As near-surface wind speed plays a key role in regulating surface evaporation and thus the hydrological cycle, it is crucial to explore its spatio-temporal characteristics. However, in-situ measurements are scarce over the Tibetan Plateau, limiting the understanding of wind speed climate across this high-elevation region. This study explores the climatology of near-surface wind speed over the Tibetan Plateau by using for the first time homogenized observations together with reanalysis products and regional climate model simulations. Measuring stations across the center and the West of the plateau are at higher elevations and display higher mean and standard deviation, confirming that wind speed increases with increasing altitude. By exploring wind characteristics (e.g. seasonal cycle) through cluster analysis, three regions of different wind regimes can be identified: (1) the center of the Tibetan Plateau, characterized by high elevation; (2) the East and the edges of the plateau; and (3) the topographic depression of the Qaidam basin, strongly influenced by the blocking effect of the surrounding mountainous terrain. Owing to the improvements in model physics and data assimilation, the ERA5 reanalysis shows closer agreement to the measured wind conditions than its predecessor ERA-Interim, by being able to reproduce the same three wind regimes. However, the newest ERA5-Land product does not show improvements compared to ERA5 as they share most of the parametrizations. In addition, the two dynamical downscalings of ERA5 analyzed here are not able to capture the observed wind statistics, showing evident biases and discrepancies also when investigating the diurnal variations. Over a region of such complex terrain, the ability of dynamical downscaling in simulating near-surface wind speed may be enhanced by improving the planetary boundary layer parameterization scheme and by better taking into account unresolved topographic features.
Research Article
Climatology of near-surface wind speed from observations and high-resolution climate modelsover the Tibetan Plateau
https://doi.org/10.21203/rs.3.rs-2470952/v1
This work is licensed under a CC BY 4.0 License
Journal Publication
published 16 Sep, 2023
You are reading this latest preprint version
Wind speed
climatology
ECMWF reanalysis
WRF downscaling
Tibetan Plateau
The Tibetan Plateau (TP), often referred to as the “Third Pole”, is the highest and most extensive upland region in the world, stretching for about 1,000 km along latitude and 2,500 km along longitude, with an average elevation over 4,000 m a.s.l. (above sea level) (Yao et al. 2012). It exerts profound thermal and orographic effects on atmospheric circulation patterns on all temporal and spatial scales (Gao et al. 1981), thus playing a crucial role in influencing regional and global climate through both thermal and dynamical mechanisms (Yang et al. 2004). The TP is also referred to as the “water tower of Asia” because it is the source region of several major rivers in Asia (e.g. the Yellow, Yangtze, and Mekong rivers), hence providing water resources to 1.5 billion of people in East and South Asia (Yao et al. 2012). As near-surface (i.e., 10 m height) wind speed (hereafter, NSWS) can alter the surface evapotranspiration rate, thus driving the exchange of heat and water vapor between the surface and the atmospheric boundary layer, it partly regulates the hydrological status and controls the local water cycle (Kang et al. 2010; Yang et al. 2014; Zheng at al. 2009). In addition, with more than 60% of the TP being arid or semi-arid, NSWS is one of the key factors in shaping landforms and the geographical environment (Dong et al. 2017). For these reasons, it is crucial to understand NSWS characteristics over the TP and its variability because of its prominent environmental implications and impacts on the water resources of 10 countries, most of them being the most densely populated countries in the world (IPCC 2019).
However, the lack of sufficient observational data limits the understanding of NSWS over the TP. In fact, compared with other terrestrial regions of the world, observations are scarce here because meteorological stations are sparsely distributed across the plateau due to its vast geographical area with complex topography, steep terrain, high elevations, and harsh environmental conditions (Ma et al. 2020). In addition, not only the density of in-situ stations is sparse, but most available stations are located in valleys (Li et al. 2018). Therefore, the sparse and biased distribution of measuring stations makes it difficult to match the high degree of landscape heterogeneity, and thus the comprehensive understanding of wind conditions across the whole TP. To overcome this challenge, more spatially-complete datasets, such as reanalysis products, could be used to explore NSWS characteristics of this region. In fact, by using a forecast model in which information from global observations are assimilated, reanalyses produce spatially-homogeneous data that goes several decades back in time, providing an accurate (i.e., spatially complete and physically coherent) description of the climate of the past (Dee et al. 2011). Among the most popular and used reanalysis products, we can find the ones produced over the past decades by the European Centre for Medium-Range Weather Forecasts (ECMWF), as ERA-Interim or ERA5 (see Section 2.2). Historical reanalyses can be used together with regional climate models to downscale information from the reanalyses and simulate more small-scale features of wind. For a complex-topography region like the TP, a high horizontal resolution is needed to better represent the heterogeneity in land surface and orography and to resolve mesoscale processes that affect wind (Li et al. 2017). Therefore, limited-domain climate model simulations with regional refinements are created by dynamically downscaling a coarser global reanalysis (AMS 2013). In particular, by downscaling the reanalysis, large-scale flow is included in the regional simulations, while regional and small-scale circulation features are generated by the regional climate model. In this way, dynamical downscaling products can capture more climatic information at different scales and at a higher resolution, being able to reduce local biases of reanalysis datasets, especially in regions with complex topography (Giorgi and Shields 1999; Gao et al. 2015; Li et al. 2017; Ou et al. 2020). However, before any reanalysis or climate model dataset can be used to study wind speed conditions in a given area, its ability in representing observed NSWS statistics must be proven. In fact, previous studies have shown that their capability in simulating wind speed is strongly dependent on the selected region and the considered time period (Ramon et al. 2019; Wohland et al. 2019; Miao et al. 2020). For this reason, it is necessary to explore the reliability of modeled outputs by comparing them with observations, so that the skills of the models in simulating wind climate can be verified (Kunz et al. 2010).
So far, the literature has largely focused on exploring NSWS changes over the TP (e.g., You et al. 2014; Duan and Xiao 2015), but only a few studies have explored the sub-regional differences in NSWS climatology and how these are simulated by reanalyses or dynamical downscaling products (Li et al. 2017; Li et al. 2021; Yao et al. 2018; Zao et al. 2019). Despite the uneven distribution of the meteorological stations over the TP, available observations can be used to investigate how wind conditions are simulated by reanalyses and climate models. Only when their reliability is verified, those simulated outputs can be employed to comprehensively investigate the NSWS climate, trends and variability over the whole heterogeneous landscape of the TP.
For all these reasons, this study aims to: (a) improve the understanding of the spatio-temporal climatology of homogenized observed NSWS over the TP by identifying the most relevant geographical features for the NSWS climatology; (b) evaluate the performance of recent reanalysis in simulating NSWS observations; and (c) assess if dynamic downscaling of current reanalyses adds value in reproducing the observed NSWS climatology. In this study, we focus on the few recent ECMWF reanalyses (i.e., ERA-Interim, ERA5, and ERA5-Land – see Section 2.2) to evaluate if and how their improvements during the last decades have affected their performance in modeling NSWS over the TP. Similarly, we only look at two different downscaling products of ERA5 to identify the potential added value of downscaling compared to the downscaled reanalysis when it comes to wind simulations. In addition, it is important to highlight that this is the first study which analyses NSWS climatology over the TP using homogenized observed series, i.e. series where their changes are only caused by variations in climate and the effect of possible non-climatic factors are removed and corrected (see Section 2.1).
The paper is structured as follows: the data used by the study (e.g., wind observations and modeled outputs) is presented in Section 2. Section 3 describes the methods applied here to investigate the wind climatology and it is presented how the comparison between wind measurements and modeled data is made. Section 4 shows and discusses the results of the analyses: in particular, after exploring the climatology of observed NSWS (Section 4.1), it is evaluated the performance of reanalysis (Section 4.2) and dynamical downscaling (Section 4.3) products in reproducing observed wind characteristics. Conclusions are drawn in Section 5.
2.1 NSWS observations
Daily means of NSWS observations from 104 stations over the TP for 1960-2020 are provided by the China Meteorological Administration (CMA). Those weather stations do not cover evenly the whole TP, but are mostly located in the East, with only a few stations in the West and center of the plateau (Figure 1). Multi-decadal NSWS series can be unrepresentative of the actual climate and its variations over time when the presence of various types of non-climatic factors, as station relocations and anemometer height and type changes, create artificial shifts (or break-points) in the observed series (Aguilar et al. 2003; Wan et al. 2010). Therefore, it is necessary to apply a homogenization protocol which is able to detect those artificial shifts and afterwards correct the biases which those inhomogeneities create. In this study, Climatol (Guijarro 2018) is applied to identify break-points, remove their biases, reject outliers, and infill missing data. Climatol is a R (R Core Team 2020) package for quality controlling, homogenizing and missing data infilling of climate series. More info about this software can be retrieved at http://www.climatol.eu/ (last accessed 11 January 2023). Climatol has been widely used for homogenizing NSWS series in various studies (e.g., Azorin-Molina et al. 2019; Shi et al. 2019; Zhang et al. 2020; Minola et al. 2021a; Minola et al. 2021b), proving to be a reliable tool for successfully homogenizing wind series. Here, homogenization is performed on monthly aggregates, using only nearest data available at each time step as reference for identifying break-points, rejecting outliers, and infilling of missing data. A homogenized dataset of 104 monthly mean NSWS series over the TP for 1960-2020 is thus created by the applied homogenization protocol.
To analyze the diurnal cycle of NSWS over the TP, sub-daily observations are retrieved from the Met Office Hadley Centre Integrated Surface Database (HadISD) (Smith et al. 2011; Dunn et al. 2012; Dunn et al. 2014; Dunn et al. 2016; Dunn 2019). In particular, this study uses the HadISD version 3.1.1 (https://www.metoffice.gov.uk/hadobs/hadisd/; last accessed 11 January 2023). The HadISD archive contains synoptic reports (sub-daily, station-based, and quality-controlled) from a large number of global measuring stations. Here, 3-hourly (i.e., at UTC 00:00, 03:00, 06:00, 09:00, etc.) measurements of mean NSWS are collected from 27 stations over the TP for 2017-2018 (Fig. 1). Those 27 stations are also included in the CMA 1960-2020 dataset previously presented.
2.2 Reanalyses
This study compares observed NSWS with wind outputs of three different reanalysis datasets from ECMWF: (a) ERA-Interim (hereafter, ERAINT); (b) ERA5; and (c) ERA5-Land. ERA5 is the fifth generation ECMWF reanalysis, which has replaced its predecessor ERAINT (Hersbach et al. 2018). Compared to ERAINT, ERA5 benefits from 10 years of improvements in the model physics and data assimilation, as well as in the increased number of ingested observations, with newly available observed datasets now assimilated. It is noteworthy to remark that in all ECMWF products near-surface wind measurements over land are excluded from the data assimilation process (Dee at el. 2011). ERA5 outputs are produced hourly at a horizontal resolution of 31 km, whereas for ERAINT climate variables are available 3-hourly at a 79 km horizontal grid spacing (Fig. 2). ERA5-Land has been produced by replaying the land component of the ERA5 climate reanalysis through global high-resolution numerical integrations (Muñoz-Sabater et al. 2021). Even though ERA5-Land shares most of the parametrizations with ERA5, its main advantage compared to ERA5 is the horizontal resolution (Fig. 2), which is enhanced globally to 9 km (while temporal resolution is hourly as in ERA5).
3-hourly (for ERAINT) and hourly (for ERA5 and ERA5-Land) 10-m height zonal (u) and meridional (v) wind components are downloaded for 1981-2018, which is the common time period for both observations and the three selected reanalyses. In particular, ERAINT wind data is downloaded from https://apps.ecmwf.int/datasets/data/interim-full-daily/levtype=sfc/ (last accessed 11 January 2023); ERA5 and ERA5-Land wind outputs are accessed from the Copernicus website (https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-single-levels?tab=overview and https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-land?tab=form, respectively; last accessed 11 January 2023). At each grid point, NSWS is calculated as the square root of the sum of the squared wind components at the land-surface single level at a given time:
Once that hourly (and 3-hourly for ERAINT) NSWS series are computed using Equation 1, monthly mean NSWS series are calculated. For evaluating the performance of the reanalyses in simulating observations, the observed NSWS series at a given measuring station is compared with the NSWS series from the closest reanalysis grid-point, under the assumption that the closest grid series matches the observed one better than any other more distant grid-point series (Minola et al. 2020).
2.3 Dynamical downscaling products
This study evaluates the performance in simulating NSWS of two high-resolution downscaling products, which are created by downscaling ERA5 using the regional Weather Research and Forecasting (WRF) model (Skamarock et al. 2008): (1) a WRF downscaling at a 9-km resolution (hereafter, WRF-9km), and (2) the High Asia Refined analysis version 2 (hereafter, HAR). In the WRF-9km (Ou et al. 2020; Sun et al. 2021), ERA5 has been dynamically downscaled at a 9-km resolution for East Asia with a focus on the TP region since 1979. More info about this product and data access at http://biggeo.gvc.gu.se/TPReanalysis/ (last accessed 11 January 2023). In a similar way, HAR (Maussion et al. 2011; Maussion et al. 2014; Wang et al. 2020) is generated by dynamical downscaling ERA5 using WRF. HAR outputs are at a 10 km grid-spacing from 1980. A more detailed description of HAR and the outputs of the model can be retrieved at this link: https://www.klima.tu-berlin.de/index.php?show=daten_har2&lan=de (last accessed 11 January 2023).
The setup of the downscaling differs between the two products. Among the differences, e.g., WRF-9km uses the Yonsei University (YSU) Planetary Boundary Layer (PBL) scheme (Hong et al. 2006), while HAR parametrize the PBL with the Mellor-Yamada-Janijic scheme (Janjić 1994). For the cumulus parameterization, HAR adopts the Grell 3D scheme (Grell 1993; Grell and Dévényi 2002); WRF-9km turned off the convective parameterization. To prevent the model from deviating too far from the forcing data (i.e., to constrain large-scale driving field and thus reducing the impact of model domain size on the regional simulation), HAR uses daily re-inizialization: each run started at 12:00 UTC and contained 36 hours, with the first 12 hours as spin-up time. Instead, WRF-9km uses as forcing strategy spectral nudging, which consists of adding a new term to the tendencies of the model variables that relaxes the selected part of the spectrum to the corresponding waves from driving fields (in this case, ERA5).
In the same way as hourly NSWS has been calculated for ERA5 and ERA5-Land, hourly NSWS series are created using Equation 1 from u and v outputs of WRF-9km and HAR. Monthly mean NSWS series are then created and the closest grid-point series is compared to the observed one of a given series. For comparison between observed, WRF-9km and HAR NSWS series, we use 1991-2018 which is the common time period between those datasets.
2.4 Global digital elevation model
To explore the influence of topography on NSWS statistics, we obtain land surface elevation data from the global digital elevation model by the Earth Resources Observation and Science (EROS) Center: the Global 30 Arc-Second Elevation (GTOPO30) dataset (Gesh et al. 1999). GTOPO30 combines several raster and vector sources of topographic information to derive elevation values on a regularly spaced grid of 30 arc seconds (approximately 1 km). GTOPO30 data is available for downloading at https://www.usgs.gov/centers/eros/science/usgs-eros-archive-digital-elevation-global-30-arc-second-elevation-gtopo30#overview (last accessed 11 January 2023).
3.1 Statistical measures for comparison
To quantify the agreement between observed and modeled NSWS, we use the following statistical tests: 1) Pearson’s correlation coefficient, which measures the degree of association (i.e., linear relationship; Gibbons and Chakrborti 2003); 2) Root Mean Squared Error (hereafter, RMSE), which mathematically expresses the vicinity between two datasets (Von Storch and Zwiers 1999); 3) bias, which identifies the tendency to constant deviate of a realization compared to another; and 4) coefficient of determination (hereafter, R2), which represents the proportion of the variance for a dependent variable that is explained by an independent variable in a regression model.
3.2 Classification of seasonal and diurnal cycle of NSWS
This study uses the k-means clustering method (Wilks 1995) to identify regional regimes in NSWS seasonality and diurnal variability over the Plateau. By applying the k-mean clustering, the mean seasonal (diurnal) cycles of NSWS of the different stations / grid-points are portioned into a given number of clusters based on feature similarity of their seasonal (diurnal) regimes. The k-means clustering is applied to the seasonal and diurnal cycle anomalies of NSWS (i.e., normalized seasonal and diurnal cycles, where the annual and daily mean is removed from each monthly and hourly mean value, respectively). In this way, the classification focuses on the differences in shape rather than differences in overall magnitude. The K-means clustering method, for example, has been used for the analysis of precipitation seasonality over the TP (Curio and Scherer 2016).
3.3 Time-scale for calculation of reanalyzed NSWS
As explained in Section 2.2, the reanalyzed NSWS is calculated from Equation 1, using the hourly (and 3-hourly for ERAINT) u and v wind components outputted by the reanalysis. Monthly mean series of NSWS are then calculated based on those hourly NSWS series (hereafter, NSWS hourly-calculated). This process demands a lot of computational resources, as it requires that data are downloaded at high temporal resolution (i.e., hourly) for multiple years in a large spatial domain, as the one of the TP. It also requires a lot of computation power in order to perform the necessary calculation of monthly averages. A less computational-costly alternative is to download already-calculated monthly means of u and v from the ECMWF archive (where they are available), and then to calculate NSWS using Equation 1 but with monthly-averaged zonal and meridional wind components (hereafter, NSWS monthly-calculated). Mathematically, differences must emerge when using Equation 1 with hourly or monthly-averaged NSWS. But the question is: how different are aggregated statistics between the NSWS hourly-calculated and the NSWS monthly-calculated? In addition, where are these differences most significant? Fig. 3 shows the spatial differences in the 1981-2020 means between the monthly-calculated NSWS and hourly-calculated NSWS. When calculating NSWS at the two different temporal resolutions, NSWS monthly-calculated is always greater than the hourly-calculated NSWS. Differences across the TP can reach 0.4 m s-1 in the higher-elevation regions, with an average difference of 0.2 m s-1 for the whole plateau in all the three reanalyses. Differences may be related to the importance given at specific spatial- and time-scales in the two NSWS calculations. In the NSWS monthly-calculated, daily or sub-daily processes are not so relevant, as it only considers the aggregated statistics of the u and v wind components. This method may be more suitable for global studies, where large-scale atmospheric circulations are more relevant than local and short-time scale processes (Landberg 2016). When computing NSWS based on hourly u and v components, daily and/or sub-daily processes can be considered. Therefore, it is relevant to use hourly-calculated NSWS only for studies on shorter-time or local scales. The TP is a region of extremely complex topography where local and short-time scale processes play a key role (Liu et al. 2009; Yang et al. 2004). Wind conditions are strongly affected by surface forcing as large-scale wind circulation is broken by localized circulations driven by valleys and mountains (Helbig et al. 2017; Rotach et al. 2015; Serafin et al. 2018). For this reason, we decide to use NSWS hourly-calculated in this study.
4.1 Climatology of observed NSWS
Fig. 4 shows the annual and seasonal spatial distribution of mean and standard deviation of observed NSWS for 1960-2020 over the TP. Annually, the few stations on the West and center of the plateau display greater mean NSWS (between ~3.5 and ~4.5 m s-1) compared to the stations in the East and South-east (~2.0 m s-1). The spatial pattern of the standard deviation of observed NSWS for 1960-2020 resembles the spatial pattern of the mean, with higher values (greater than ~1.0 m s-1) for stations in the West and center of the plateau compared to stations in the East (Yao et al. 2018). Following this spatial distribution, it appears that stations with higher mean NSWS display also the higher standard deviation, as shown by the scatter plot of mean versus standard deviation in Fig. S1 in the supplementary material. Seasonally, the spatial distribution with lower values of mean and standard deviation in the West does not change. Overall, there are greater values recorded for all the measuring stations during spring compared to winter, summer and autumn, and stations with greater mean are also the ones with greater standard deviation.
To identify the geographical features behind the detected spatial pattern, the mean NSWS of each station is plotted against the station elevation (Fig. 5a). It is evident that stations at higher elevations are the ones with greater mean NSWS. By grouping stations into classes with increasing averaged elevation, stations display greater mean and standard deviation of NSWS (e.g., red star vs. light-blue star; Fig. 5b). As stations in the West and center of the plateau are the ones located at higher elevation (Fig. 5c), the spatial pattern identified in Fig. 1a and Fig. 1b is explained. Therefore, elevation differences driven by the topography are responsible for the spatial distribution of mean and standard deviation of observed NSWS, with station mean and standard deviation increasing for increasing elevation. Note that mean NSWS decreases from stations in the West to stations in the East when plotting the mean of NSWS of each station against the station longitude (Fig. S1 in the supplementary material). However, this zonal dependency of mean NSWS results from having just a few stations at high elevation in the West of the plateau. In fact, when using the GTOPO30 data to calculate the elevation meridional-mean and to see how elevation changes according to the longitude over the TP (Fig. 5c), it is shown that the plateau has greater altitude in the West compared to the East: the apparent zonal dependency is driven by the real elevation dependency. The elevation dependency of NSWS (i.e., terrain enhances wind speed and high-elevations regions more influences by strong large-scale synoptic flows) is not something new, and various studies (McVicar et al. 2007; Miller and Davenport 1998; Wood 2000) have already shown that wind speed increases exponentially with increasing altitude, especially over the TP (Yao et al. 2018).
To further explore the climatology of NSWS over the TP, the seasonal cycle for 1960-2020 is analyzed. As shown by Fig. S2 in the supplementary material, greater NSWS is recorded during spring compared to the rest of the seasons, in line with what has been shown by Li et al. (2017), Li et al. (2021), and Zhao et al. (2019). Wind is stronger during spring because it is the time of the year when extra-tropical cyclones are active (Li et al. 2021). By using the k-means clustering, differences in the seasonal cycle anomalies among the observed NSWS series are explored. We identify three classes of stations with different seasonality (Fig. 6). To note that we chose three partitions in the cluster analysis because with this number it is reached a good balance between coherent patterns and sufficient distinction between classes. One class (hereafter, Cluster 1, which includes most of the stations – 65 stations; Fig. 6a) displays a NSWS maximum during spring (i.e., March and April). Another class (Cluster 2; Fig. 6b) includes 18 NSWS series that have a maximum in February-March-April and a minimum in July-August-September. The third class (Cluster 3; Fig. 6c) includes 21 stations with greater NSWS during spring and early-summer (i.e., from March to July) compared to the rest of the year. When also looking at the magnitude differences in the NSWS seasonal cycle between the three classes (i.e., when plotting the mean NSWS annual cycle for each cluster identified using the seasonal cycle NSWS anomalies; Fig. 7a), the mean Cluster 2 seasonal cycle is overall greater (~2.0-3.8 m s-1) than the one for Cluster 1 (~1.8-2.5 m s-1). Similar to the highest NSWS values recorded in Cluster 2, Cluster 3 stations reach on average ~1.8-2.5 m s-1 during late spring. Fig. 7b shows the spatial distribution of the seasonal cycle clusters. Stations belonging to Cluster 1 are mostly located in the East of the TP, while stations from Cluster 2 are found in the center of the plateau, with only a few stations in the East (mostly South-east). Cluster 3 stations are located all along the western and southern border of the TP, in the North, and a few are also found in the East. Unfortunately, the sparse and biased distribution of measuring stations over the TP is a challenge for the comprehensive understanding of the spatial pattern of NSWS clusters across the plateau. For example, does Cluster 2 include all stations at the higher elevations of the center of the plateau? For answering to such types of questions, it is necessary to analyze a more spatially complete dataset, as the reanalysis outputs (see Section 4.2).
4.2 Observations vs. reanalyses
To investigate the performance of the selected ECMWF reanalyses in simulating NSWS, the spatial distribution of mean NSWS for 1981-2018 is plotted in Fig. 8 for observations, ERAINT, ERA5 and ERA5-Land. Despite the differences in the horizontal resolution, all the three datasets are able to capture the higher NSWS in the center of the plateau. But, thanks to the higher resolution, better model physics, more data assimilated, and its more advanced assimilation method, ERA5 and ERA5-Land show differences in mean NSWS due to more-local topographic features, which ERAINT cannot reproduce. For example, especially in the South-east of the TP, the latest reanalyses can differentiate between NSWS conditions in valleys (lower values) and along ridges (higher values). In general, ERAINT tends to overestimate mean NSWS over the TP (Fig. 9). Even if both ERA5 and ERA5-Land slightly underestimates the NSWS conditions, the latter dataset better matches the spatial differences in mean NSWS over the plateau compared to ERAINT and ERA5 (R2 for ERA5-Land is 0.48, greater than the 0.42 and 0.45 of ERAINT and ERA5, respectively). Statistics shown in Fig. 10 enable a more detailed comparison between the performance of the three reanalysis datasets in simulating NSWS during 1981-2018. Mean Pearson’s correlation does not differ notably among the reanalyses (between 0.4 and 0.5), even when the seasonal cycle is removed from the monthly NSWS series (mean correlation drops almost equally by ~0.12 for all the three datasets). Mean RMSE is greater for ERAINT and ERA5-Land (0.96 and 0.89 m s-1, respectively) compared to ERA5 (0.8 m s-1). The large RMSEs in ERAINT and ERA5-Land are partly caused by the large biases, which is positive for the former (0.46 m s-1) and negative for the latter dataset (0.55 m s-1). As already seen from Fig. 9, similar to ERA5-Land, ERA5 tends to underestimate NSWS, with a mean bias of 0.3 m s-1. Overall, the performance of the three reanalyses in simulating monthly NSWS series for 1981-2018 does not differ significantly in terms of mean correlation, mean RMSE, and mean bias. Only ERA5 displays a lower mean RMSE and bias compared to both ERAINT and ERA5-Land, even if it is also negatively biased.
The improvements in NSWS simulation of ERA5 and ERA5-Land can be detected when looking at the mean seasonal cycle. Following results of Fig. 9, mean NSWS seasonal cycle of ERA5 and ERA5-Land is negatively biased, but captures the higher NSWS conditions observed during March and April (Fig. S3 in the supplementary material). In addition, when applying the k-mean clustering to modelled NSWS anomaly series, only the seasonal cycles of ERA5 and ERA5-Land grid-series can be clustered in three classes (Fig. 11). Within the ERAINT data, just two classes with different seasonality can be identified. Fig. 12 compares the cluster-averaged (i.e., average over all the stations/closest grid-points in the cluster) seasonal cycles of NSWS calculated using observations with the ones calculated using ERAINT, ERA5, and ERA5-Land. Here we can see that ERAINT, captures the annual variations in the two classes that it simulates, even if it overestimates the annual mean of NSWS. In contrast, both ERA5 and ERAINT are negatively biased when simulating the NSWS seasonal cycle of Cluster 1 and partly lack in capturing the greater wind magnitudes during March-April. But, in Cluster 2 and Cluster 3, the two latest ECMWF reanalyses, even if positively biased (bias smaller than the one shown by ERAINT), properly follow the observed annual variations of NSWS.
The spatially-homogeneous reanalyses can now be used to explore how the different classes are distributed over the TP (Fig. 13), something we could not do using the spatially-heterogeneous observation datasets (see Section 4.1). Cluster 1 (red areas) in all the three datasets surrounds the plateau and it largely occupies the eastern side of the plateau. Instead, Cluster 2 (blue areas) is dominant in the center-west of the TP and over a few limited areas in the East. Cluster 3, which appears in ERA5 and ERA5-Land, occupies a large area in the northeastern plateau, which corresponds to the Qaidam basin, the largest topographic depression inside the TP (Yin et al. 2008). Fig. 14 plots the 1981-2018 mean NSWS of each grid point in a given dataset with its modelled elevation, with the scatter points colored according to the class they belong to. In all the reanalyses, grid-points from Cluster 1 go from the highest to the lowest elevations, while grid-points in Cluster 2 are the ones with the higher elevations and the greatest mean NSWS. Cluster 3, which appears only in ERA5 and ERA5-Land, includes grid-points in a shallow band with elevation between 2800-3500 m a.s.l. (i.e., the depression of the Qaidam basin) with mean NSWS between 2 and 4.5 m s-1 (generally higher than the one of Cluster 1 and in the same range of Cluster 2). The clear differences in the seasonal cycle at the Qaidam basin can be associated with the blocking effect of the surrounding mountainous terrain (Zhao et al. 2019). In fact, as the Quidam Basin is aligned North-west, it shows greater wind conditions during summer when the dominant southwesterly winds prevail strengthened by the blocking effect of the surrounding terrain.
Not only measuring stations located in the Qaidam basin are classified in Cluster 3, but various other stations over the TP are grouped in Cluster 3 (see Fig. 13). Similarly, a few grid points of ERA5 and ERA5-Land outside the Qaidam basin appear to belong to Cluster 3. The reason behind such localized Cluster 3 stations and grid points could be related to the valley and mountain orientations: for example, the main wind direction coincides with the path of the westerly jet stream and the orientation of valleys and mountains (Yao et al. 2018). In ERA5 and ERA5-Land, where complex topography can only be modeled partly, Cluster 3 grid points outside the Qaidam basin are uncommon.
The reason why ERAINT only identifies two classes when clustering the seasonality, while the latest datasets can see three classes (i.e., the Qaidam basin’s wind conditions), may be related to the lower resolution, the worse modeled physics, less data assimilated, and the less advanced assimilation method compared to ERA5 and ERA5-Land (Dee et al. 2011). For example, thanks to the higher resolution, the orography is better simulated in ERA5 and ERA5-Land: the mean elevation difference between actual station elevations and modeled elevations of the closest grid-point of ERAINT is -645.3 m, and it decreases for ERA5 and ERA5-Land (-599.7 m and -450.2 m, respectively). The more realistic topographic representation may help to better identify the different NSWS conditions in the comparatively low-elevated area of the Qaidam basin.
To summarize, ERA5 better simulates NSWS over the TP compared to ERAINT probably due to its considerable increase in horizontal and vertical resolution and a decade of improvements in representation of model processes and data assimilation (Hersbach et al. 2018). The simulated NSWS in ERA5-Land do not match better the observed one compared to ERA5, as ERA5-Land shares with ERA5 most of the parametrizations and does not benefit of any changes in the physics of the model or in the data assimilation (Muñoz-Sabater et al. 2021). To notice that the elevation of the measuring stations is generally lower than the one of the nearest grid cells, especially in the mountainous southern TP (Li et al. 2017). This is because most of the stations across the Plateau are located in valleys, where it is easier to access. Therefore, the station elevation may not be a representation of the closest grid point, and such elevation mismatch could contribute to the differences between observed and simulated wind.
4.3 The added value of downscaling
To evaluate if the downscaling adds value to the NSWS simulations compared to the downscaled product, the spatial distribution of mean NSWS for 1991-2018 is plotted in Fig. 15 for observations, ERA5, and the two downscaled products WRF-9km and HAR. As already seen from Fig. 8, ERA5 shows higher NSWS values in the center of the plateau (as also seen in the observations) and simulates NSWS conditions following the complex terrain features (e.g., the large valleys in the South-east of the TP). Instead, WRF-9km only shows strong mean winds in the southern part of the plateau, with weaker winds from the center to the North. Mean NSWS in HAR is greater in the West, and it decreases by moving to the East. HAR, similar to WRF-9km, does not display the feature with higher wind conditions in the center-west of the TP, and a general lower NSWS in all its surroundings. Fig. S4 in the supplementary material confirms those discrepancies between observed NSWS and NSWS from the two downscaled products. When plotting the mean observed NSWS against the simulated one, WRF-9km poorly simulates wind conditions: the scatterplot has R2 = 0.17 and all the scatter-points are greatly positively biased (i.e., NSWS higher in WRF-9km compared to observations). For HAR, R2 is still lower than for ERA5 (0.42 and 0.46, respectively) and modeled NSWS means are in general higher than the observed ones. When it comes to the modeling of the seasonal cycle (Fig. S5 in the supplementary material), in a similar way both WRF-9km and HAR overestimate the observed one, not showing any improvements compared to the downscaled ERA5. For these reasons, when it comes to aggregated climatology, the two downscaled products WRF-9km and HAR do not better simulate NSWS over the TP compared to ERA5. But the added value of downscaled products should be searched in the better modeling of shorter spatial- and time- scale processes, as the diurnal cycle (Ou et al. 2020).
Therefore, we explore here the mean diurnal cycle of observed NSWS and then we compare it with the one simulated by ERA5 and the two downscaling products (Fig. 16). Fig. 16a shows that the mean 3-hourly sub-daily variations of NSWS calculated using the 27 stations of the HadISD dataset peaks at 9 UTC (~15-16 local time), in line with what shown by Zhao et al. (2019). The annual averaged diurnal cycle from ERA5 and WRF-9km also peaks around 9 UTC. Instead, the diurnal cycle of HAR peaks earlier (around 6 UTC). WRF-9km displays the lowest correlation value compared to ERA5 and HAR. In addition, WRF-9km largely overestimates the magnitude of the diurnal variations (mean positive bias and RMSE of ~2.5 m s-1). Even if for HAR there is a large mean RMSE (~1.5 m s-1), it is mostly caused by the high mean positive bias (~1.3 m s-1). ERA5 displays a much smaller mean RMSE compared to WRF-9km and HAR, and such mean error is largely caused by the average negative bias of ~0.5 m s-1.
To further explore the observed 3-hourly diurnal cycle of NSWS, we apply the k-mean clustering to the dataset of 27 observed series. By conducting the cluster analysis with an increasing number of classes (Fig. 17 - top row), the anomalies of the mean diurnal cycle do not drastically change in shape (always peak at 9UTC), but the peak becomes more pronounced. The more pronounced diurnal cycle seems to be partly related to station elevation: the classes which peaks more, are the ones with greater mean elevation. When looking at the clusters identified in the simulated diurnal cycles (Fig. 17 - bottom row), all the classes in ERA5 peaks at 9 UTC and differ only in how this peak is pronounced. This is partly shown also in HAR, but, especially in WRF-9km, the identified classes do not only differ in how the 9 UTC peak is pronounced, but also in the time when it is recorded and in the overall shape of the mean diurnal variations.
Zhao et al. (2019) previously investigated diurnal variations in NSWS over the TP, finding remarkable regional characteristics with respect to the diurnal variations. For example, in the eastern of the plateau, the minimum and the maximum in the recorded wind speed occur about 1 hour later than in the west; or the diurnal cycle at the stations in the Qaidam Basin differ significantly from the one observed at the other stations elsewhere on the plateau. Unfortunately, for this study hourly data were not available, and the use of 3-hourly outputs for exploring the diurnal cycle characteristics may mask possible regional differences.
Therefore, for what is shown here, HAR and WRF-9km do not improve the simulation of the mean diurnal cycle over the TP, even if additional analysis (using hourly statistics) may be needed. To conclude, both HAR and WRF-9km do not show a significant added value in the simulation of NSWS statistics over the TP compared to the downscaled product ERA5. The dynamical downscalings may improve their performance in simulating NSWS over the TP by enhancing the momentum transports in the planetary boundary parameterization scheme and better considering the unresolved topographic features (Li et al. 2015). As shown by Gómez-Navarro et al. (2015), WRF models perform better in simulating NSWS over complex terrain when an appropriate planetary parameterization scheme (i.e., one that explicitly considers the effects of sub-grid topography) is chosen. In addition, considering the complicated geographical environment, the biases could arise from the planetary boundary parameterization that does not pass the decreased momentum in the upper levels of the atmosphere to the near-surface layers over the TP. Moreover, improvements could be reached when dynamic downscaling will use dynamic land surface characteristics rather than static ones.
This study explored the climatology of NSWS over the TP by using both observations and high-resolution climate model simulations. Observed mean and standard deviation is greater at the stations located in the center of the plateau, proving that wind speed increases with increasing altitude over the TP as altitude enhances wind speed. In addition, according to their wind characteristics (e.g., mean values, seasonal cycle, etc.), three regions of different wind regimes can be identified: (1) the center areas of the plateau, with a higher average elevation than the rest of the TP; (2) the East and the edges of the TP; and (3) the topographic depression of the Qaidam basin, where the blocking effect of the surrounding mountainous terrain plays a key role.
ERA5 is the dataset which matches better the measured NSWS. Compared to its predecessor ERAINT, which can only represent two regions of different wind conditions, ERA5 grid cells can be grouped into three regions, similar to the observations. This is most likely due to its higher spatial resolution, better model physics, more data assimilated, and the more advanced assimilation method. However, the newest ERA5-Land does not show improvements compared to ERA5, as the ERA5-Land data is a replay of the land component of ERA5 with a finer spatial resolution (without any improvement of the parametrization scheme; i.e., no better model physics).
To check the added values of the dynamical downscaling, observations are compared to two different downscaling products of ERA5, i.e. WRF-9km and HAR. Both dynamical downscaling products do not capture the observed spatial distribution characteristics of NSWS. In a similar way, they also do not improve the simulation of the 3-hourly diurnal cycle compared to ERA5. The biases of NSWS in the dynamical downscalings may result from planetary boundary parameterization or the characteristics of land surface. Therefore, the ability of dynamical downscaling to simulate NSWS may be improved by enhancing the momentum transports in the planetary boundary parametrization scheme and better considering the unresolved topographic features.
Funding
This study contributes to the strategic research areas of Modelling the Regional and Global Earth system (MERGE). Lorenzo Minola is funded by the International Postdoc grant from the Swedish Research Council (2021-00444). Kaiqiang Deng is supported by the National Natural Science Foundation of China (42275020).
Competing interests
The authors have no relevant financial or non-financial interests to disclose.
Author contributions
Conceptualization: Lorenzo Minola, Deliang Chen; Data curation: Lorenzo Minola, Gangfeng Zhang, Tinghai Ou, Jose A. Guijarro; Methodology: Lorenzo Minola, Julia Kukulies, Julia Curio, Cesar Azorin-Molina; Formal analysis and investigation: Lorenzo Minola, Julia Kukulies, Julia Curio; Writing – original draft preparation: Lorenzo Minola; Writing – review and editing: Lorenzo Minola, Gangfeng Zhang, Tinghai Ou, Julia Kukulies, Julia Curio, Jose A. Guijarro, Kaiqiang Deng, Cesar Azorin-Molina, Cheng Shen, Alessandro Pezzoli, Deliang Chen; Funding acquisition: Deliang Chen.
Data availability
The dataset of homogenized monthly mean near-surface wind speed series generated and analysed during this study is available from this repository: https://biggeo.gvc.gu.se/Pubdata/NSWS_TP.nc. Sub-daily (3-hourly) wind observations are retrieved from the Met Office HadISD dataset available at https://www.metoffice.gov.uk/hadobs/hadisd/. The GTOPO30 dataset is available for downloading at https://www.usgs.gov/centers/eros/science/usgs-eros-archive-digital-elevation-global-30-arc-second-elevation-gtopo30#overview. ERAINT wind data is accessed from https://apps.ecmwf.int/datasets/data/interim-full-daily/levtype=sfc/. ERA5 and ERA5-Land wind outputs are downloaded from the Copernicus website: https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-single-levels?tab=overview and https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-land?tab=form, respectively. The WRF-9km dataset is available from: http://biggeo.gvc.gu.se/TPReanalysis/. HAR outputs are retrieved from https://www.klima.tu-berlin.de/index.php?show=daten_har2&lan=de.
- Aguilar E, Auer I, Brunet M, Peterson TC, Wieringa J (2003) Guidelines on climate metadata and homogenization. WMO Technical Document No. 1186. https://library.wmo.int/doc_num.php?explnum_id=10751 Accessed 11 January 2023
- American Meteorological Society (AMS) (2013) Regional climate model. Glossary of Meteorology. https://glossary.ametsoc.org/wiki/Regional_climate_model Accessed 11 January 2023
- Azorin-Molina C, Guijarro, JA, McVicar TR, Frost BCT, Chen D (2019) An approach to homogenize daily peak wind gusts: An application to the Australian series. Int J Climatol 39:2260-2277. https://doi.org/10.1002/joc.5949
- Curio J, Scherer D (2016) Seasonality and spatial variability of dynamic precipitation controls on the Tibetan Plateau. Earth Syst Dyn 7:767-782. https://doi.org/10.5194/esd-7-767-2016
- Dee DP, Uppala SM, Simmons AJ, Berrisford P, Poli P, Kobayashi S et al (2011) The ERA-Interim reanalysis: configuration and performance of the data assimilation system. Q J R Meteorol Soc 137:553-597. https://doi.org/10.1002/qj.828
- Dong Z, Hu G, Qian G, Lu J, Zhang Z, Luo W, Lyu P (2017) High-altitude aeolian research on the Tibetan Plateau. Rev Geophys 55:864-901. https://doi.org/10.1002/2017RG000585
- Duan A, Xiao Z (2015) Does the climate warming hiatus exist over the Tibetan Plateau? Sci Rep 5:13711. Https://doi.org/10.1038/srep13711
- Dunn RJH, Willet KM, Thorne PW, Woolley EV, Durre I, Dai A et al (2012) HadISD: a quality-controlled global synoptic report database for selected variables at long-term stations from 1973-2011. Clim Past 8:1649-1679. https://doi.org/10.5194/cp-8-1649-2012
- Dunn RJH, Willet KM, Morice CP, Parker DE (2014) Pairwise homogeneity assessment of HadISD. Clim Past 10:1501-1522. https://doi.org/10.5194/cp-10-1501-2014
- Dunn RJH, Willet KM, Parker DE, Mitchell L (2016) Expanding HadISD: quality-controlled, sub-daily station data from 1931. Geosci Instrum Methods and Data Syst 5:473-491. https://doi.org/10.5194/gi-5-473-2016
- Dunn RJH (2019) HadISD version 3: monthly updates. Met Office Hadley Centre Technical Notes. https://digital.nmla.metoffice.gov.uk/digitalFile_13890750-fb6f-42c7-92df-1c4504621fae/ Accessed 11 January 2023
- Gao Y-X, Tang M-C, Luo S-W, Shen Z-B, Li C (1981) Some aspects of recent research on the Qinghai-Xizang Plateau meteorology. Bull Am Meteorol Soc 62:31-35. https://doi.org/10.1175/1520-0477(1981)062<0031:SAORRO>2.0.CO;2
- Gao Y, Xu J, Chen D (2015) Evaluation of WRF mesoscale climate simulations over the Tibetan Plateau during 1979-2011. J Clim 28:2823-2841. https://doi.org/10.1175/JCLI-D-14-00300.1
- Gesch DB, Verdin KL, Greenlee SK (1999) New land surface digital elevation model covers the Earth. Eos 80:69-70. https://doi.org710.1029/99EO00050
- Giorgi F, Shields C (1999) Tests of precipitation parametrizations availavle in the latest version of NCAR regional climate model (RegCM) over continental United States. J Geophys Res Atmos 104: 6353-6375. https://doi.org/10.1029/98JD01164
- Gómez-Navarro JJ, Raible CC, Dierer S (2015) Sensitivity of the WRF model to PBL parametrisations and nesting techniques: evaluation of wind storms over complex terrain. Geosci Model Dev 8:3349-3363. https://doi.org/10.5194/gmd-8-3349-2015
- Grell GA (1993) Prognostic evaluation of assumptions used by cumulus parameterization. Mon Weather Rev 121:764-787. https://doi.org/10.1175/1520-0493(1993)121<0764:PEOAUB>2.0.CO;2
- Grell GA, Dévényi D (2002) A generalized approach to parameterizing convection combining ensemble and data assimilation techniques. Geophys Res Lett 29:38-1-38-4. https://doi.org/10.1029/2002GL015311
- Guijarro JA (2018) Homogenization of climatic series with Climatol. www.climatol.eu/homog_climatol-en.pdf Accessed 11 January 2023
- Helbig N, Mott R, van Herwijnen A, Winstral A, Jonas T (2017) Parameterizing surface wind speed over complex topography. J Geophys Res Atmos 122:651-667. https://doi.org/10.1002/2016JD025593
- Hersbach H, de Rosnay P, Bell B, Schepers D, Simmons A, Soci C et al (2018) Operational global reanalysis: progress, future directions and synergies with NWP. ERA Report Series. https://www.ecmwf.int/node/18765 Accessed 11 January 2023
- Hong S-Y, Noh Y, Dudhia J (2006) A new vertical diffusion package with an explicit treatment of entrainment processes. Mon Weather Rev 134:2318-2341. https://doi.org/10.1175/MWRR3199.1
- IPCC (2019) IPCC special report on the ocean and cryosphere in a changing climate. Cambridge Press, Cambridge
- Janjić ZI (1994) The step-mountain eta coordinate model: Further developments of the convection, viscous sublayer, and turbulence closure schemes. Mon Weather Rev 122:927-945. https://doi.org/10.1175/1520-0493(1994)122<0927:TSMECM>2.0.CO;2
- Kang S, Xu Y, You Q, Flügel W-A, Pepin N, Yao T (2010) Review of climate and cryospheric change in the Tibetan Plateau. Environ Res Lett 5:015101. https://doi.org/10.1088/1748-9326/5/1/015101
- Kunz M, Mohr S, Rauthe M, Lux R, Kottmeier C. (2010) Assessment of extreme wind speeds from Regional Climate Models – Part 1: Estimation of return values and their evaluation. Nat Hazards Earth Syst Sci 10:907-922. https://doi.org/10.5194/nhess-10--907-2010
- Landberg L (2016) Meteorology for wind energy – An introduction. John Wiley & Sons Ltd, Chichester
- Li X, Gao Y, Pan Y, Xu Y (2017) Evaluation of near-surface wind speed simulations over the Tibetan Plateau from three dynamical downscalings based on WRF model. Theor Appl Climatol 134:1399-1411. https://doi.org/10.1007/s00704-017-2353-9
- Li P, Furtado K, Zhou T, Chen H, Li J, Guo Z, Xiao C (2018) The diurnal cycle of East Asian summer monsoon precipitation simulated by the Met Office Unified Model at convection-permitting scales. Clim Dyn 55:131-151. https://doi.org/10.1007/s00382-018-4368-z
- Li Z, Xiao Z-N, Zheng C-W (2021) Observation analysis of wind climate in China for 1971-2017 under the demand of wind energy evaluation and utilization. Energy Rep 7:3535-3546. https://doi.org/10.1016/j.egyr.2021.06.012
- Liu X, Bai A, Liu C (2009) Diurnal variations of summertime precipitation over the Tibetan Plateau in relation to orographically-induced regional circulations. Environ Res Lett 4:045203. https://doi.org/10.1088/1748-9326/4/4/045203
- Ma Y, Hu Z, Xie Z, Ma W, Wang B, Chen X et al (2020) A long-term (2005-2016) dataset of hourly integrated land-atmosphere interaction observations on the Tibetan Plateau. Earth Syst Sci Data 12:2937-2957. https://doi.org/10.5194/essd-12-2937-2020
- Maussion F, Scherer D, Finkelnburg R, Richters J, Yang W, Yao T (2011) WRF simulation of a precipitation event over the Tibetan Plateau, China – an assessment using remote sensing and ground observations. Hydrol Earth Syst Sci 15:1795-1817. https://doi.org/10.5194/hess-15-1795-2011
- Maussion F, Scherer D, Mölg T, Collier E, Curio J, Finkelnburg R (2014) Precipitation seasonality and variability over the Tibetan Plateau as resolved by the High Asia Reanalysis. J Climate 27:1910-1927. https://doi.org/10.1175/JCLI-D-13-00282.1
- McVicar TR, Van Niel TG, Li LT, Hutchinson MF, Mu XM, Liu ZH (2007) Spatially distributing monthly reference evapotranspiration and pan evaporation considering topographic influences. J Hydrol 338:196-220. https://doi.org/10.1016/j.hydrol.2007.02.018
- Maio H, Dong D, Huang G, Hu K, Tian Q, Gong Y (2020) Evaluation of Northern Hemisphere surface wind speed and wind power density in multiple reanalysis datasets. Energy 200:117382. https://doi.org/10.1016/j.energy.2020.117382
- Miller CA, Davenport AG (1998) Guidelines for the calculation of wind speed-ups in complex terrain. J Wind Eng Ind Aerodyn 74-76:189-197 https://doi.org/10.1016/S0167-6105(98)00016-6
- Minola L, Zhang F, Azorin-Molina C, Safaei Pirooz AA, Flay RGJ, Hersbach H et al (2020) Near-surface mean and gust wind speeds in ERA5 across Sweden: Towards an improved gust parametrization. Clim Dyn 55:887-907. https://doi.org/10.10077s00382-020-05302-6
- Minola L, Azorin-Molina C, Guijarro JA, Zhang G, Son S-W, Chen D (2021a) Climatology of near-surface daily peak wind gusts across Scandinavia: Observations and model simulations. J Geophys Res Atmos 126:e2020JD033534. https://doi.org/10.1029/2020JD033534
- Minola L, Reese H, Lai H-W, Azorin-Molina C, Guijarro JA, Son S-W et al (2021b) Wind stilling-reversal across Sweden: The impact of land-use and large-scale atmospheric circulation changes. Int J Climatol 42:1049-1071. https://doi.org/10.1002/joc.7289
- Muñoz-Sabater J, Dutra E, Agustí-Panareda A, Albergel C, Arduini G, Balsamo G et al (2021) ERA5-Land: a state-of-the-art global reanalysis dataset for land applications. Earth Syst Sci Data 13:4349-4383. https://doi.org/10.5194/essd-13-4349-2021
- Ou T, Chen D, Chen X, Lin C, Yang C, Lai H-W et al. (2020) Simulation of summer precipitation diurnal cycles over the Tibetan Plateau at the gray-zone grid spacing for cumulus parametrization. Clym Dyn 54:3525-3539. https://doi.org/10.1007/s00382-020-05181-x
- R Core Team (2020) R: A language and environment for statistical computing. R Fundation for Statistical Computing. https://www.R-project.org/ Accessed 11 January 2023
- Ramon J, Lledó L, Torralba V, Soret A, Doblas-Reyes FJ (2019) What global reanalysis best represents near-surface winds? Q J R Meteorol Soc145:3236-3251. https://doi.org/10.1002/qj.3616
- Rotach MW, Gohm A, Lang MN, Leukauf D, Stiperski I, Wagner JS (2015) On the vertical exchange of heat, mass, and momentum over complex, mountainous terrain. Front Earth Sci 3:76. https://doi.org/10.3389/feart.2015.00076
- Serafin S, Adler B, Cuxart J, De Wekker SFJ, Gohm A, Grisogono B et al (2018) Exchange processes in the atmospheric boundary layer over mountainous terrain. Atmos 9:102. https://doi.org/10.3390/atmos9030102
- Shi P, Zhang G, Kong F, Chen D, Azorin-Molina C, Guijarro JA (2019) Variability of winter haze over the Beijing-Tianjin-Hebei region tied to wind speed in the lower troposphere and particulate sources. Atmos Res 215:1-11. https://doi.org/10.1016/j.atmosres.2018.08.013
- Skamarock WC, Klemp JC, Dudhia J, Gill DO, Barker DM, Duda MG et al (2008) A description of the advanced research WRF version 3. NCAR Technical note (NCAR/TN-475+STR). https://doi.org/10.5065/D68S4MVH Accessed 11 January 2023
- Smith A, Lott N, Vose R (2011) The Integrated Surface Database: Recent developments and partnerships. Bull Am Meteorol Soc 92:704-708. https://doi.org/10.1175/2011BAMS3015.1
- Sun H, Su F, He Z, Ou T, Chen D, Li Z et al (2021) Hydrological evaluation of high-resolution precipitation estimates from The WRF model in the Third Pole river basins. J Hydrometeorol 22:2055-2071. https://doi.org/10.1175/JHM-D-20-0271.1
- Wan H, Wang XL, Swail VR (2010) Homogenization and trend analysis of Canadian near-surface wind speeds. J Clim 23:1209-1225. https://doi.org/10.1175/2009JCLI3200.1
- Wang X, Tolksdorf V, Otto M, Scherer D (2020) WRF-based dynamical downscaling of ERA5 reanalysis data for High Mountain Asia: Towards a new version of the High Asia Refined analysis. In J Climatol 41:743-762. https://doi.org/10.1002/joc.6686
- Wilks DS (1995) Statistical methods in the atmospheric sciences. Academic Press, San Diego
- Wohland J, Omrani N-E, Witthaut D, Keenlyside NS (2019) Inconsistent wind speed trends in current Twentieth Century reanalyses. J Geophys Res Atmos 124:1931-1940. https://doi.org/10.1029/2018JD030083
- Wood N (2000) Wind flow over complex terrain: a historical perspective and the prospect for large-eddy modelling. Bound-Layer Meteorol 96:11-32. https://doi.org/10.1023/A:1002017732694
- Yang K, Koike T, Fujii H, Tamura T, Xu X, Bian L et al (2004) The daytime evolution of the atmospheric boundary layer and convection over the Tibetan Plateau: Observations and simulation. J Meteorol Soc Jpn 82:1777-1792. https://doi.org/10.2151/jmsj.82.1777
- Yang K, Wu H, Qin J, Lin C, Tang W, Chen Y (2014) Recent climate changes over the Tibetan Plateau and their impacts on energy and water cycle: A review. Glob Planet Change 112:79-91. https://doi.org/10.1016/j.gloplacha.2013.12.001
- Yao T, Thompson LG, Mosbrugger V, Zhang F, Ma Y, Luo T et al (2012) Third Pole Environment. Environ Dev 3:52-64. https://doi.org/10.1016/j.envdev.2012.04.002
- Yao Z, Li X, Xiao J (2018) Characteristics of daily extreme wind gusts on the Qinghai-Tibet Plateau, China. J Arid Land 10:673-685. https://doi.org/10.1007/s40333-018-0094-y
- Yin A, Dang Y-Q, Zhang M, Chen X-H, McRivette MW (2008) Cenozoic tectonic evolution of the Qaidam basin and its surrounding regions (Part 3): Structural geology, sedimentation, and regional tectonic reconstruction. Geol Soc Am Bull 120:847-876. https://doi.org/10.1130/B26232.1
- You Q, Fraedrich K, Min J, Kang S, Zhu X, Pepin N et al (2014) Observed surface wind speed in the Tibetan Plateau since 1980 and its physical causes. In J Climatol 34:1873-1882. https://doi.org/10.1002/joc.3807
- Zhang G, Azorin-Molina C, Chen D, Guijarro JA, Kong F, Minola L et al (2020). Variability of daily maximum wind speed across China, 1975-2016. J Clim 33: 2793-2816. https://doi.org/10.1175/JCLI-D-19-0603.1
- Zhao Y, Li J, Zhang Q, Jiang X, Feng A (2019) Diurnal variations in surface wind over the Tibetan Plateau. Atmos 10:112. https://doi.org/10.3390/atmos10030112
- Zheng H, Liu X, Liu C, Dai X, Zhu R (2009) Assessing contributions to panevaporation trends in Haihe River Basin, China. J Geophys Res Atmos 114:D24105. https://doi.org/10.1029/2009JD012203
No competing interests reported.