Document Type : Research Article
Authors
1
Department of Physical Geography, Faculty of Social Sciences, University of Mohaghegh Ardabili, Ardabil, Iran.
2
Department of Physical Geography, Faculty of Social Sciences, University of Mohaghegh Ardabili, Ardabil, Iran
Abstract
This study was conducted to evaluate the accuracy of snow depth estimation derived from two reanalysis datasets, ERA5-Land and MERRA-2, across northwestern Iran, using data from ten selected synoptic meteorological stations over a 20-year period (2004–2023). Ground-based observational data were obtained from the Iranian Meteorological Organization, while reanalysis data were acquired from the NASA repository for MERRA-2 and the Copernicus Climate Data Store for ERA5-Land. To downscale the outputs of both reanalysis datasets and reduce systematic bias, the Quantile Mapping (QM) method was employed. The performance of ERA5-Land and MERRA-2 was statistically evaluated by comparing their estimated snow depth values with in situ observations from the synoptic stations on a monthly time scale. Four statistical indicators Kling Gupta Efficiency (KGE), Taylor diagram metrics, Pearson correlation coefficient, and Root Mean Square Error (RMSE) were calculated using the R statistical environment. The results demonstrated that the Quantile Mapping bias-correction technique provides reliable adjustment for snow-related parameters and can effectively improve reanalysis accuracy. Overall, the ERA5-Land dataset showed a higher capability in detecting wintertime snow depth variations across northwestern Iran compared to MERRA-2. The best performance of ERA5-Land was observed at the Mahabad, Khoy, Tabriz, Ahar, Urmia, and Maragheh stations. Furthermore, the analysis revealed a spatial pattern of uncertainty, with an increase in ERA5-Land snow depth uncertainty from the western toward the eastern parts of northwestern Iran. These findings indicate that ERA5-Land provides robust and reliable estimates of snow depth in winter conditions, particularly over the complex topography of northwestern Iran, although its accuracy decreases slightly eastward due to localized climatic and terrain influences.
Introduction
Snow is an important phenomenon in mountainous areas and at high latitudes. Since snow causes transportation and agricultural problems due to temperature drops, it requires careful investigation. Snow evaluation without reliable data cannot produce effective and meaningful results. Due to the constraints of mountainous regions in the construction and maintenance of ground stations and the limited number of synoptic stations, data coverage is often insufficient. One suitable alternative is the use of reanalysis datasets, which can provide valuable information for researchers if they have appropriate spatial resolution and are compatible with the study area. Therefore, the accuracy and effectiveness of these datasets must be evaluated prior to their application.
Material and Methods
After obtaining monthly data from both reanalysis datasets, the snow depth parameter was converted from meters to centimeters to ensure consistency. The accuracy of the datasets was initially evaluated using a Taylor diagram and compared with observational data. For this purpose, 30% of the study stations (three stations) were selected based on their spatial distribution, climatic variability, distance between stations (km), and elevation differences. Due to the weak performance of raw snow depth data from both datasets in the Taylor diagram, the outputs of the two reanalysis datasets were downscaled using the Quantile Mapping method. In the next step, the downscaled results for the three selected stations were compared with the raw snow depth values for the same stations. The calculations indicated that the downscaled results performed better; therefore, subsequent analyses were conducted using downscaled data for all 10 stations. Considering the geographical location and mountainous topography of the study area, as well as favorable conditions for snowfall during winter, snow depth analyses were carried out for December, January, and February. Statistical indices including the Pearson correlation coefficient, Root Mean Square Error (RMSE), and Kling Gupta Efficiency (KGE) were calculated using the R software to evaluate the seasonal and spatial variability of snow depth based on ERA5-Land and MERRA-2 reanalysis datasets.
For each station, the Percentage Bias Index (PBIAS) was calculated separately for the outputs of the two reanalysis datasets for each winter month, and the difference between the two datasets was averaged for each station. The resulting bias differences were zoned and mapped in a GIS environment for each three-month period. Finally, to provide a clear visual comparison of snow depth estimates from the two reanalysis datasets across the 10 stations in northwestern Iran, box plots were generated using R software.
Results and Discussion
According to the Taylor diagram results, downscaling of snow depth data from both reanalysis datasets is necessary before application. In this study, the ERA5-Land reanalysis dataset performed better than MERRA-2 in both spatial (station-based) and temporal (seasonal) analyses. Based on the Taylor diagram, the downscaled outputs from both datasets were used for further snow depth evaluation.
Bar graphs were produced for each synoptic station to represent winter snow depth over the 20-year study period. Among the 10 synoptic stations, Tabriz, Khalkhal, Meshkin Shahr and Urmia stations representing different elevations, geographical locations, and climatic conditions were selected for detailed analysis of the winter season.
In January, the correlation coefficient, RMSE and KGE values indicated that snow depth estimates from the ERA5-Land reanalysis dataset were more accurate than those from MERRA-2. Compared with MERRA-2, ERA5-Land exhibited lower RMSE values and higher correlation and KGE values. The highest accuracy of the ERA5-Land dataset was observed at the Tabriz synoptic station. In December, the best performance of ERA5-Land was recorded at the Urmia station, with a KGE value of 0.9.
The seasonal spatial distribution based on bias differences between the two datasets showed that the maximum spatial differences in snow depth occurred in January at the Tabriz and Maragheh stations (central part of the study area) and in February and December at the Ahar station (northern part of the study area).
Conclusion
The results showed that the raw outputs of ERA5-Land and MERRA-2 are not suitable for direct snow depth estimation in northwestern Iran, making downscaling unavoidable prior to application. The findings confirmed the reliability of the Quantile Mapping (QM) bias-correction method for snow-related variables. Overall, ERA5-Land demonstrated a higher capability in detecting winter snow depth compared with MERRA-2, which showed limited reliability except at a few stations and months; therefore, caution is required when using MERRA-2 data in the study area. ERA5-Land exhibited its best performance at the Mahabad, Khoy, Ahar, Urmia and Maragheh stations, with KGE values close to unity, whereas the weakest performance of both datasets occurred at the Sarab station. Temporally, ERA5-Land performed best in December, while the poorest performance was observed for MERRA-2 in February. The maximum bias difference between the two reanalysis datasets over the 20-year period reached 542 cm in December, with the largest spatial bias differences occurring at the Ahar station (December and February) and at the Tabriz and Maragheh stations (January). The minimum spatial bias difference was detected at the Ardabil station in January.
In addition, the uncertainty of ERA5-Land snow depth estimates increased from west to east across northwestern Iran. The results are largely consistent with previous studies reporting the superior performance of ERA5 over MERRA-2 in mountainous regions, while partially differing from studies suggesting systematic underestimation or latitude-dependent uncertainty patterns. Overall, this study provides a robust seasonal-scale assessment of reanalysis-based snow depth over a long-term period (2004–2023) using a multi-metric evaluation framework.
Keywords
Subjects
