimum temperature data of 24 and 48-hour simulations with observed data in synoptic stations of Zanjan and Ardabil, a scatter plot was used. Also, in order to compare predicted values through WRF model and observed values in stations, statistical indicators like MAE, RMSE, MBE and R were made use of.
Results and Discussion
By comparing data obtained from WRF model and data observed through statistical indicators, we concluded that compared to 48-hour simulation, 24-hour simulation of 2m temperature was more accurate in both provinces. In addition, regarding the statistical indicators, performance of WRF model in simulating the temperature and predicting late-spring frost was better in Ardabil Province than in Zanjan Province. It should of course be noted that the complex mountainous topography can account for the low accuracy of 2m data obtained from WRF model in these two provinces. Correlation coefficient (R) in Ardabil in both 24 and 48-hour simulations was about 0.8 while it was about 0.3 in Zanjan for both simulations. Assessment indicators of MAE, RMSE, and MBE of 24-hour simulation in Ardabil Province equaled 2.3, 3, and 1.4, respectively. In Zanjan Province, they were 4.3, 5.9, and 3.1, respectively. These results show that compared to 48-hour simulation, 24-hour simulations of 2m temperature in both provinces were more accurate. Also, based on the values of the above-mentioned indicators, performance of WRF model in simulating the temperature and predicting late-spring frost has been better in Ardabil Province than in Zanjan Province. In Zanjan Province, the weak correlation between the data of the model and the observed data and high error rate prove WRF model performance in 2m temperature simulation unacceptable. Of course, considering the results of this research and other researches done on accuracy of WRF predictions, it can be inferred that overall, the low accuracy of 2m data obtained from WRF model in Ardabil and Zanjan provinces results from uneven, complicated, mountainous topography of these regions.
Conclusion
This study illustrated the capacity of WRF model in 24 and 48-hour simulations in Ardabil and Zanjan provinces in predicting the incidence of late-spring frost. Spring-frost prediction is an efficient method in reducing the damage to agricultural crops. Therefore, examining and evaluating various methods of late-spring frost prediction plays an important role in selecting the most accurate method to predict this hazard.
|
Article number: 10 |
| Keywords: Late-Spring Frost, minimum temperature simulation, Verification, WRF model |
|
|
Full-Text [PDF 2057 kb]
(3 Downloads)
|
Type of Study: Research |
Subject:
Special
Received: 2024/02/9 | Accepted: 2024/11/5 | Published: 2025/10/8
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
|
|
