Evaluation of the ECMWF Precipitation Product Over Various Regions of Iran

+ Author Affiliations + Find other works by these authors

PDF

  • Easy access to accurate and reliable climate data is a crucial concern in hydrological modeling. In this regard, gridded climate data have recently been provided as an alternative to observational data. However, those data should be first evaluated and corrected to guarantee their validity and accuracy. This study offered a new approach to assess the ECMWF gridded precipitation data based on some indicators including, correlation coefficient (CC), and normalized root-mean-square error (NRMSE), and absolute error (AE) in daily and monthly intervals (2007–2017) across different climatic and geographical areas of Iran. Besides, an artificial neural network (ANN) model was utilized to correct the ECMWF precipitation product. According to the results, NRMSE was less than 2 (in 93% of stations) and 5 (in 63% of stations) on monthly and daily scales, respectively. Moreover, CC was above 0.6 in 58% and 94% of stations on daily and monthly scales, respectively. The AEs values were between −0.5 to 0.5, in 80% (daily scale) and 50% (monthly scale) of stations. Having corrected the ECMWF precipitation product by ANN, the number of stations with NRMSE < 5 increased from 63% to 74% on a daily scale, whereas the number of stations with NRMSE < 2 reached 95% from 93% on a monthly scale. The results also showed that the number of stations with CC > 0.6 increased from 58% to 87% on a daily scale.
  • 加载中
  • Fig. 1.  The distribution map of ground synoptic stations and different studied regions.

    Fig. 2.  The flowchart of the research methodology.

    Fig. 3.  The ANN structure used in this study.

    Fig. 4.  Distribution map of CC, NRMSE, and AE on daily and monthly scales between gridded precipitation and in-situ measurements within the 2007–2017 period.

    Fig. 5.  Daily and monthly CC, NRMSE, and AE at different longitudes, latitudes, altitudes in Iran.

    Fig. 6.  Temporal variation of performance metrics employed in the current study.

    Fig. 7.  Improvement map of CC, NRMSE on daily and monthly scales.

    Fig. 8.  Comparing CC, NRMSE, and AE on daily and monthly scales in eight different precipitation regions and different longitudes, latitudes, altitudes.

    Table 1.  Zonation of different climatic regions of Iran (Modarres, 2006)

    RegionDescription
    G1Arid and semi-arid regions in Central Iran
    G2High-altitude regions around the regions of the first group
    G3Cold regions of the northwest (Urmia, Ardabil, Zanjan, and Tabriz)
    G4Regions along the Persian Gulf in the south of Iran
    G5Regions located in Zagros mountains with relatively lower precipitation than the seventh group
    G6Coastal regions of the Caspian Sea in the north of Iran with relatively higher precipitation than the eighth group
    G7Regions located in Zagros mountains with relatively higher precipitation than the fifth group
    G8Coastal regions of the Caspian Sea in the north of Iran with relatively lower precipitation than the sixth group
    Download: Download as CSV
  • [1]

    Abdolmanafi, A., B. Saghafian, and S. Aminyavari, 2021: Evaluation of global ensemble prediction models for forecasting medium to heavy precipitations. Meteor. Atmos. Phys., 133, 15–26. doi: 10.1007/s00703-020-00731-8.
    [2]

    Alizadeh-Choobari, O., P. Adibi, and P. Irannejad, 2018: Impact of the El Niño–Southern Oscillation on the climate of Iran using ERA-Interim data. Climate Dyn., 51, 2897–2911. doi: 10.1007/s00382-017-4055-5.
    [3]

    Aminyavari, S., B. Saghafian, and M. Delavar, 2018: Evaluation of TIGGE ensemble forecasts of precipitation in distinct climate regions in Iran. Adv. Atmos. Sci., 35, 457–468. doi: 10.1007/s00376-017-7082-6.
    [4]

    Bai, L., C. X. Shi, L. H. Li, et al., 2018. Accuracy of CHIRPS satellite-rainfall products over mainland China. Remote Sens., 10, 362, doi: 10.3390/rs10030362.
    [5]

    Fallah, A., G. Rakhshandehroo, P. Berg, et al., 2020: Evaluation of precipitation datasets against local observations in southwestern Iran. Int. J. Climatol., 40, 4102–4116. doi: 10.1002/joc.6445.
    [6]

    Guo, Y. Q., C. C. Huang, J. L. Pang, et al., 2017: Reconstruction palaeoflood hydrology using slackwater flow depth method in the Yanhe River valley, middle Yellow River basin, China. J. Hydrol., 544, 156–171. doi: 10.1016/J.JHYDROL.2016.11.017.
    [7]

    Hu, Z., Q. Hu, C. Zhang, et al., 1955. Journal of geophysical research. Nature, 175, 238, doi: 10.1038/175238c0. (查阅所有网上资料,未找到对应的作者信息,请联系作者确认).
    [8]

    Hu, Z. Y., Q. M. Zhou, X. Chen, et al., 2018: Evaluation of three global gridded precipitation data sets in central Asia based on rain gauge observations. Int. J. Climatol., 38, 3475–3493. doi: 10.1002/joc.5510.
    [9]

    Hyndman, R. J., and A. B. Koehler, 2006: Another look at measures of forecast accuracy. Int. J. Forecast., 22, 679–688. doi: 10.1016/j.ijforecast.2006.03.001.
    [10]

    Javadi, S., S. M. H. Shahdany, A. Neshat, et al., 2020: Multi-parameter risk mapping of Qazvin aquifer by classic and fuzzy clustering techniques. Geocarto Int., 1–23. doi: 10.1080/10106049.2020.1778099..
    [11]

    Kharanagh, S. G., M. E. Banihabib, and S. Javadi, 2020: An MCDM-based social network analysis of water governance to determine actors’ power in water-food-energy nexus. J. Hydrol., 581, 124382. doi: 10.1016/j.jhydrol.2019.124382.
    [12]

    Kolachian, R., and B. Saghafian, 2019: Deterministic and probabilistic evaluation of raw and post processed sub-seasonal to seasonal precipitation forecasts in different precipitation regimes. Theor. Appl. Climatol., 137, 1479–1493. doi: 10.1007/s00704-018-2680-5.
    [13]

    Kurtzman, D., S. Navon, and E. Morin, 2009: Improving interpolation of daily precipitation for hydrologic modelling: Spatial patterns of preferred interpolators. Hydrol. Process., 23, 3281–3291. doi: 10.1002/hyp.7442.
    [14]

    Lin, X. S., S. S. Yu, and H. Y. Wang, 2012: Spatial simulation of precipitation resources in Chongqing. Adv. Mater. Res., 518–523,4261-4265. doi: 10.4028/WWW.SCIENTIFIC.NET/AMR.518-523.4261.
    [15]

    Lloyd, C. D., 2005: Assessing the effect of integrating elevation data into the estimation of monthly precipitation in Great Britain. J. Hydrol., 308, 128–150. doi: 10.1016/j.jhydrol.2004.10.026.
    [16]

    Malmir, M., S. Javadi, A. Moridi, et al., 2021: A new combined framework for sustainable development using the DPSIR approach and numerical modeling. Geosci. Front., 12, 101169. doi: 10.1016/j.gsf.2021.101169.
    [17]

    Medina, H., D. Tian, F. R. Marin, et al., 2019: Comparing GEFS, ECMWF, and postprocessing methods for ensemble precipitation forecasts over Brazil. J. Hydrometeorol., 20, 773–790. doi: 10.1175/JHM-D-18-0125.1.
    [18]

    Moazami, S., S. Golian, Y. Hong, et al., 2016: Comprehensive evaluation of four high-resolution satellite precipitation products under diverse climate conditions in Iran. Hydrol. Sci. J., 61, 420–440. doi: 10.1080/02626667.2014.987675.
    [19]

    Modarres, R., 2006: Regional precipitation climates of Iran. J. Hydrol, , 45,15–29.
    [20]

    Modarres, R., and A. Sarhadi, 2009: Rainfall trends analysis of Iran in the last half of the twentieth century. J. Geophys. Res.: Atmos., 114, D03101. doi: 10.1029/2008JD010707.
    [21]

    Raziei, T., and F. Sotoudeh, 2017: Investigation of the accuracy of the European Center for Medium Range Weather Forecast (ECMWF) in forecasting observed precipitation in different climates of Iran. J. Earth Sp. Phys., 43, 133–147. doi: 10.22059/JESPHYS.2017.57958.(inFarsi.
    [22]

    Raziei, T., I. Bordi, L. S. Pereira, et al., 2010: Space-time variability of hydrological drought and wetness in Iran using NCEP/NCAR and GPCC datasets. Hydrol. Earth Syst. Sci., 14, 1919–1930. doi: 10.5194/hess-14-1919-2010.
    [23]

    Razmi, R., S. Balyani, and M. R. M. Daneshvar, 2017: Geo-statistical modeling of mean annual rainfall over the Iran using ECMWF database. Spat. Inf. Res., 25, 219–227. doi: 10.1007/s41324-017-0097-3.
    [24]

    Saeedi, M., A. Sharafati, and A. Tavakol, 2021: Evaluation of gridded soil moisture products over varied land covers, climates, and soil textures using in situ measurements: A case study of Lake Urmia Basin. Theor. Appl. Climatol., 145, 1053–1074. doi: 10.1007/s00704-021-03678-x.
    [25]

    Sagar, S. K., M. Rajeevan, S. V. B. Rao, et al., 2017: Prediction skill of rainstorm events over India in the TIGGE weather prediction models. Atmos. Res., 198, 194–204. doi: 10.1016/j.atmosres.2017.08.017.
    [26]

    Salimi, A. H., J. M. Samakosh, E. Sharifi, et al., 2019: Optimized artificial neural networks-based methods for statistical downscaling of gridded precipitation data. Water, 11, 1653. doi: 10.3390/w11081653.
    [27]

    Sharafati, A., S. Nabaei, and S. Shahid, 2020: Spatial assessment of meteorological drought features over different climate regions in Iran. Int. J. Climatol., 40, 1864–1884. doi: 10.1002/joc.6307.
    [28]

    Sharifi, E., R. Steinacker, and B. Saghafian, 2016: Assessment of GPM-IMERG and other precipitation products against gauge data under different topographic and climatic conditions in Iran. Remote Sens., 8, 135. doi: 10.3390/rs8020135.
    [29]

    Shobeiri, S., A. Sharafati, and A. Neshat, 2021: Evaluation of different gridded precipitation products in trend analysis of precipitation features over Iran. Acta Geophys., 69, 959–974. doi: 10.1007/s11600-021-00595-5.
    [30]

    Shrestha, N. K., F. M. Qamer, D. Pedreros, et al., 2017: Evaluating the accuracy of Climate Hazard Group (CHG) satellite rainfall estimates for precipitation based drought monitoring in Koshi basin, Nepal. J. Hydrol.: Reg. Stud., 13, 138–151. doi: 10.1016/j.ejrh.2017.08.004.
    [31]

    Sodoudi, S., A. Noorian, M. Geb, et al., 2010: Daily precipitation forecast of ECMWF verified over Iran. Theor. Appl. Climatol., 99, 39–51. doi: 10.1007/s00704-009-0118-9.
    [32]

    Taylor, J. R., 1997: An Introduction to Error Analysis: The Study of Uncertainties in Physical Measurements. 2nd ed. University Science Books, Sausalito, CA. (查阅所有网上资料,未找到对应的页码信息,请联系作者确认).
  • 加载中
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Evaluation of the ECMWF Precipitation Product Over Various Regions of Iran

    Corresponding author: Ahmad Sharafati, asharafati@gmail.com
  • 1. Department of GIS/RS, Faculty of Natural Resources and Environment, Science and Research Branch, Islamic Azad University, Tehran, Iran
  • 2. Department of Civil Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran

Abstract: Easy access to accurate and reliable climate data is a crucial concern in hydrological modeling. In this regard, gridded climate data have recently been provided as an alternative to observational data. However, those data should be first evaluated and corrected to guarantee their validity and accuracy. This study offered a new approach to assess the ECMWF gridded precipitation data based on some indicators including, correlation coefficient (CC), and normalized root-mean-square error (NRMSE), and absolute error (AE) in daily and monthly intervals (2007–2017) across different climatic and geographical areas of Iran. Besides, an artificial neural network (ANN) model was utilized to correct the ECMWF precipitation product. According to the results, NRMSE was less than 2 (in 93% of stations) and 5 (in 63% of stations) on monthly and daily scales, respectively. Moreover, CC was above 0.6 in 58% and 94% of stations on daily and monthly scales, respectively. The AEs values were between −0.5 to 0.5, in 80% (daily scale) and 50% (monthly scale) of stations. Having corrected the ECMWF precipitation product by ANN, the number of stations with NRMSE < 5 increased from 63% to 74% on a daily scale, whereas the number of stations with NRMSE < 2 reached 95% from 93% on a monthly scale. The results also showed that the number of stations with CC > 0.6 increased from 58% to 87% on a daily scale.

    • Precipitation directly affects human life and plays an undeniable role in growing and developing most countries. Therefore, it is essential to accurately estimate precipitation in hydrological models, water resources management, hydrological studies, groundwater modeling, and prediction of floods and droughts to prevent natural disasters (Javadi et al., 2020; Kharanagh et al., 2020; Sharafati et al., 2020; Malmir et al., 2021). Therefore, the accurate estimation of precipitation is a serious challenge.

      With an acceptable scope and a variety of data, gridded precipitation products can be used as reliable data to efficiently and practically remove limitations on access to precipitation data with spatial-temporal resolution. Satellites can provide easily accessible and usable data on precipitation distribution and rate directly through analysis of microwaves, indirectly through measurement of visible and infrared waves, or a combination of both methods (Shobeiri et al., 2021).

      Using ground rain gauges to measure precipitation, especially in developing countries, would face certain problems such as high costs, incomplete coverage of seas, deserts, and mountainous regions, and the pointiness nature of data. There are no full-coverage precipitation measurement networks regarding radars due to the high costs of establishing and maintaining infrastructure and coverage limitations in mountainous regions (Medina et al., 2019; Abdolmanafi et al., 2021).

      Gridded precipitation products are precious due to their high spatial-temporal resolution and globally pervasive coverage. They can complete ground rain gauges and radar-based precipitation data; however, the extensive capabilities of gridded products to estimate precipitation depends on both identification and correction of error factors (Aminyavari et al., 2018; Fallah et al., 2020).

      (Raziei et al., 2010) analyzed GPCC and NCEP data to study the diversity of drought and water sufficiency based on hydrology. According to their results, those data yielded acceptable outputs for studying drought in the southeast and northwest of Iran. (Sodoudi et al., 2010) evaluated ECMWF data in daily precipitation forecasting models in Iran and reported that ECMWF data produced better outputs in high-altitude and mountainous areas than in low-altitude regions. (Sharifi et al., 2016) analyzed ECMWF, TRMM, and GPM products compared with ground-based data in different climatic and altitudinal conditions. They reported that IMERG data outperformed the other types of gridded data. (Sagar et al., 2017) used the TIGGE databases (e.g., ECMWF, NCEP, and UKMO) in forecasting models and reported that ECMWF data had the lowest error rate and the highest CC. (Razmi et al., 2017) employed ECMWF data in bio-statistical models of average annual precipitation in Iran and reported that precipitation model diversity was in direct correlation with the contour line of every region in Iran. (Raziei and Sotoudeh, 2017) drew a comparison between ECMWF product and observational precipitation in different climates of Iran. According to their results, different ECMWF gridded precipitation data yielded different outputs in various regions of Iran. (Alizadeh-Choobari et al., 2018) used ECMWF (ERA-INTERIM) data to analyze the effect of El Niño–Southern Oscillation (ENSO) on Iran’s climate. Their results indicated that the ENSO cycle affected precipitation diversity in Iran. (Aminyavari et al., 2018) employed precipitation forecasting models to evaluate TIGGE outcomes in different climates of Iran. According to their results, NCEP data showed a better performance in most stations. (Medina et al., 2019) used GEFS and ECMWF data in forecasting models and reported that the ECMWF forecasting model outperformed the GEFS model. They also stated that the processed GEFS data yielded better results than the processed ECMWF data after data processing operations were performed. (Fallah et al., 2020) analyzed gridded precipitation data in comparison with observational data in the southwest of Iran and reported that ECMWF data yielded better outcomes in most stations. (Abdolmanafi et al., 2021) used ECMWF product in Iran to analyze precipitation forecasting models. According to their results, ECMWF outperformed the other types of products in forecasting models.

      The possibility of forecasting precipitation has significantly increased by using gridded precipitation data with global coverage on the scale of a few hours. Nevertheless, due to the uncertainty of gridded precipitation data, it is necessary to evaluate and correct gridded precipitation data in comparison with ground rain gauge data on a regional scale with respect to the effects of different factors on gridded precipitation estimation such as geographical location, topography, humidity, seasons, and precipitation rate. In this way, (Salimi et al., 2019) used an ANN to statistically downscale satellite-based precipitation data. According to their results, forecasting precipitation was greatly affected by the cloud optical thickness.

      This study considered spatial differences of precipitation rates on different temporal scales across Iran to evaluate, calibrate, validate, and analyze ECMWF precipitation product. The NRMSE and CC are two statistical indices used in this study for assessing gridded data and analyze their consistency with observational data. After that, an artificial neural network (ANN) was employed to correct the gridded precipitation. According to the literature review of gridded precipitation products, it can be claimed that analysis and correction of ECMWF data are unique at this level in Iran.

    2.   Study area and data
    • With an area of 1,648,195 km2, Iran is located in the southern half of the northern hemisphere between latitudes 25º3’ and 39º47’ N and 44º5’ and 63º18’ E. Iran is a mountainous country having two major mountain ranges, i.e., Alborz and Zagros, as well as many altitudinal variations in its mountain ranges. The altitudes of Iran’s mountains prevent the effects of wet winds blown from the Caspian Sea, the Mediterranean Sea, and the Persian Gulf to Central Iran. Therefore, the outer slopes of these mountains have a humid climate, whereas the inner slopes are dry and arid. As a result, the spatial-temporal distribution of precipitation is not uniform across Iran, with many climatic diversities (Modarres, 2006). There are many differences in annual precipitation rates of various regions, in some of which the total precipitation of a year occurs during only a few hours (Modarres and Sarhadi, 2009).

      (Modarres, 2006) evaluated the spatial distribution of precipitation across Iran using the hierarchical cluster technique and divided Iran into eight regions based on ground rain gauges to determine the regional precipitation patterns (Fig. 1).

      Figure 1.  The distribution map of ground synoptic stations and different studied regions.

      Table 1 demonstrates the eight regions used in this study for the spatial analysis of ECMWF precipitation data.

      RegionDescription
      G1Arid and semi-arid regions in Central Iran
      G2High-altitude regions around the regions of the first group
      G3Cold regions of the northwest (Urmia, Ardabil, Zanjan, and Tabriz)
      G4Regions along the Persian Gulf in the south of Iran
      G5Regions located in Zagros mountains with relatively lower precipitation than the seventh group
      G6Coastal regions of the Caspian Sea in the north of Iran with relatively higher precipitation than the eighth group
      G7Regions located in Zagros mountains with relatively higher precipitation than the fifth group
      G8Coastal regions of the Caspian Sea in the north of Iran with relatively lower precipitation than the sixth group

      Table 1.  Zonation of different climatic regions of Iran (Modarres, 2006)

      In this study, the ECMWF precipitation product was used as the gridded dataset. In addition, the reference precipitation data were collected from 70 synoptic stations of the Iran Meteorological Organization for the 2007–2017 period. The set of ground gauge precipitation data is hereafter referred to as the in-situ measurements.

      This study also used ECMWF precipitation data (downloaded in the grib format from ECMWF’s website) with a 0.5º × 0.5º spatial resolution. Moreover, the set of ECMWF precipitation data is hereafter referred to as the gridded precipitation. This is because the spatial distance of every datum from the next datum is 0.5º in a pointy distribution of data.

    • Previous studies have been used the “pixel-to-pixel” and “point-to-pixel” methods to evaluate satellite precipitation products. In the first method, the in-situ observation data are determined through the pixels by an interpolation method. The resulting pattern is then compared with satellite precipitation data by the pixel-to-pixel method. The second method, known as the point-to-pixel approach, compares the ground precipitation data with that obtained in the pixel of satellite precipitation data. Moreover, it is possible to compare the interpolated satellite precipitation data with ground data through the point-to-pixel approach (Shrestha et al., 2017).

      (Bai et al., 2018) indicated that the point-to-pixel method provided more accurate results in comparison with the pixel-to-pixel approach. Hence, the point-to-pixel method was used in this study to evaluate ECWMF data against ground observational data due to the limited number of meteorological stations. Interpolation is an approach to estimate the precipitation depth at the site of a station using the precipitation data collected from the nearest neighboring points. The literature (Kurtzman et al., 2009) reported the excellent performance of the inverse distance weighting (IDW) method in the provision of the precipitation pattern than other methods such as kriging, regression, and Thiessen methods. The inverse distance weighted (IDW) interpolation method assumes that precipitation loses impact as distance increases; in other words, the precipitation points closer to stations have greater impacts than the precipitation points farther. In this method, distance is used as a known variable in estimating the precipitation depth of stations which are calculated by averaging precipitation in the nearest points (Shobeiri et al., 2021). Finally, satellite precipitation data are estimated by the following equation using the IDW method at the ground station:

      $$ R={\sum }_{1}^{n}{r}_{i}{\omega }_{i}. $$ (1)

      where R represents satellite precipitation interpolated value at the ground station, n the number of all satellite points within the penetration radius, $ {\omega }_{i} $ the distance and weight coefficient, $ {r}_{i} $ the satellite precipitation value within the ground station penetration radius. The weight coefficient is calculated as follows:

      $$ {\omega }_{i}=\frac{\frac{1}{{d}_{i}^{2}}}{\displaystyle\sum _{1}^{n}\frac{1}{{d}_{i}^{2}}}. $$ (2)

      where $ {d}_{i} $ represents the distance of the ground station and the satellite precipitation point, and the exponent 2 shows the distance power. Several studies (Lloyd, 2005; Lin et al., 2012) have shown that precipitation interpolation errors are reduced by considering an exponent of 2 for the distance power.

    3.   Method
    • The observational precipitation data of 70 synoptic stations in the period of 2007–2017 period was collected from the Iran Meteorological Organization. Reliable data are the prerequisite for any hydrological and water resource studies. Given continuous and discontinuous (discretized) gaps in most hydrological data, such as precipitation data, due to the lack of statistics recording, elimination of incorrect statistics, and the failure or destruction of measurement devices, it is necessary to estimate and evaluate these data. Obviously, before ensuring the accuracy and quality of observational and time-series data, such data cannot be used to extract subsequent results (Saeedi et al., 2021). About 3.5% of data in this study were obtained from ungauged observational stations or those with suspicious data. The linear regression method was used to reproduce data considering corresponding values in adjacent stations. The weight factor can be estimated from the correlation coefficient or the ratio of areas or distances between ungauged and gauged stations (Guo et al., 2017). After data correction and reconstruction, the time series of observational data was developed. In this study, observational data were used as the reference for precipitation data evaluation. After the gridded precipitation data were collected, the precipitation depth was interpolated at the sites of stations.

      Moreover, a few statistical indices were employed to evaluate and compare gridded data with observational data. The ANN model was then utilized to correct the ECWMF data obtained in the location of stations. When data correction was performed, all precipitation data were validated to finally allow for the use of gridded precipitation data by analyzing the results from different temporal scales in various regions of Iran. Figure 2 shows the flowchart of the methodology used in the current study.

      Figure 2.  The flowchart of the research methodology.

    • After the gridded precipitation data were extracted, they were compared with the in-situ measurements through statistical indices on daily, monthly, and yearly scales at observational stations. In this way, two different performance indices, including CC (Eq.4), and NRMSE (Eq.5), and AE (Eq.6), were employed. In fact, CC indicates the intensity and type of relationship between gridded precipitation and in-situ measurements. This coefficient ranges between −1 and +1. The closer the CC to −1 or +1, the stronger the relationship, whereas the closer the CC to 0, the weaker the relationship (Taylor, 1997).

      Furthermore, NRMSE is the normalized version of RMSE, obtained by dividing the RMSE by the mean of in-situ measurements. NRMSE denotes the standard deviation of differences between gridded precipitation and in-situ measurements and indicates the precipitation estimation error. It ranges from zero to infinity. The closer the NRMSE to zero, the smaller the estimation error is than the observational data (Hyndman and Koehler, 2006). The third index used in this study was the absolute error (AE), which should be ideally equal to zero. Positive and negative AEs respectively show overestimation and underestimation of the model. This parameter is indicative of the method’s precision and the error value (Hu et al., 1955, 2018).

      $$ \mathrm{C}\mathrm{C}=\frac{\sum _{i=1}^{n}({\mathrm{P}}_{\mathrm{T}\left(\mathrm{i}\right)}-\widehat{{\mathrm{P}}_{\mathrm{T}}})({\mathrm{P}}_{\mathrm{G}\left(\mathrm{i}\right)}-\widehat{{\mathrm{P}}_{G}})}{\sqrt{{\sum _{i=1}^{n}({\mathrm{P}}_{\mathrm{T}\left(\mathrm{i}\right)}-\widehat{{\mathrm{P}}_{\mathrm{T}}})}^{2}}*{\sum _{i=1}^{n}({\mathrm{P}}_{\mathrm{G}\left(\mathrm{i}\right)}-\widehat{{\mathrm{P}}_{\mathrm{G}}})}^{2}} , $$ (3)
      $$ \mathrm{N}\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}=\frac{\sqrt{\frac{1}{\mathrm{n}}\sum _{\mathrm{i}=1}^{\mathrm{n}}{\left({\mathrm{P}}_{{\mathrm{T}}_{\mathrm{i}}}-{\mathrm{P}}_{{\mathrm{G}}_{\mathrm{i}}}\right)}^{2}}}{\stackrel-{{\mathrm{P}}_{\mathrm{G}}}} , $$ (4)
      $$ \mathrm{A}\mathrm{E}=\frac{\sum \left({\mathrm{P}}_{{\mathrm{T}}_{\mathrm{i}}}-{\mathrm{P}}_{{\mathrm{G}}_{\mathrm{i}}}\right)}{n} . $$ (5)

      In the above equations, $ {\mathrm{P}}_{{\mathrm{G}}_{\mathrm{i}}} $ and $ {\mathrm{P}}_{{\mathrm{T}}_{\mathrm{i}}} $ refer to the ith daily, monthly, or yearly in-situ measurements and gridded precipitation data, respectively, whereas n indicates the interval (n = 4015 on a daily scale, n = 132 on a monthly scale, and n = 11 on a yearly scale).

    • An ANN model was employed to correct gridded-based precipitation data. The ANN was created for all climates of Iran to correct gridded precipitation data on daily and monthly scales separately. Gridded precipitation was then corrected, and statistical indices were calculated.

      Inspired by the human brain, an ANN is a calculating system that consists of a few processing units (neurons) that link the input set to the output set. The ANN components are neurons, input layer, hidden layer, an output layer, weight, and transfer function. Usually, an ANN is trained to determine a specific output based on a specific input. Generally, an ANN is utilized for estimation, pattern recognition, and classification (Moazami et al., 2016).

      This study employed a cascade–forward neural network (CFN) having a hyperbolic tangent sigmoid transfer function with three neurons in the hidden layer and a linear transfer function with one neuron in the output layer (Fig. 3).

      Figure 3.  The ANN structure used in this study.

      The in-situ measurements were used to validate the raw and corrected gridded precipitation data over daily and monthly scales at 70 synoptic stations within the 2007–2017 period. The validation results of the corrected gridded precipitation data were compared with the in-situ measurements to extract the appropriate model for daily and monthly precipitation patterns in different climates.

    4.   Results and discussion
    • In this section, raw gridded-based precipitation data are first evaluated using CC in different spatial-temporal intervals to determine the usability or non-usability of gridded-based precipitation data in various regions on different temporal scales. After that, an ANN is employed to correct gridded-based precipitation data. Finally, the efficiency of the ANN model is also analyzed.

    • To compare the original gridded-based precipitation data with in-situ measurements based on the different time scales, several performance metrics, including CC, NRMSE, and AE, were obtained. The CC was above 0.5 in more than 92% and 96% of stations on daily and monthly scales, respectively. Therefore, CC calculation indicated a significant relationship between in-situ measurements and gridded precipitation on daily and monthly scales. In contrast, yearly data were excluded from the analysis process due to their low correlation.

      Figure 4 depicts the pattern of CC, NRMSE, and AE on daily and monthly scales. The daily CC is often greater than 0.5, and most of the regions with the highest CC were observed in mainly the southern low-altitude areas of Iran. In contrast, the lowest CC was reported in the coastal areas of the Caspian Sea. However, the lowest NRMSE was mainly observed in the high-altitude regions. (Sharifi et al., 2016) reported that ECMWF data did not have enough quality to replace ground precipitation data in the coastal regions of the Caspian Sea. According to the results, the lowest and highest daily AE were respectively observed in the low-altitude central regions of Iran in G1 and the northwest regions of Iran in G3. The AE values were between −0.5 to 0.5 in 56 stations.

      Figure 4.  Distribution map of CC, NRMSE, and AE on daily and monthly scales between gridded precipitation and in-situ measurements within the 2007–2017 period.

      At the same time, the analysis of NRMSE indicated acceptable results in the marginal areas of the Persian Gulf and G2. According to (Aminyavari et al., 2018; Abdolmanafi et al., 2021), the highest CC was observed in mountainous areas of G5 and G2, whereas the highest NRMSE was mainly reported in the marginal areas of the Caspian Sea. Thus, the findings of both studies are consistent with the results of this study. The monthly CC was mainly above 0.7 in all areas except for a few regions in the northern margin of the Caspian Sea in G6 and G1, whereas the lowest NRMSE was observed in all regions except for G3 and G1. Like the daily AE, the highest monthly AE was also observed in the central desert areas of Iran in G1, and the lowest monthly AE was observed in the G3 region. (Aminyavari et al., 2018) analyzed TIGGE data and found the highest AE in the northwest areas of Iran. According to (Sharifi et al., 2016), the highest CC consistency was observed in marginal areas of the Persian Gulf, whereas the best results of analyzing NRMSE were reported in G5.

      Moreover, (Kolachian and Saghafian, 2019) stated that a region of G2 located in the northeast of Iran had the highest monthly CC. This finding was consistent with the results of this study. In general, the pattern of daily and monthly CCs confirmed the highest correlation of precipitation data on a monthly scale.

      Figure 5 demonstrates the values of CC, NRMSE, and AE between gridded precipitation and in-situ measurements at different longitudes, latitudes, and altitudes on daily and monthly scales. As shown in Fig. 5a, the highest CC values were reported 0.65 and 0.84 on daily and monthly scales, respectively, at the longitude 59º, whereas the lowest values were obtained 0.57 and 0.72 at the longitude 53º. Figure 5b indicates that the highest values of NRMSE between gridded precipitation and in-situ measurements were reported 5.02 at the longitude 51º and 1.53 at the longitude 55º on daily and monthly scales, respectively. The lowest values were obtained 1.8 at the longitude 59º and 0.9 at the longitude 51º. As shown in Fig. 5c, the highest daily and monthly AEs were found at the longitude less than 47°, and the lowest corresponding values were observed at the longitude in the range of 53° to 55°. According to the results, there was no specific relationship between CC and longitudinal variations. (Fallah et al., 2020) indicated that the highest and lowest CCs were observed at the longitudes 51º and 48º, respectively.

      Figure 5.  Daily and monthly CC, NRMSE, and AE at different longitudes, latitudes, altitudes in Iran.

      From Fig. 5d, it is clear that the highest daily CC was reported 0.71 at the latitude 31º, whereas the highest monthly CC was obtained 0.89 at the latitude 29º. Moreover, the lowest values of CC were reported 0.51 and 0.68 on daily and monthly scales, respectively, at the latitude 37º. According to (Fallah et al., 2020), the highest CC was observed at the latitude of 31º. Figure 5e shows that the highest daily values of NRMSE were respectively reported 5.02 and 1.53 at the latitudes 31º and 35º, whereas the lowest monthly values were obtained 1.73 and 0.9 at the latitudes 37º and 31º, respectively. As shown in Fig. 5f, the highest daily and monthly AEs were observed at a latitude of more than 37° and the lowest corresponding values of less than 27°.

      From Fig. 5g, the highest daily and monthly values of CC were obtained 0.73 and 0.9, respectively, at an altitude of 600 m, whereas the lowest values of CC were reported 0.59 and 0.74 at altitudes of 300 m and 1500 m. Moreover, Fig. 5h indicates that the highest daily and monthly values of NRMSE were respectively reported 4.37 and 1.58 at altitudes of 300 m and 600 m. In contrast, the lowest daily and monthly values were obtained 2.04 and 1.01 at altitudes of 600 m and 1800 m, respectively. As shown in Fig. 5i, the lowest daily and monthly AEs were observed at an altitude of less than 300 m and the highest values in the range of 1200 to 1500 m.

      To assess the temporal variation of performance metrics employed in the current study, their monthly values over Iran are presented in Fig. 6. Figure 6a shows that the highest and lowest consistency between in-situ measurements and gridded data was observed in June and January, respectively. On the other hand, the ECMWF dataset has more agreement with observation over dry months. In general, an incrassating trend is observed in CC values from January (CC: 0.37) to June (CC: 0.62), while uniform variations were found in other months, and CC values were in the range of 0.48 to 0.59. From Fig. 6b, it is evident that the highest and lowest error

      Figure 6.  Temporal variation of performance metrics employed in the current study.

      in ECMWF data was observed in the summer (NRMSE: 2.46 to 2.8) and winter seasons (NRMSE: 1.21 to 1.34), respectively. However, the ECMWF data has good accuracy in rainy months. As shown in Fig. 6c, it is clear that ECMWF overestimates the rainfall depths in all months except August and September. In this regard, the highest overestimation was found in March (AE: 15.75 mm).

    • The artificial neural network model was developed for 70 synoptic states separately to analyze different climates and classifications based on longitudes, latitudes, and altitudes to independently correct gridded precipitation data on daily and monthly scales.

      To measure the improvement provided by the ANN model, the performance indices (e.g., CC, NRSME) were calculated on daily and monthly scales at 70 synoptic stations after the ANN model corrected the gridded precipitation data. Finally, the improvement coefficient (IM) was calculated at every station on daily and monthly scales as follows:

      $$ {IM}_{CC}=\frac{{CC}_{cor}-{CC}_{\mathrm{o}\mathrm{r}\mathrm{g}}}{{CC}_{org}}\times 100\mathrm{ }, $$ (6)
      $$ {IM}_{NRMSE}=\frac{{NRMSE}_{\mathrm{o}\mathrm{r}\mathrm{g}}-{NRMSE}_{\mathrm{c}\mathrm{o}\mathrm{r}}}{{NRMSE}_{\mathrm{o}\mathrm{r}\mathrm{g}}}\times 100\mathrm{ }. $$ (7)

      where, $ {CC}_{org} $ and $ {CC}_{cor} $ are the CC values extracted based on original and corrected gridded precipitation data, respectively. $ {NRMSE}_{org} $ and $ {NRMSE}_{cor} $ are the $ NRMSE $ values obtained based on original and corrected gridded precipitation data, respectively.

      The IM shows how much CC, NRMSE improved after correction. The improvement map was then drawn CC, NRMSE on daily and monthly scales (Fig. 7). The daily IM is usually above zero and shows an increase in $ {CC}_{Cor} $ as opposed to $ {CC}_{org} $. The highest $ {IM}_{\mathrm{C}\mathrm{C}} $ was observed mainly in desert regions of Central Iran, whereas its lowest value was reported in the marginal areas of the Sea of Oman in the southeast of Iran. Similarly, the lowest $ {IM}_{\mathrm{N}\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}} $ was observed in the southern areas of Iran overlooking the Sea of Oman and the Persian Gulf, whereas the highest value was reported in Central Iran and the high-altitude regions of the northwest of Iran.

      Figure 7.  Improvement map of CC, NRMSE on daily and monthly scales.

      The monthly $ {IM}_{\mathrm{C}\mathrm{C}} $ was often positive; however, it was negative only in the marginal areas of the Caspian Sea and only a few regions in the northeast of Iran. It was positive in other areas. The highest monthly $ {IM}_{\mathrm{C}\mathrm{C}} $ was observed along the Zagros Mountain and in the eastern areas of Iran, whereas the lowest monthly $ {IM}_{\mathrm{N}\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}} $ was mainly reported in the southern regions of Iran overlooking the Sea of Oman and the Persian Gulf. Finally, its highest value was reported in the southwest of Iran.

      Fig. 8a indicates corrected gridded precipitation data on daily and monthly scales over different regions in Iran. The lowest and highest values of CC-Daily (cor) were reported 0.47 and 0.66, respectively, in G6 located in the southern coastal strip of the Caspian Sea and G8 located in the eastern coastal strip of the Caspian Sea. Similarly, the lowest and highest values of CC-Monthly (cor) were reported 0.64 and 0.85, respectively, in G6 and G8 on the margin of the Caspian Sea.

      Figure 8.  Comparing CC, NRMSE, and AE on daily and monthly scales in eight different precipitation regions and different longitudes, latitudes, altitudes.

      The minimum and maximum values of NRMSE-Daily (cor) (Fig. 8b) were reported 2.23 in G8 on the west of the coastal strip of the Caspian Sea and 5.78 in G4 on the coastal areas of the Persian Gulf. The minimum and maximum values of NRMSE-Monthly (cor) were obtained 0.46 and 1.27, respectively, in G8 in the east of the coastal strip of the Caspian Sea and G1 in deserts of Central Iran. As shown in Fig. 8c, the highest and lowest AE- Monthly (cor) of 9.45 and −3.47 were respectively observed in G3 and G6. The highest and lowest AE- Daily (cor) of 0.31 and −0.11 were respectively observed in G3 and G6.

      From Fig. 8d, it is evident that the highest values of CC (cor) were reported 0.65 and 0.84 on daily and monthly scales, respectively, at the longitude 51º. In contrast, its lowest values were obtained 0.54 and 0.7 at longitudes 55º and 53º, respectively. Figure 8e shows the highest values of NRMSE (cor) were reported 6.57 and 1.48 on daily and monthly scales at the longitude 55º, whereas its lowest values were obtained 3.11 and 0.71 at longitudes 47º and 49º, respectively.

      According to Fig. 8f, the highest monthly and daily AEs (cor) of 7.5 and 0.25 were respectively observed at the longitudes less than 47°. The lowest monthly and daily AEs (cor) of 1.2 and 0.04 were observed between the longitude 53° to 55°. Figure 8g indicates that the highest CC (cor) values were reported 0.67 and 0.84 on a daily scale at the latitude 31º. In contrast, its lowest values were obtained 0.49 and 0.71 at latitudes 27º and 37º on daily and monthly scales, respectively. From Figure 8h, it is clear that the maximum values of NRSME (cor) were reported 10.07 and 1.83 on daily and monthly, respectively, scales at the latitude 27º. In contrast, the minimum values were obtained at 2.87 and 0.64 on daily and monthly scales, respectively, at the latitude 37º. According to Fig. 8i, the highest monthly and daily AEs (cor) of 8.5 and 0.28 were observed at the latitude of more than 37°. On the other hand, the lowest monthly and daily AEs (cor) of 0.09 and 0.01 were observed at the latitude less than 27°. As shown in Fig. 8j, the highest CC (cor) values were reported 0.67 and 0.86 on daily and monthly scales at an altitude above 1800 m, whereas its lowest values were obtained 0.55 and 0.72 at altitudes 900 m and 300 m, respectively. Moreover, the maximum values of NRMSE (cor) were reported at 5.88 and 1.47 on daily and monthly scales at altitudes 300 m and 600 m, respectively. The minimum values of NRSME (cor) were obtained 3.56 and 0.72 on daily and monthly scales at the altitude 1800 m (Fig. 8k). The highest monthly and daily AEs (cor) of 6.4 and 0.21 were observed at altitudes between 1200 to 1500 m. The lowest monthly and daily AEs (cor) of −0.03 and −0.01 were found at altitudes between 600 to 900 m (Fig. 8l).

    5.   Conclusion
    • This study proposes an approach to assess and correct the ECMWF precipitation product using 70 synoptic stations over different regions in Iran. According to the results of evaluating original ECMWF data against in-situ measurements, there was a satisfactory correlation between those precipitation datasets on daily and monthly scales. According to (Kolachian and Saghafian, 2019), the highest monthly correlation coefficient was observed in part of G2 in the northeast of Iran, consistent with this study’s results. The CC values between original ECMWF data and in-situ measurements were above 0.6 at 58.6% of stations on a daily scale and 94.3% of stations on a monthly scale. However, there was no specific correlation between CC variations and longitudinal variations. In contrast, there was a good correlation between daily and monthly precipitations at different altitudes of Iran. (Aminyavari et al., 2018) also reported the highest CC was observed in mountainous areas of G5 and G2. A lower CC was observed in a few high-altitude regions than in the other mountainous regions. This finding was due to disregarding snowfall at high altitudes in calculating precipitation data (Sharifi et al., 2016). According to the results, the lowest AE was observed in low-altitude central regions in G1 and the highest AE in the western margins of the Caspian Sea in G8. (Aminyavari et al., 2018) analyzed TIGGE data and observed the highest MAE in Guilan.

      According to the separate results of calculating NRMSE at the sites of 70 synoptic stations, it was below 2 at 93% of stations on a monthly scale, whereas it was below 5 at 63% of stations on a daily scale. Most of the regions with the lowest NRMSE were located in high-altitude mountainous areas, a finding which confirmed the high accuracy of gridded-based data in those regions. The reason lies in regular and high-volume precipitations in the mountainous areas of Iran, which simplifies the estimation accuracy of these precipitations compared to low-precipitation areas. According to (Sharifi et al., 2016), the lowest results of analyzing NRMSE were reported in G5, a high-altitude mountainous region, which is consistent with the results of this study.

      According to the results of correcting original ECMWF data through the ANN, this model offered an adequate performance to correct data at the sites of most stations and different geographical and climatic conditions. As the research findings showed, the CC of above 0.6 increased from 58% to 87% on a daily scale. However, the CC values obtained from monthly corrected ECMWF data in some regions were lower than those obtained from original data. Overall, the results indicated that using an ANN to correct data is most applicable to bias correction of ECMWF precipitation data over Iran.

Reference (32)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return