محسن جوادي

The Helmand/Hirmand River Basin is a significant watershed for three countries: Afghanistan, Iran, and Pakistan. In this basin, the Hirmand River originates from the Hazarajat mountain ranges. The objective of this study was to eva‎luate and improve the seasonal precipitation forecast data from ECMWF for the Hirmand River Basin. This product provides precipitation forecasts in time series for the next 1 to 6 months. Ground-based precipitation observations are essential for eva‎luating and improving forecast data. However, in the Afghan section of the basin, the only available observational data comprises monthly precipitation records from 1960 to 1980. To address the challenge of limited ground-based data, reanalysis data were adjusted using observations from 1960-1980 to create an extended corrected time series (1960-2022), which was then used as a reference for improving forecast data. This study employed precipitation data products from CRU, ERA5, and GLDAS. A deep learning-based model with a Bi-LSTM network structure was developed to enhance these datasets. The results indicated the model's effective performance. When compared to ground observations, the median NSE of the uncorrected CRU, ERA5, and GLDAS datasets were 0.2, -0.25, and -0.4, respectively, which improved to 0.6, 0.6, and 0.5 after correction. Additionally, the Mann-Kendall test was conducted to analyze precipitation trends over the 63-year period. The trend analysis revealed that most products exhibited no significant precipitation trends. However, by comparing all aspects of trend analysis, eva‎luation metrics, and spatial precipitation maps, it was found that the corrected ERA5 data provided the best results and alignment with observations. Moreover, the corrected ERA5 data demonstrated good alignment with monthly, seasonal, and annual observations in the Iranian section of the basin for the years 1981-2022. Therefore, in the absence of ground precipitation data, the corrected ERA5 data was used as the basis for eva‎luating and improving forecast data. The results of improving the forecast data also showed the model's capacity to enhance eva‎luation metrics effectively. For instance, the median NSE of the uncorrected precipitation forecast time series was typically -1, which improved to approximately 0.6 after correction. In conclusion, the corrected precipitation forecast and reanalysis data generated in this study can serve as a valuable resource for various water resource management studies in the Hirmand River Basin.

حوضه آبريز هيرمند , اصلاح داده بارش , يادگيري عميق , تحليل روند , محصولات بازتحليل , محصول پيش‌بيني ECMWF

Hirmand River Basin , precipitation data bias correction , Deep Learning , Trend analysis , reanalysis products , ECMWF forecast product

Mohsen Javadi

Dr. Hosein Alizade