High-Resolution Dynamical Downscaling Reveals Multi-Scale Evolution of the Surface Wind Field over Hainan Island (1961–2022)

Abstract

Wind fields on tropical islands are among the most complex systems in atmospheric science, simultaneously influenced by large-scale monsoons, tropical cyclones, local sea-land circulation, and island topography. These interactions result in extremely complex responses to climate change, posing significant challenges for detailed assessment. This study examines how multi-scale processes have shaped the long-term evolution of the near-surface wind speed over Hainan, China’s largest tropical island. We developed a new high-resolution (5 km, hourly) regional climate reanalysis spanning 1961–2022, based on the WRF model and ERA5 data. Our analysis reveals three key findings: First, the long-term trend of wind speed over Hainan exhibits significant spatial heterogeneity, characterized by “coastal stilling and inland strengthening.” Wind speeds in coastal areas have decreased by −0.03 to −0.09 m/s per decade, while those in the mountainous interior have paradoxically increased by up to +0.06 m/s per decade. This pattern arises from the interaction between the weakening East Asian Winter Monsoon and the island’s complex terrain. Second, the frequency of extreme wind events has undergone seasonal reorganization: days with strong winds linked to the winter monsoon have significantly decreased (−0.214 days per decade), whereas days linked to warm-season tropical cyclones have increased (+0.097 days per decade), indicating asynchronous evolution of climate extremes. Third, the risk from 100-year extreme wind events is undergoing geographical redistribution, shifting from the coast to the mountainous interior (with an increase of 0.4–0.7 m/s in inland areas), posing a direct challenge to existing engineering design standards. Taken together, these findings demonstrate that local topography can significantly influence large-scale climate change signals, underscoring the critical role of high-resolution modeling in understanding the climate response of such complex systems.

1. Introduction

Hainan Island, China’s largest tropical island and only provincial-level Free Trade Port, occupies a unique geographical position with increasing strategic economic importance. Situated at the confluence of the East Asian monsoon and Western Pacific tropical circulation, the island experiences frequent impacts from tropical cyclones and other severe weather systems. Its complex inland topography, dominated by high central mountain ranges, creates a distinct tropical island climate [1,2]. With the comprehensive advancement of the Hainan Free Trade Port, large-scale infrastructure projects (e.g., seaports, cross-sea bridges, offshore wind farms) and climate-dependent industries such as shipping, tourism, and tropical agriculture have created unprecedented demand for high-resolution meteorological information, particularly regarding the spatiotemporal evolution of the near-surface wind field [3,4]. Surface wind speed is not only the core variable determining wind energy potential but also a critical environmental factor influencing maritime safety, structural engineering design, pollutant dispersion, and agricultural productivity [5,6]. Therefore, scientific understanding of Hainan’s wind field evolution represents both a frontier issue in regional climatology and a matter of profound practical significance for ensuring sustainable regional development.

Over the past several decades, widespread decreases in near-surface terrestrial wind speeds—a phenomenon termed “global terrestrial stilling”—have been observed globally [7,8]. Studies focused on China have confirmed the prevalence of this trend, especially in northern regions and along eastern coasts [8]. The causes of this stilling are primarily attributed to two factors: systematic changes in large-scale circulation, such as significant weakening of the East Asian Winter Monsoon (EAWM) [9], and modifications of the underlying surface, such as increased surface roughness due to rapid urbanization and large-scale afforestation [10,11]. However, since approximately 2010, some studies have indicated pauses or even “stilling reversals” in global or regional trends [12], adding complexity to long-term wind speed evolution. To date, the vast majority of research on wind speed changes in China has concentrated on the continental interior; systematic understanding of how regions like Hainan, with profound marine influence and complex terrain, respond to these large-scale changes remains a significant knowledge gap.

In contrast to the general stilling over land, wind speeds over global oceans have exhibited widespread strengthening trends during the satellite era [13,14]. The South China Sea, where Hainan is located, shows particularly complex spatiotemporal wind speed variations closely linked to air-sea interaction processes such as the El Niño–Southern Oscillation (ENSO) [15]. As the dominant system driving Hainan’s annual climate cycle, the long-term evolution of the East Asian monsoon, especially weakening of the winter component, is undoubtedly a key large-scale forcing factor for the region’s wind field [16]. Concurrently, as one of the most active typhoon regions in the Northwest Pacific, Hainan’s extreme wind climate is largely dictated by typhoon activity [17]. In the context of global warming, while trends in typhoon frequency remain debated, there is broad consensus that typhoon intensity and associated extreme precipitation are increasing [18,19,20]. Consequently, Hainan’s wind field evolution represents the net result of multiple interacting factors—terrestrial stilling signals, oceanic wind trends, monsoon system evolution, and changing typhoon activity—all modulated by the island’s complex topography. Disentangling how these factors interact to shape final wind field patterns constitutes a key scientific challenge.

Addressing these issues requires high-quality, high-resolution meteorological data, yet existing products have significant limitations. In situ meteorological stations, while providing “ground truth” observations, are too sparsely and unevenly distributed in mountainous regions like Hainan to capture the fine-scale spatial structure of wind fields, and their historical records often suffer from inhomogeneities due to instrument changes and station relocations [21]. Global reanalyses like ERA5, though temporally homogeneous, have coarse spatial resolutions (~25–30 km) that cannot resolve steep wind gradients and local circulations induced by Hainan’s inland topography [22]. Meanwhile, high-resolution gridded products for China, such as CLDAS or CMFD, while performing well over the mainland, have limitations in representing marine characteristics and physical processes over complex terrain, as their wind fields are often based on sparse station interpolation rather than physical modeling [23,24]. Therefore, employing dynamical downscaling to construct high-resolution, purpose-built climate datasets is the necessary pathway to overcome these data limitations and conduct in-depth investigations of Hainan’s wind field evolution.

This study, therefore, aims to address the following core scientific questions through high-resolution regional climate modeling. First, utilizing a new long-term, high-resolution dataset, we systematically characterize multi-scale (annual, seasonal, diurnal) spatiotemporal evolution of Hainan’s near-surface wind speed field, with particular focus on revealing heterogeneous spatial patterns. Second, we investigate key physical mechanisms driving these changes, especially interactions between large-scale circulation and local topography. Finally, we evaluate the resulting impacts of these changes on the frequency of extreme wind events (both strong and weak) and regional wind energy resources. This study focuses specifically on wind speed characteristics, while wind direction analysis will be addressed in future research.

2. Materials and Methods

2.1. Study Area and Climatic Zoning

This study focuses on Hainan Island (approximately 18.1° N–20.2° N, 108.6° E–111.1° E), a large tropical island at China’s southernmost extent. The island exhibits a distinct tropical monsoon climate regime. To establish an objective, quantitative framework for analyzing spatial climate heterogeneity, we performed climate regionalization using the k-means clustering algorithm.

The classification employed feature vectors designed to capture primary geographic, topographic, and climatic controls on the island’s environment, including: longitude, latitude, topographic complexity, elevation, slope, aspect, distance from coastline, and monthly mean wind speeds for all 12 months averaged over the 1991–2020 climatological period from our Hainan Reanalysis (HNR) dataset. The optimal cluster number was determined as three using the Elbow Method, resulting in the partitioning of Hainan Island into three distinct subregions: Coastal Plains, Hills, and Mountains. This zoning framework provides the fundamental geographic reference for spatial analyses throughout this study (depicted by contours in Figure 1).

2.2. Data and Methodology

2.2.1. The Hainan Reanalysis (HNR) Dataset

To support a detailed investigation of the long-term, fine-scale characteristics of the surface wind field over Hainan Island, we developed a new regional climate reanalysis dataset, herein referred to as the Hainan Reanalysis (HNR). This dataset provides hourly meteorological fields at 5-km horizontal resolution, spanning 1940–2022. The HNR was generated through dynamical downscaling using the Weather Research and Forecasting (WRF) model version 4.3, driven by the fifth-generation European Centre for Medium-Range Weather Forecasts (ECMWF) global atmospheric reanalysis (ERA5).

As summarized in

Table 1. Comparison of Major Global Atmospheric Reanalysis Datasets.

Dataset	Provider	Temporal Coverage	Spatial Resolution (Atmosphere)	Key Advantages and Suitability for Regional Wind Studies in China
ERA5	ECMWF	1940–Present	~31 km (0.28125°)	Highest resolution, hourly data, advanced assimilation. Widely validated and shown to perform well for wind studies in China.
ERA-Interim	ECMWF	1979–2019	~79 km (0.75°)	Predecessor to ERA5. It was a benchmark but is now superseded; lower resolution and less advanced model.
MERRA-2	NASA	1980–Present	~50 km (0.5° × 0.625°)	High resolution and includes aerosol-weather interactions. A strong alternative, but ERA5 generally has higher spatial/temporal resolution.
JRA-55	JMA	1958–Present	~55 km (0.5625°)	Long-term, consistent, and high-quality dataset. Good performance, but slightly coarser resolution than ERA5.
NCEP/NCAR R1	NCEP/NCAR	1948–Present	~210 km (2.5°)	Very long time series, but the very coarse resolution is unsuitable for detailed regional or local wind analysis.
NCEP-DOE R2	NCEP/DOE	1979–Present	~210 km (2.5°)	An update to R1 with fixes for known errors, but retains the same coarse resolution, limiting its use for regional studies.
CFSR	NCEP	1979–2010	~38 km (0.5°)	High resolution for its time, but the time series is discontinued, making it unsuitable for long-term climatological studies extending to the present.

, datasets like NCEP R1/R2 are not suitable for our study due to their very coarse spatial resolution. Other datasets, like CFSR, do not cover the required time period. While MERRA-2 and JRA-55 are high-quality datasets, ERA5’s superior spatiotemporal resolution makes it the most appropriate choice for a detailed, multi-scale wind analysis [25].

The WRF model was configured with a domain of 200 × 200 horizontal grid points and 45 vertical levels. Key physics parameterizations essential for regional climate simulation included the Kain-Fritsch (KF) cumulus scheme, the Thompson microphysics scheme, and the Noah-MP land surface model [26,27,28]. To enhance the simulation’s fidelity and reduce model biases, a sophisticated data assimilation and initialization scheme was implemented. First, to ensure the simulation remained consistent with large-scale atmospheric circulation, four-dimensional data assimilation (FDDA) in the form of grid nudging was applied to the upper-atmospheric fields (wind, temperature, and moisture) throughout the integration. Second, to establish a more realistic surface state at the beginning of each forecast cycle, the initial conditions (t = 0) for near-surface variables were not taken from ERA5 directly. Instead, they were replaced by fields generated from an objective analysis of observations (10-m winds, 2-m temperature, and 2-m humidity) from 19 national meteorological stations (detailed in Section 2.2.3).

Our simulation strategy employed continuous 36-h cycles, with the initial 12 h discarded as model spin-up to minimize errors from initial condition imbalances; the subsequent 24 h constituted the final reanalysis product. Throughout the simulations, sea surface temperature (SST) boundary conditions were continuously updated using ERA5 data to ensure a realistic representation of air-sea thermodynamic exchanges. While the HNR dataset spans 1940–2022, our analyses focus on the period 1961–2022 to ensure data quality and reliability, as most observational stations began continuous operation after 1958.

2.2.2. Datasets for Inter-Comparison and Verification

The performance of the HNR dataset was rigorously evaluated through comparison with four widely used reanalysis and gridded observational products: ERA5 reanalysis (0.25° resolution, hourly, 1940–2022), which also serves as the driving data for our downscaling [29]; China Meteorological Forcing Dataset (CMFD; 0.1° resolution, 3-hourly, 1979–2018) [30]; CN05.1 gridded observation dataset interpolated from station data (0.25° resolution, daily, 1961–2022) [31]; and NOAA-CIRES-DOE 20th Century Reanalysis V3 (NOAA 20C), a long-term global reanalysis (1.0° resolution, 3-hourly, 1836–2015) [32].

2.2.3. In Situ Observations for Assimilation and Verification

Surface observational data for model initialization and subsequent verification were sourced from two distinct networks provided by the China Meteorological Administration (CMA). The first network consists of 19 national meteorological stations, which provide high-quality, long-term hourly wind speed records, with most data spanning from 1958 to 2022. These high-fidelity stations serve a dual and critical role in this study. First, their observations were used to enhance the model’s initial conditions by replacing the near-surface fields (10-m winds, 2-m temperature, and 2-m humidity) at the start of each simulation cycle (t = 0). Second, their long-term, homogeneous records serve as a key reference for verifying the decadal trends captured by the HNR dataset [33]. A statistical summary of the observational network’s development, including annual station counts and data volume, is provided in Appendix A. The second network is a much denser array of 362 regional Automatic Weather Stations (AWS), providing hourly data for the contemporary period of 2016–2022 (station locations are shown in Figure 1). This dataset, which was not used in any part of the model initialization process, serves as the primary basis for the independent verification of the HNR’s ability to capture fine-scale spatiotemporal wind patterns. It is important to note, however, that the siting and maintenance standards for the regional AWS network can differ from the stringent requirements for the national stations. Their immediate surroundings may not always adhere to standard open-field conditions, leading to a greater susceptibility to localized environmental influences (e.g., from nearby trees and buildings). Consequently, these factors can introduce representativeness errors or biases into the wind speed measurements. This context is crucial for the proper interpretation of the verification results presented in Section 3.

2.2.4. Large-Scale Circulation and Climate Indices

To investigate the broader climatic context and potential drivers of observed changes in Hainan’s surface wind field, we analyzed 130 climate system indices from the CMA National Climate Center [34].

2.3. Methods

2.3.1. Physical Basis for Seasonal Division

The annual cycle of Hainan’s surface wind field is governed predominantly by the tropical monsoon system rather than the thermal transitions characteristic of temperate latitudes. Consequently, a standard four-season division is not physically representative. To better align our analysis with the underlying atmospheric dynamics, we partition the year into three distinct seasons based on prevailing monsoon regimes:

Cold Season (December–March): Dominated by the Siberian-Mongolian High, this period features cool, dry northeasterly winds originating from the Asian continent. This season corresponds to Hainan’s dry season and exhibits the strongest persistent winds of the year.

Warm Season (May–October): Driven by intense continental heating and thermal low formation, the regional pressure gradient reverses, producing warm, moist southwesterly or southeasterly monsoonal flows from surrounding seas. This period constitutes Hainan’s rainy season and experiences frequent tropical cyclone impacts.

Transitional Seasons (April and November): These months represent periods of large-scale circulatory adjustment when winter and summer monsoon systems compete for dominance. Wind direction is often variable, and mean wind speeds typically reach their annual minimum, marking monsoon onset and retreat phases.

2.3.2. Analysis of the Diurnal Cycle

To comprehensively characterize the diurnal evolution of the surface wind field and its long-term changes, our analysis focuses on four key synoptic times (Beijing Standard Time, BJT) selected to capture peak and transitional phases of diurnal wind circulation:

• 14:00 BJT: Maximum land-sea thermal contrast period with strongest sea breeze development;
• 02:00 BJT: Mature nocturnal land breeze phase following sufficient radiative cooling;
• 07:00 BJT: Near sunrise, marking the critical transition from nocturnal land breeze to daytime sea breeze regime;
• 19:00 BJT: Near sunset, when sea breeze circulation rapidly decays due to cessation of solar radiation.

2.3.3. Long-Term Trend Analysis

Long-term trends in wind speed and other climatic variables were quantified using the Theil-Sen estimator, a non-parametric method robust to outliers that ensures trend calculation stability. Statistical significance of identified trends was assessed using the Mann-Kendall (MK) test.

2.3.4. Definition of Widespread Strong and Weak Wind Events

To objectively identify days with spatially coherent, anomalous wind conditions, we established definitions for widespread strong and weak wind events:

Strong Wind Days: Identified using dual-threshold criteria: (1) daily mean wind speed at a given grid point must exceed its 95th percentile (P95) relative to the 1991–2020 climatological baseline, and (2) total area of grid points meeting this intensity threshold must exceed 20% of the island’s area. The specific P95 threshold field and spatial distribution are provided in Appendix C.

Weak Wind Days: Begin with an intensity threshold where daily mean wind speed falls below the 5th percentile (P05) of the 1991–2020 baseline. However, since a 5% spatial threshold would merely identify statistically average days, we employed a more stringent 7% spatial threshold (P07) to isolate genuinely widespread synoptically driven stagnant conditions. This threshold deliberately exceeds the statistical baseline to filter for physically significant calm-wind events.

Trend analysis robustness was confirmed through sensitivity tests using various spatial thresholds (10–30% for strong wind days; 3%, 5%, and 10% for weak wind days), yielding highly consistent results in both trend direction and magnitude.

2.3.5. Return Period Analysis of Extreme Winds

To evaluate changes in high-impact, infrequent wind event intensity, we applied Generalized Extreme Value (GEV) theory. For each grid point, annual maximum wind speed series were extracted using the block maxima approach. GEV distributions were fitted to estimate wind speeds corresponding to specific return periods (20-, 50-, and 100-year events). This analysis was performed separately for the 1961–1990 and 1991–2020 periods to quantify decadal shifts in extreme wind climatology.

2.3.6. Identification of Sea-Land Breeze Days (SLBDs)

Leveraging the high-resolution HNR dataset, we developed objective criteria to identify local Sea-Land Breeze Days (SLBDs) at each grid point. A day qualifies as an SLBD if it satisfies two conditions: (1) daily mean wind speed < 10 m/s (excluding days dominated by strong synoptic systems like tropical cyclones), and (2) diurnal wind speed range (difference between daily maximum and minimum) > 3 m/s (ensuring significant thermally-driven circulation presence). These criteria adapt established sea breeze research methodologies [18,35,36].

2.3.7. Wind Power Resource Assessment

Wind energy resource availability was quantified using Wind Power Density (WPD), calculated as WPD = ½$\mathsf{\rho }$v³, where $\mathsf{\rho }$ represents air density (standard value 1.225 kg/m³) and v is wind speed. All Wind Power Density (WPD) calculations were based on the 10-m wind fields from the HNR dataset to maintain consistency and minimize uncertainties associated with wind shear parameterization. Although extrapolation to typical turbine hub heights (e.g., 70 m) was tested using the power law with a shear coefficient (α = 0.14) derived from radiosonde data, the final analysis was restricted to 10-m winds to ensure robustness. Interannual wind resource stability was assessed using the Coefficient of Variation (CV), defined as the ratio of standard deviation to long-term mean annual WPD.

2.3.8. Teleconnection Analysis

To explore relationships between local wind variability and large-scale atmospheric forcing, we employed Spearman rank correlation coefficients. This non-parametric approach was selected for its robustness, as some data series deviated from normal distributions. Key seasonal wind metrics (mean wind speed, SLBD frequency) were correlated with 130 climate indices from the CMA National Climate Center. For each wind metric, the top 6 most correlated indices were identified through correlation analysis, with the top 3 classified as primary drivers and indices 4–6 as secondary factors, enabling systematic linkage between local climate variations and large-scale circulation patterns.

3. Results

3.1. Verification of the HNR Dataset

The following analysis represents a verification of HNR’s added value rather than a canonical validation, as the availability of truly independent datasets is limited. This added value is particularly evident when examining long-term wind speed anomalies. Wind speed anomalies were calculated relative to the 1961–1990 climatological mean, with a 5-year Savitzky-Golay low-pass filter applied to emphasize decadal-scale variability. Station observations represent averages from 19 long-term monitoring stations distributed across the island. As shown in Figure 2, for the post-1990 period, most datasets, including HNR, exhibit consistent patterns of island-averaged interannual variability, with wind speed anomalies generally confined within ±0.2 m/s and no significant long-term trends, characteristic of the tropical monsoon climate regime. During 1960–1967, in situ station data display extreme positive anomalies that decay sharply from nearly +1.5 m/s to +0.5 m/s. This abrupt change far exceeds the plausible range of natural climate variability and indicates data inhomogeneity. In contrast, both HNR and its driving field (ERA5) maintain stable anomaly states close to 0.0 m/s throughout this period, with minor fluctuations largely within ±0.1 m/s. This demonstrates the effectiveness of the dynamical reanalysis approach in producing climatologically homogeneous long-term time series, free from spurious trends that contaminate the original instrumental record.

Furthermore, HNR demonstrates significant added value over both its parent reanalysis (ERA5) and other gridded products. While HNR’s long-term trend aligns with ERA5 before 1990, its finer 5-km resolution enables much more precise representation of local topographically forced circulations. For the recent, higher-quality observation period (post-1990), HNR’s agreement with station observation is markedly superior to ERA5.

In comparison, other datasets show notable limitations for this region. CMFD v1.0 exhibits anomalous positive peaks after 2010 that are inconsistent with all other data sources. The CN05.1 interpolated product consistently underestimates the magnitude of recent wind speed anomalies, while the coarse 1.0° resolution of NOAA 20C reanalysis leads to excessive smoothing of regional variability. Therefore, we conclude that the HNR dataset provides an excellent balance of long-term homogeneity and high-fidelity representation of recent climate variability, making it a robust foundation for the detailed analyses presented in this study.

The model evaluation reveals a complex performance profile characterized by systematic biases and spatially varying skill. The HNR dataset achieves a moderately strong temporal correlation with observations (median r = 0.62), successfully capturing principal modes of wind speed variability (Figure 3a). However, performance is significantly undermined by severe systematic positive bias, with wind speeds consistently overestimated across the island (median bias = +2.40 m/s) (Figure 3b). This large positive bias is the primary contributor to high overall Root Mean Square Error (median RMSE = 4.1 m/s) (Figure 3c). This bias demonstrates high temporal consistency throughout the study period (Figure 2), indicating that while the absolute wind speeds are overestimated, the integrity of the long-term trend analysis is not compromised. Spatial analysis reveals a paradoxical relationship between model skill in different regions: coastal plains exhibit “high correlation, high error,” while central mountainous areas show “low correlation, low error.” In coastal regions, the wind field is governed by well-defined large-scale systems like seasonal monsoons and diurnal sea-land breezes. The model adeptly simulates the temporal rhythm of these predictable phenomena, resulting in relatively high correlation coefficients (typically r > 0.6). However, these same high-wind regions produce the largest errors, likely due to the model’s inherently smoother topography and inadequate land use parameterization of surface roughness for complex coastal and tropical vegetation. This leads to underestimated surface friction and insufficient kinetic energy dissipation, causing severe wind speed overestimation. Conversely, within the complex central mountain terrain, the model’s 5-km resolution cannot explicitly resolve microscale wind fields generated by topographic blocking, flow channeling, and local thermal gradients. This scale mismatch results in poor temporal agreement between grid-averaged simulations and point observations, yielding low correlation coefficients (typically r < 0.6). Paradoxically, because observed wind speeds in these sheltered regions are inherently low, even with the model’s overestimation tendency, absolute bias and RMSE values remain much smaller than in high-wind coastal zones. In summary, the HNR dataset exhibits systematic wind speed overestimation, most pronounced in high-wind coastal regions, likely originating from insufficient simulation of topographic form drag and land-surface frictional effects.

3.2. Multi-Scale Climatology and Trends

3.2.1. Annual Mean and Long-Term Trends

The annual mean wind climatology of Hainan Island exhibits distinct topographically controlled patterns (Figure 4). Both long-term mean and maximum wind speeds display high spatial coherence, characterized by concentric structures with high winds along the coast and low-wind cores over the mountainous interior. In coastal plains, where terrain is open and surface roughness is minimal, direct marine airstream influence produces mean wind speeds exceeding 4.0 m/s and maximum speeds often surpassing 8.0 m/s. Conversely, central mountain ranges, including Wuzhi (109.6° E, 18.8° N) and Limu Mountains (109.7° E, 19.2° N), create significant barriers that generate sheltered zones where topographic blocking and enhanced friction reduce mean wind speeds below 3.0 m/s and maximum speeds under 6.0 m/s. This sharp coast-to-interior wind speed gradient manifests the interaction between prevailing tropical monsoon flow and complex terrain.

Comparison of Figure 4a,creveals that maximum wind speed spatial variability (median: 5.5 m/s) considerably exceeds that of mean wind speed (median: 3.6 m/s), indicating that topography exerts more pronounced control on extreme wind events, particularly in coastal regions most exposed to typhoons and strong monsoonal surges.

Long-term trends in annual mean and maximum wind speeds exhibit significant spatially divergent patterns. Notable “wind stilling” occurs across northeastern coastal plains, with statistically significant mean wind speed decreases of −0.03 to −0.09 m/s decade⁻¹ (Figure 4b). In striking contrast, the island’s interior—central mountains and southern hills with the lowest climatological wind speeds—shows significant strengthening trends often exceeding 0.06 m/s decade⁻¹. This dipole pattern is evident in maximum wind speeds (Figure 4d), where high-wind coastal areas in the east and north exhibit significant weakening trends (up to −0.15 m/s decade⁻¹), while low-wind central and southern interior regions show strengthening trends often above 0.05 m/s decade⁻¹.

Despite these strong regional trends, the island-averaged trend remains nearly neutral (median trend: −0.004 m/s decade⁻¹ for mean speed). This “coastal stilling, inland strengthening” pattern demonstrates that the island’s wind field is not undergoing uniform change but rather responding to long-term climatic shifts in a spatially complex and heterogeneous manner.

3.2.2. Seasonal Patterns and Trends

The annual “coastal stilling, inland strengthening” trend pattern reflects distinct seasonal dynamics, each governed by different phases of the tropical monsoon system.

During the warm season (May–October), the prevailing Southwest Monsoon establishes classic topographically controlled wind patterns (Figure 5a). The southwestern and southeastern coasts, serving as primary windward shores, experience mean wind speeds generally exceeding 4.0 m/s. However, the overall wind field is less energetic than during the cold season, with an island-wide median speed of 3.4 m/s. Central mountain ranges present formidable barriers to monsoonal flow, creating extensive low-wind zones (<3.0 m/s) on leeward (northeastern) slopes and within the mountainous core. The long-term trend during this season (Figure 5b), analyzed using the Theil-Sen robust regression method for 1961–2022, exhibits a striking inverse relationship with mean climatology. The windiest coastal plains, particularly along northern, eastern, and western shores, show significant stilling trends of −0.06 to −0.12 m/s decade⁻¹. Conversely, the calmest interior regions—central mountains and southern hills—display significant strengthening trends of 0.03 to 0.1 m/s decade⁻¹. This pronounced dipole pattern produces a near-zero island-averaged trend (−0.016 m/s decade⁻¹).

The cold season (December–March) represents Hainan’s windiest period, with an island-wide median wind speed of 3.9 m/s, driven by strong northeasterly flow of the East Asian Winter Monsoon and frequent cold air outbreaks (Figure 5c). As direct windward shores, northern and eastern coasts experience exceptionally high wind speeds, with large areas exceeding 5.0 m/s and some localities surpassing 6.0 m/s, forming the island’s primary wind energy resource zone. When this powerful northeasterly flow encounters central mountains, it is forced to ascend and diverge, creating pronounced “wind shadows” in leeward southwestern regions where speeds drop sharply below 3.5 m/s. The resulting spatial wind speed gradient is the steepest among all seasons, illustrating classic orographic effects of windward enhancement and leeward sheltering. The long-term trend pattern (Figure 5d) closely mirrors the warm season, confirming the robust “high-wind areas weaken, low-wind areas strengthen” dipole. The most intense stilling coincides precisely with climatologically windiest northern and eastern coasts, with rates of −0.06 to −0.12 m/s decade⁻¹, suggesting potential weakening of the broader East Asian Winter Monsoon system. Concurrent strengthening over mountainous interior regions is equally evident during this season.

Transitional months (April and November) represent complex interactions between retreating winter and advancing summer monsoons. The wind field reflects this dynamic struggle, exhibiting hybrid cold and warm season characteristics (Figure 5e). The island-wide median wind speed (3.8 m/s) is comparable to the cold season, indicating continued influence of strong synoptic systems. While high winds persist along coasts, the extreme high-wind zone in the northeast contracts compared to the cold season, while southern coastal winds strengthen, reflecting shifting monsoonal regime balance. The low-wind core over central mountains remains stable, highlighting persistent topographic barrier effects regardless of background flow. Critically, long-term trend analysis reveals highly significant features during this period (Figure 5f). While the “coastal stilling, inland strengthening” pattern persists, stilling magnitude is strongest during transitional months, with an island-wide median trend of −0.025 m/s decade⁻¹. Wind speed decline rates along coasts—particularly in the north, east, and southwest—are more pronounced than in any other season, with some areas exceeding −0.12 m/s decade⁻¹. This finding strongly suggests that long-term decline in Hainan’s wind speeds is not temporally uniform but is most acute during critical monsoon onset and retreat periods. This may indicate that ongoing climate change affects not only the mean monsoon strength but also profoundly alters seasonal transition dynamics.

3.2.3. Diurnal Evolution of the Wind Field

The diurnal cycle of Hainan’s wind field reveals fundamentally different controlling mechanisms between the two main seasons, highlighting the interplay between local thermal forcing and large-scale monsoon flow.

Nocturnal and Early Morning Regime (02:00 and 07:00 BJT)

During the warm season, late night and sunrise periods are characterized by classic radiatively driven quiescent conditions (Figure 6a,g). With sea-breeze circulation fully dissipated, the island experiences weak and variable winds, typically 2.0–3.0 m/s. The absence of defined spatial gradients between the coast and the interior signifies minimum kinetic energy periods. This is quantitatively confirmed by wind speed histograms showing low island-wide median values (~3.0 m/s) and narrow, sharply peaked distributions, indicating spatially homogeneous calm conditions across the island. In contrast, the cold season’s nocturnal and sunrise wind fields remain under firm synoptic-scale Northeast Monsoon control, with minimal local diurnal effects (Figure 6b,h). Spatial patterns closely resemble 24-h means: strong, persistent northeasterly winds (>4.0 m/s) impact windward northern and eastern coasts, while pronounced wind shadows with low speeds (<3.0 m/s) persist on leeward southwestern slopes. Wind speed histograms for this period are broad without sharp peaks, with higher island-wide medians (~3.5 m/s), reflecting significant persistent spatial heterogeneity from strong monsoon-topography interactions.

Afternoon Peak and Sea-Breeze Development (14:00 BJT)

Afternoon periods mark peak diurnal wind activity, driven by different mechanisms each season. In the warm season, wind fields undergo dramatic transformation driven by intense solar heating (Figure 6e). Powerful island-wide sea-breeze circulation develops, forming coherent rings of strong onshore winds (4.5–5.5 m/s) around the entire coastline. This period exhibits the highest warm-season island-wide median wind speed (4.3 m/s) and greatest spatial variability, shown by very broad histogram distributions capturing large differences between strong coastal winds and relatively calm mountainous interiors. In the cold season, afternoon conditions produce the most intense annual wind speeds through constructive superposition of local sea breezes and powerful synoptic monsoons (Figure 6f). On northern and eastern coasts where flows are co-directional, background monsoons are significantly amplified, creating extensive areas with speeds exceeding 6.0–7.0 m/s. Resulting histograms are the broadest of all analyzed periods and skewed toward high values, indicating widespread, powerful winds.

Evening Transition (19:00 BJT)

Sunset periods demonstrate differential decay of these wind systems. In the warm season, thermally driven sea breezes collapse rapidly with solar radiation cessation (Figure 6c). Strong coastal wind belts dissipate, and the entire island quickly returns to quiescent, spatially uniform states, with histograms reverting to narrow, sharply peaked nocturnal forms. In the cold season, while additional sea-breeze velocity components vanish, underlying synoptic monsoons persist (Figure 6d). Windward coastal wind speeds decrease from afternoon peaks but remain strong (>4.0 m/s), and wind speed distributions, while narrower than afternoon patterns, remain significantly broader than warm season conditions, reflecting continued dominance of topographically controlled synoptic flow.

Long-term trends in Hainan’s diurnal wind cycle consolidate into two robust, season-dependent spatial patterns that provide insights into underlying drivers of change.

Throughout the cold season, persistent and powerful trend dipoles are observed at all hours. This pattern is characterized by significant wind stilling on windward (northeastern) coasts and robust strengthening on leeward (southwestern) slopes. This “windward weakening, leeward strengthening” dipole is remarkably stable, evident during late night (Figure 7h), morning (Figure 7b), and evening (Figure 7d), reaching maximum spatial contrast and intensity during afternoon (Figure 7f). This consistent 24-h pattern strongly suggests governance by long-term changes in large-scale synoptic systems—namely, systematic weakening of the East Asian Winter Monsoon. Observed stilling on directly exposed windward coasts directly manifests this weakening. More complexly, wind strengthening in historically sheltered “wind shadow” regions indicate significant changes in flow-topography interactions. We hypothesize that this results from dynamic adjustment: as powerful monsoon flow weakens, its mountain interactions change. Weaker flow becomes more susceptible to generating enhanced turbulence and stronger lee-side vortices, manifesting as increased mean wind speeds. Concurrently, weakening dominant monsoons reduce their ability to suppress smaller-scale local circulations (e.g., valley winds) in leeward regions, allowing greater contributions to local wind fields and producing overall wind speed increases.

In contrast, warm season trends are governed by the long-term evolution of local thermal circulation (i.e., sea breeze). This manifests primarily as widespread, albeit weaker, wind stilling, most pronounced and spatially coherent in coastal plains (Figure 7a,g). While faint “coastal stilling, inland strengthening” dipoles remain discernible, particularly in the morning (Figure 7a), the dominant signal is broad weakening. Critically, stilling trend magnitude is directly coupled to diurnal thermal forcing strength. Trends are most significant and widespread during the afternoon (Figure 7e), coinciding with climatological sea breeze peaks. Conversely, they are weakest and most spatially fragmented during night and evening (Figure 7c,g), when thermal forcing is minimal or absent. This strong correspondence suggests that warm-season stilling primarily results from long-term sea breeze weakening. This is consistent with hypotheses of reduced land-sea thermal contrast under global warming scenarios, which would diminish the primary driver of this local circulation system, representing sensitive local climate system responses to large-scale changes.

3.3. Frequency and Trends of Extreme Wind Events

3.3.1. Strong Wind Days

During the warm season (May–October), the time series of widespread strong island-wide wind days from 1961 to 2022 exhibits a subtle but statistically significant increasing trend (Figure 8a). Theil-Sen trend analysis reveals that the frequency of these events has increased at 0.097 days per decade⁻¹. This long-term increase is superimposed on strong interannual and decadal variability, with notably active periods in the early 1970s, around 1990, and after 2010, when annual frequency could exceed 10 days. Conversely, the early 2000s represent a relatively quiescent period.

To ensure this trend is not an artifact of our event definition, we conducted a sensitivity analysis using various spatial thresholds (P10 to P30). The resulting time series shows high synchrony in variability and trends, confirming the robustness of our findings. This suggests systematic increases in either frequency or spatial extent of strong wind events during the warm season. In contrast, the cold season (December–March) displays pronounced and significant decreasing trends in widespread strong island-wide wind days over the entire 62-year period (Figure 8b). The decline rate, calculated via Theil-Sen analysis, is −0.214 days decade⁻¹—more than double the magnitude of the warm-season increase. These cold-season events are primarily driven by intense cold air outbreaks associated with southward winter monsoon advances. Therefore, this sharp negative trend provides compelling evidence for the well-documented multi-decadal weakening of the East Asian Winter Monsoon system. Although the series contains notable peaks, such as in the late 1960s and mid-1990s, the overall downward trajectory is unequivocal, with event frequency in the latter half of the record being substantially lower than in the first half.

During the warm season (May–October), strong wind day frequency exhibits significant spatial heterogeneity, with patterns influenced by local topographic features and potentially convective activity (Figure 9a). To examine representative spatial distributions, we analyzed composite patterns from years with the highest island-wide frequencies (1969, 1973, 1977, 1989, 1994, 1997, 2013, 2018, and 2020, as identified from Figure 8). The highest frequencies (13–17 days year⁻¹) concentrate in the island’s central-western and southern mountainous and hilly regions. In contrast, the lowest frequencies (7–11 days year⁻¹) occur in northeastern and southeastern plains and hills. While the island-wide median frequency is 11.8 days year⁻¹, spatial contrast is considerable, with mountainous areas experiencing 3–8 more strong wind days annually than plains.

During the cold season (December–March), strong wind day distribution becomes highly organized, clearly reflecting dominant Northeast Monsoon interactions with island topography (Figure 9b). A continuous coastal belt of high frequency (12–16 days year⁻¹) extends across the entire western and southern plains, which are leeward of mountains relative to prevailing northeasterlies but may experience flow acceleration. The lowest frequencies (6–8 days year⁻¹) occur in the northeastern hills and mountains. Although the island-wide median frequency is identical to the warm season (11.8 days year⁻¹), the cold season distribution histogram is noticeably broader, reflecting a more polarized spatial pattern with steeper gradients between high- and low-frequency zones. The difference between most and least active regions can reach 8–10 days year⁻¹, considerably greater than warm season ranges. This pattern represents a classic manifestation of synoptic-scale flow interacting with complex terrain.

3.3.2. Weak Wind Days

During the warm season (May–October), the frequency of widespread weak wind days on Hainan Island exhibits no significant long-term trend from 1961 to 2022 (Figure 10a). The Theil-Sen trend line is nearly horizontal, with a calculated rate of −0.007 days decade⁻¹, statistically indistinguishable from zero. This indicates that large-scale stagnant weather event occurrence during the warm season has remained stable over the past six decades. These events, physically associated with South China Sea summer monsoon breaks, subtropical high control periods, or calm conditions between tropical cyclone passages, show considerable interannual and decadal variability. Notably active periods occurred in the mid-1970s, mid-1990s, and around 2010.

Sensitivity tests using various spatial thresholds (P03, P05, and P10) all confirm this lack of significant long-term trends, underscoring conclusion robustness. Thus, while mean wind speeds may have changed, the frequency of extreme island-wide calm events in the warm season has been historically stable. Conversely, the cold season (December–March) frequency of widespread weak island-wide wind days shows a subtle but statistically significant decreasing trend, with a Theil-Sen rate of −0.021 days decade⁻¹ (Figure 10b). These events correspond to East Asian Winter Monsoon lulls—periods without strong cold air outbreaks when synoptic conditions are unusually placid. This finding presents an intriguing dynamic: in conjunction with the sharp decline in strong wind days, it suggests that the cold season is experiencing reduced extreme events at both ends of the wind speed spectrum.

This seemingly paradoxical result—simultaneous decreases in both the windiest and calmest days—points toward more complex winter climate adjustment than simple monsoon weakening. A plausible hypothesis is that the overall variability of the winter monsoon system is decreasing. While the most intense cold air surges (driving strong wind days) have become less frequent, the most quiescent, stable atmospheric conditions allowing widespread calm are also occurring less often. This suggests wind speed distribution contraction, leading to more days within moderate ranges rather than simple downward shifts of the entire wind regime.

During the warm season (May–October), weak wind day frequency displays clear concentric patterns, with the highest frequency in the island’s interior, decreasing radially toward coasts (Figure 11a). Peak frequencies, exceeding 11 days year⁻¹, concentrate in central hills and mountains, largely coinciding with the highest terrain. From this core, frequency decreases to 9–11 days year⁻¹ in surrounding hills and 5–9 days year⁻¹ in coastal plains, with the eastern coast experiencing fewest events (<7 days year⁻¹). The island-wide median frequency is 9.0 days year⁻¹, and the relatively symmetric histogram distribution indicates that while inland maxima exist, weak wind days are common features with moderate spatial variability across the island during this season.

This pattern is completely inverted during the cold season (December–March), characterized by spatially homogeneous, exceptionally infrequent weak wind events (Figure 11b). For the vast majority of the island, annual weak wind day frequency is less than 2 days year⁻¹. This is reflected in the island-wide median of just 1.3 days year⁻¹—an 86% reduction compared to the warm season median. The highly right-skewed histogram quantitatively demonstrates this rarity, with over 75% of the island’s area experiencing fewer than 2 weak wind days annually. The only exceptions are small, isolated pockets in southwestern mountains and hills, which can experience 4–8 events annually. This island-wide suppression of calm conditions underscores the pervasive influence of the strong winter monsoon, which maintains windy conditions nearly everywhere, making widespread stagnation a rare phenomenon.

3.3.3. Changes in Extreme Wind Speed Return Periods

Baseline Climatology (1991–2020)

The climatology of extreme wind events during the recent period (1991–2020), estimated by Generalized Extreme Value (GEV) analysis, exhibits spatially coherent patterns consistent across all return periods. This pattern is characterized by high return-level winds along coasts and significantly lower values in mountainous interiors. For the 20-year return period (Figure 12a), a ring of high-velocity winds (>11 m/s) encircles coastal plains, with the most intense values concentrated along northeastern shorelines. This directly reflects the dominant contribution of tropical cyclones to the island’s extreme wind climate. In contrast, central Wuzhi and Limu mountain ranges create pronounced sheltering effects, where 20-year return-level winds fall below 7 m/s. Spatial patterns for 50-year (Figure 12c) and 100-year (Figure 12e) return periods are nearly identical but with systematically higher wind speeds, as indicated by island-wide median values increasing from 8.7 m/s for 20-year events to 9.9 m/s and 10.8 m/s for 50- and 100-year events, respectively.

Decadal Changes in Return Levels (1991–2020 vs. 1961–1990)

Comparison between the two 30-year climate normals reveals robust, spatially divergent patterns of change in extreme wind return levels that are remarkably consistent across all frequencies. This pattern mirrors long-term trends observed in mean wind climatology. For the 20-year return period (Figure 12b), widespread and significant increases in extreme wind speeds have occurred across central mountains and southern hills, with the most pronounced strengthening (+0.4 to +0.7 m/s) in south-central mountains. Conversely, a continuous belt of decreasing extreme wind speeds (−0.2 to −0.5 m/s) is evident along eastern and northeastern coastal plains, from Wenchang (110.8° E, 19.6° N) to Wanning (110.4° E, 18.8° N).

This same dipole pattern—interior strengthening and eastern coastal weakening—is consistently observed for 50-year (Figure 12d) and 100-year (Figure 12f) return periods. Change magnitudes remain similar across different return levels, with central mountain strengthening consistently reaching or exceeding +0.7 m/s and eastern coastal weakening remaining around −0.5 m/s for 100-year events.

3.4. Climatology and Trends of Sea-Land Breeze Days (SLBDs)

During the warm season (May–October), when synoptic forcing is weak and solar radiation is strong, sea-land breeze circulation is a dominant feature of Hainan’s coastal climate. The climatological frequency of SLBDs exhibits pronounced topographic gradients, decreasing from coast to interior (Figure 13a). In coastal plains, SLBDs are exceptionally frequent, occurring on 140–180 days, accounting for 76–98% of all seasonal days. This frequency drops sharply in the central mountains to just 40–80 days (22–43% of the season). The long-term trend in warm-season SLBDs (Figure 13b) displays a familiar spatial pattern: decreases in coastal plains, most significantly along the northeastern coast (−1.0 to −7.0 days decade⁻¹), and concurrent increases in inland hills (+1.0 to +3.0 days decade⁻¹). The island-wide median trend is a slight decrease of −1.0 days decade⁻¹, suggesting net weakening of this local circulation system in areas where it is most prevalent.

In the cold season (December–March), the powerful Northeast Monsoon consistently suppresses local thermal circulation development, leading to dramatic SLBD frequency reductions (Figure 13c). The number of SLBDs in coastal plains drops to 80–120 days, while mountainous interiors experience only 20–60 events. The island-wide median frequency falls to 82.7 days, a 35% reduction (44.2 fewer days) compared to the warm season. This seasonal suppression is statistically evident as a wholesale shift in the frequency distribution toward lower values. The long-term trend during the cold season (Figure 13d) reveals spatial patterns distinct from the warm season. The most significant decreases concentrate in western coastal plains (−3.0 to −5.0 days decade⁻¹), while notable increases occur in south-central hills and along southern coastal plains (+1.0 to +3.0 days decade⁻¹). This asymmetrical “west-decrease, south-increase” pattern contrasts sharply with the more concentric warm season trend pattern, suggesting different mechanisms of long-term change, even though the island-wide median trend (−1.0 days decade⁻¹) is identical.

3.5. Spatial Patterns and Stability of Wind Power Resources

Assessment of near-surface (10-m) wind power potential on Hainan Island reveals stark spatial disparities in both resource abundance and interannual stability. While actual wind turbine hub heights typically operate around 70 m where wind speeds are higher, this assessment employs 10-m wind speeds to maintain consistency with standard meteorological observations and provide a conservative, uncertainty-reduced baseline for resource evaluation. The annual mean Wind Power Density (WPD), a metric for resource richness, exhibits pronounced spatial patterns (Figure 14a). Coastal plains possess the most significant resources, with WPD values generally ranging from 170 to 260 W m⁻². Several hotspots, particularly along northeastern, eastern, and northern coastlines, exceed 260 W m⁻². In sharp contrast, mountainous and hilly interiors are resource-poor, with WPD values below 80 W m⁻² and often as low as 50 W m⁻². The island-wide median WPD is only 55.1 W m⁻², and the highly right-skewed distribution indicates that the most valuable wind resources are geographically concentrated in a relatively small portion of the island’s total area.

The Coefficient of Variation (CV) of annual mean WPD, which quantifies resource interannual stability, displays spatial patterns inverse to WPD itself (Figure 14b). The resource-poor interior is characterized by high instability, with CV values typically ranging from 0.3 to 0.6, indicating significant year-to-year uncertainty. Conversely, resource-rich coastal plains exhibit high temporal stability, with CV values generally below 0.2. This analysis reveals critical spatial coupling: areas with the highest resource potential also exhibit the greatest temporal stability (low CV), while resource-poor regions are simultaneously characterized by high interannual uncertainty (high CV). This robust “high-resource, high-stability” versus “low-resource, low-stability” relationship provides a clear quantitative basis for strategic energy planning. It identifies coastal plains as prime locations for wind energy development, not only for their resource abundance but also for their high reliability and year-to-year predictability. The interior, however, presents dual challenges of both resource scarcity and high uncertainty.

3.6. Teleconnection Patterns Analysis

Cold season mean wind speed variability is primarily governed by the strength and position of subtropical highs and mid-latitude circulation patterns (

Table 2. Primary and secondary climate drivers for cold-season wind characteristics over Hainan Island (1961–2022).

Local Metric	Factor Tier	Climate Factor	r
Cold Season Mean Wind Speed	Primary	Pacific Subtropical High Ridge Position	+0.42 **
		Indian Subtropical High Ridge Position	+0.37 **
		North Pacific Pattern (NP)	−0.35 **
	Secondary	South China Sea Subtropical High Ridge Position	+0.34 *
		East Asian Trough Intensity	−0.30 *
		Pacific Polar Vortex Intensity	+0.30 *
Cold Season SLBDs Count	Primary	Tropical-Northern Hemisphere Pattern (TNH)	−0.53 **
		Area of the Indian Ocean Warm Pool	+0.47 **
		Tibet Plateau Region 2 Index	+0.42 **
		Northern Hemisphere Polar Vortex Intensity	−0.41 **
	Secondary	Intensity of the Indian Ocean Warm Pool	+0.40 *
		East Pacific 850 mb Trade Wind	−0.31 *
		East Asian Trough Intensity	+0.31 *

Notes: * p < 0.05; ** p < 0.01. The complete list of climate indices is provided in Appendix B, Table A1.

). The most significant drivers include the ridge position of the North Pacific Subtropical High (r = +0.42) and Indian Subtropical High (r = +0.37), along with the North Pacific (NP) teleconnection pattern (r = −0.35). Strong positive correlations with subtropical high ridge positions indicate that more northward placement of these systems steepens pressure gradients across Southern China, thereby enhancing northeasterly background flow over Hainan. Conversely, the negative correlation with the NP index reflects East Asian Winter Monsoon modulation; a positive NP phase, corresponding to a weaker Aleutian Low, leads to weaker winter monsoons and lower mean wind speeds. Secondary yet significant drivers include the South China Sea Subtropical High position (r = +0.34) and the intensities of the East Asian Trough (r = −0.30) and Pacific sector polar vortex (r = +0.30).

In contrast, factors promoting local sea-land breeze formation are those that suppress large-scale mean flow, creating quiescent conditions. SLBD frequency is most strongly correlated with tropical heat sources and high-latitude variability. Dominant drivers include the Tropical-Northern Hemisphere (TNH) pattern (r = −0.53), Indo-Pacific Warm Pool area (r = +0.47), Tibetan Plateau-2 Index (r = +0.42), and Northern Hemisphere polar vortex strength (r = −0.41). The strong positive correlation with Indo-Pacific Warm Pool area is particularly insightful: a larger warm pool enhances regional deep convection, which through atmospheric teleconnections weakens northeasterly monsoon flow over Hainan, creating more favorable environments for local thermal circulation development.

Large-scale drivers of Hainan’s warm-season wind field differ markedly from the cold season, revealing shifts toward higher-latitude influences and more complex factor interactions (

Table 3. Primary and secondary climate drivers for warm-season wind characteristics over Hainan Island (1961–2022).

Local Metric	Factor Tier	Climate Factor	r
Warm Season Mean Wind Speed	Primary	Pacific Transition Pattern (PT)	−0.42 **
	Secondary	Tibet Plateau Region 2 Index	−0.24 *
		Pacific Polar Vortex Area	+0.24 *
		Tibet Plateau Region 1 Index	−0.23 *
Warm Season SLBDs Count	Primary	North American Polar Vortex Intensity	−0.36 **
		Atlantic-European Polar Vortex Intensity	−0.35 **
		Northern Hemisphere Polar Vortex Intensity	−0.33 **
		Pacific Transition Pattern (PT)	−0.32 *
	Secondary	Pacific Polar Vortex Intensity	−0.30 *
		Eurasian Zonal Circulation	−0.29 *
		Asia Polar Vortex Intensity	−0.26 *

Notes: * p < 0.05; ** p < 0.01. The complete list of climate indices is provided in Appendix B, Table A1.

).

Variability in warm season mean wind speed is most significantly linked to the Pacific Transition (PT) teleconnection pattern (r = −0.42). Secondary influences include indices related to the Tibetan Plateau’s thermal state (e.g., Tibetan Plateau-2 Index, r = −0.24), reflecting the complex modulating effect of the plateau on downstream monsoon systems. A positive correlation with Pacific-sector polar vortex area (r = +0.24) also suggests that high-latitude circulation anomalies can remotely influence low-latitude monsoon wind fields.

Drivers for warm-season SLBD frequency are again distinct and overwhelmingly dominated by high-latitude circulation systems. A suite of polar vortex-related indices—including those for North America, North Atlantic-Europe sector, Northern Hemisphere as a whole, and Pacific sector—all show significant negative correlations (r ranging from −0.53 to −0.32). This strong, consistent negative relationship is a key finding: it implies that stronger, more consolidated polar vortices are associated with large-scale atmospheric conditions that suppress local thermally driven sea breeze formation over Hainan, likely by enabling more frequent or stronger mid-latitude disturbances to reach the subtropics.

A clear seasonal dichotomy emerges in large-scale controls of Hainan’s wind field. The cold season regime is robustly governed by the strength and position of lower-latitude subtropical high-pressure systems, which directly dictate winter monsoon force. In contrast, the warm season regime is more subtly influenced by combinations of mid-latitude teleconnections and, critically, strong “top-down” influences from high-latitude polar vortex systems, which appear to be key factors in modulating conditions required for local circulations like sea breezes. Generally weaker correlations and multiple influencing factors during the warm season suggest lower climatic predictability for summer wind conditions compared to winter, which is dominated by more direct and powerful synoptic drivers.

4. Discussion

Before discussing the scientific implications of our findings, it is crucial to address the systematic positive wind speed bias of ~+2.40 m/s identified during the verification process. This overestimation is not an isolated issue but reflects a persistent and well-documented challenge in regional climate modeling with WRF, particularly when driven by reanalysis data over complex terrain [37,38,39]. The most significant cause of wind speed overestimation in complex terrain is the model’s inability to resolve true topographic features. The model’s gridded elevation data inherently “smooths” out the landscape, replacing sharp peaks and deep valleys with gentler slopes [40,41]. This process neglects the immense momentum loss caused by subgrid-scale form drag—the force exerted by airflow blocking, separating, and flowing around real-world rugged terrain. In essence, the model simulates wind over a much smoother surface than reality, leading to insufficient drag and consequently higher wind speeds. Similar to topographic drag, the friction from vegetation and buildings is represented by a “roughness length” parameter tied to land-use categories [42,43]. In a heterogeneous landscape like Hainan, a single value per grid cell is often an oversimplification that underestimates the true aerodynamic resistance of the surface. This underestimation of friction contributes directly to the positive wind speed bias [42,44].

A primary finding of this study is the robust “coastal stilling, inland strengthening” dipole pattern in wind speed trends that persists across annual, seasonal, and diurnal scales from 1961 to 2022. The wind stilling observed in Hainan’s coastal plains is consistent with the widely documented global phenomenon, which is generally attributed to both weakening large-scale atmospheric circulation and increases in surface roughness [11]. Crucially, our study’s methodology provides a unique opportunity to disentangle these two drivers. By employing a static land-use and land-cover (LULCC) map for the entire 62-year simulation, our approach effectively creates a numerical experiment designed to isolate and quantify the independent contribution of large-scale circulation changes, free from the confounding effects of historical LULCC. Our results, therefore, demonstrate that the documented long-term weakening of the East Asian Winter Monsoon system is, by itself, sufficient to produce the significant stilling effect observed on the coasts of Hainan. This is a critical point, as it allows our work to quantify the circulation-driven portion of the “coastal stilling” phenomenon, which is understood in the real world to be a composite of both weakening circulation and increased roughness [12].

However, the concurrent wind strengthening in Hainan’s inland mountainous areas represents a significant departure from the global stilling narrative. This enhancement does not result from strengthened background flow but rather reflects complex dynamic adjustments arising from weakened monsoon-terrain interactions. As the powerful northeasterly monsoon flow diminishes, its interaction with the central mountain ranges fundamentally changes: weaker flow becomes more susceptible to generating enhanced local turbulence and lee-side vortices in the southwestern “wind shadow” regions, while local thermal circulations such as valley winds that were previously suppressed by strong background winds can now develop more freely. These processes collectively produce paradoxical increases in mean wind speeds within formerly sheltered areas, demonstrating that topographic modulation can reverse basin-scale climate signals. This discovery underscores that in regions of complex terrain, climate change impacts on wind fields exhibit non-linear responses that deviate substantially from large-scale patterns, necessitating high-resolution dynamical downscaling to detect climate change signals masked or altered by topographic interactions.

The contrasting evolution of extreme wind events reveals distinct governing physical systems: significant decreases in cold-season strong wind days versus slight increases during the warm season. The decline in cold-season extremes directly corresponds with the weakening East Asian Winter Monsoon and reduced frequency of intense cold air outbreaks, corroborating similar trends across East Asia. Conversely, the slight increase in warm-season strong wind days may reflect changes in tropical cyclone activity, though this requires dedicated investigation. The subtle but significant decrease in cold-season weak wind days presents an intriguing paradox that suggests “moderation” of the entire winter monsoon system, where both extreme high-wind events and extreme calm conditions are becoming less frequent—a contraction of the wind speed probability distribution from both tails that may characterize transitional climate states.

Our analysis of extreme wind speed return periods reveals systematic geographic redistribution of wind risk. While maintaining the baseline “coastal high, inland low” pattern, comparison between 1961–1990 and 1991–2020 periods shows significant changes: central mountains and western hills exhibit increases of +0.4 to +0.7 m/s in return-level wind speeds, while eastern coastal plains show decreases of −0.2 to −0.5 m/s. These statistically robust changes across 20-, 50-, and 100-year return periods indicate systematic shifts demanding a transition from uniform to geographically differentiated design wind speeds. Infrastructure in central and western regions should incorporate additional safety margins of +0.4 to +0.7 m/s, while new construction in eastern coastal plains could optimize wind-loading requirements by 2–5% without compromising safety.

The quantified changes in sea-land breeze frequency provide precise parameters for climate-adaptive infrastructure design. The warm-season average of 126.9 sea-land breeze days offers significant natural ventilation potential for coastal buildings, though the observed declining trend necessitates increased redundancy in mechanical ventilation systems. For transportation infrastructure, the reduction in cold-season sea-land breeze days indicates more persistent background wind fields, potentially increasing crosswind events at airports and operational disruptions at ports.

These findings demonstrate the critical importance of incorporating high-resolution, regionally specific climate change signals into infrastructure planning. The transition from static design standards to dynamic, adaptive strategies based on quantified climate trends is essential for building truly resilient infrastructure systems in tropical coastal regions. While this study focuses on Hainan Island, the methodological framework and physical insights have broader applicability to similar island and coastal mountain environments worldwide, particularly in monsoon-influenced regions undergoing rapid development. To build upon these insights, future work should focus on providing more direct evidence for the proposed dynamic adjustment mechanism by performing dedicated diagnostic analyses—for instance, by examining vorticity fields and vertical wind profiles on the leeward side of the mountains during contrasting strong and weak monsoon periods.

5. Conclusions

This study utilized a newly developed, high-spatiotemporal resolution (5-km/hourly) regional climate reanalysis dataset (HNR) to conduct a systematic analysis of multi-scale climate characteristics, long-term trends, and extreme events of Hainan Island’s near-surface wind field from 1961 to 2022. The principal conclusions are as follows:

(1)

The HNR dataset provides a robust foundation for regional climate analysis. Comprehensive verification against multi-source datasets and in-situ observations demonstrates that the HNR exhibits superior long-term homogeneity and high-fidelity representation of recent conditions. It successfully overcomes biases present in raw station data and accurately reproduces the topographically governed “strong coastal, weak inland” climatological wind patterns. Furthermore, its high spatiotemporal resolution makes it a valuable asset for future, event-based case studies, such as the detailed hindcasting of specific tropical storms.

(2)

Hainan’s long-term wind speed trends exhibit significant and spatially heterogeneous dipole patterns of “coastal stilling, inland strengthening.” Over the past six decades, annual mean wind speed in coastal plains has decreased at rates of −0.03 to −0.09 m/s decade⁻¹, while wind speed in mountainous interiors has paradoxically increased, with rates up to +0.06 m/s decade⁻¹. This unique pattern represents a composite result of systematic East Asian Winter Monsoon weakening interacting with local topographically forced dynamics.

(3)

Long-term wind speed trends are strongly modulated by seasonal and diurnal cycles. On seasonal scales, the stilling trend is most pronounced during the cold season (with coastal rates of −0.06 to −0.12 m/s decade⁻¹) and monsoon transition periods. On diurnal scales, trend signals are clearest during afternoon hours when sea breezes peak and at night when background synoptic flow dominates, revealing a complex interplay between large-scale climate change and local cyclic phenomena.

(4)

The frequency of extreme wind events shows complex and asynchronous evolution. Widespread strong wind days have significantly decreased during the cold season (−0.214 days decade⁻¹) but show a slight increasing trend in the warm season (+0.097 days decade⁻¹). Furthermore, widespread weak wind days also exhibit subtle decreasing trends in the cold season (−0.021 days decade⁻¹) while remaining stable in the warm season, suggesting nuanced changes in regional atmospheric stability.

(5)

The spatiotemporal evolution of the wind field has direct implications for regional wind resources and engineering design. Risk associated with 100-year return period extreme winds has undergone geographical redistribution, weakening along eastern coasts (by −0.2 to −0.5 m/s) but strengthening significantly in mountainous interiors (by +0.4 to +0.7 m/s). Concurrently, sea-land breezes, critical local climate resources, show decreasing trends in coastal regions where they are most frequent, with reductions of up to 3–7 days decade⁻¹ in some areas.