Post-processing of the UKMO ensemble precipitation product over various regions of Iran: integration of long short-term memory model with principal component analysis

An accurate forecast of precipitation can significantly enhance the management of water resources. While the data originated from ground-based synoptic stations are known to be the most accurate inputs of hydrological models, it is mostly unavailable in developing countries. So, the other approaches, such as numerical weather predictions (NWPs), are considered proper alternatives. This study utilized the precipitation data of the UK Meteorological Office (UKMO) model over eight different regions of Iran. The eleven ensemble data of UKMO at 47 ground-based synoptic stations from 2007 and 2017 were chosen as the input variables, while the ground-based precipitation was considered the output variable. The long short-term memory (LSTM) model was used as the predictive model, and the three proposed input strategies were evaluated using correlation coefficient (CC) and normalized-root mean squared error (NRMSE). The results showed that the combination of LSTM and the principal component analysis (PCA) approaches in post-processing of the UKMO data (PPUKMOD) enhances CC and NRMSE by 9% compared to the raw UKMO dataset. Besides, the most performance in PPUKMOD is found in the G7 (Zagros Highlands) region. Moreover, the Zagros mountain and the northern-eastern part of Iran showed better performance in PPUKMOD based on the evaluation of longitude, latitude, and elevation ranges. The temporal assessment also revealed that the highest performance in PPUKMOD was observed in the cold and rainy months (CCaverage = 0.59 and NRMSEaverage = 0.74) where November was the first rank. The proposed methodology for post-processing the UKMO ensemble sources aligns well with Iran's observed precipitations. Subsequently, it can be used as the input of hydrological models.