Long-Term Hourly Temperature Dynamics on Tropical Hainan Island (1940–2022)

Abstract

With global warming, tropical islands, as sensitive areas to climate change, exhibit new and significant temperature variation characteristics. Using the high-resolution Hainan Island Regional Reanalysis (HNR) dataset and multi-source data, this study analyzes temperature changes on Hainan Island from 1900 to 2022, focusing on spatiotemporal trends, diurnal patterns, and probability distribution shifts. The findings reveal significant periodic temperature changes: weak warming (0.02–0.08 °C/decade) from 1900 to 1949, a temperature hiatus from 1950 to 1979, and accelerated warming (0.14–0.28 °C/decade) from 1979 to 2022. Coastal plains (0.11 °C/decade) warm faster than inland mountains (0.08 °C/decade), reflecting oceanic and topographic effects. Diurnal temperature variations show topographic dependence, with a maximum range (8–9 °C) in the north during the warm season, and a southwest–northeast gradient in the cold season. Probability density function analysis indicates that the curves for transitional and cold seasons show a noticeable widening and rightward shift, reflecting the increasing frequency of extreme temperature events under the trend of temperature rise. The study also finds that the occurrence time of daily maximum temperature over coastal plains is advancing (−0.05 to −0.1 h/decade). This study fills gaps in understanding tropical island climate responses under global warming and provides new insights into temperature changes over Hainan Island.

1. Introduction

Climate change has emerged as a global challenge, with escalating impacts on ecosystems, human health, and socioeconomic systems worldwide [1,2,3]. As these effects intensify, there is an urgent need for a more nuanced understanding of climate dynamics at fine temporal resolutions, particularly in regions with unique geographical sensitivities such as tropical islands [4]. The Fifth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) has highlighted the observed changes in extreme weather and climatic events since 1950, emphasizing their varying impacts across different regions and seasons [5].

Global observations indicate a clear warming trend, accompanied by an increasing frequency and intensity of extreme high-temperature events [6,7,8]. This trend is consistent across various continents, from North America to Asia, underscoring the pervasive nature of temperature anomalies and their profound implications for regional climate patterns [9,10,11,12]. The spatiotemporal distribution of these temperature trends reflects significant influences from large-scale atmospheric circulations, as demonstrated by diverse studies across regions [13,14].

A comprehensive review of climate events has revealed emerging patterns of earlier and more simultaneous occurrences of heat extremes worldwide [15]. This changing landscape of climate extremes necessitates a shift towards more granular climate analyses, particularly at hourly scales, which have been historically underexplored [16]. Fine-scale investigations can provide new insights into diurnal temperature variations and their cumulative effects on tropical island climates, which are inherently different due to their oceanic influences and smaller land masses [17].

Although numerous studies have examined temperature trends and extreme temperature events at daily or monthly scales, there remains an urgent need for finer temporal resolution analysis, particularly in tropical island environments. Due to the high heat capacity of oceans, these regions face unique climate change risks, such as sea-level rise and changes in tropical cyclone activity [4]. Analyzing long-term hourly temperature data offers several advantages, including a more detailed examination of extreme temperature events and the study of changes in the timing of daily maximum and minimum temperatures, which have significant ecological and agricultural implications and are more aligned with human perceptions of temperature changes.

Recent advancements in climate modeling and data assimilation techniques have made it possible to analyze long-term hourly temperature trends. Regional climate models (RCMs) have been particularly useful in capturing small-scale meteorological features that interact with complex terrain or land cover, playing a considerable role in determining climate and weather at regional and local scales [18]. These high-resolution models can provide valuable insights into the diurnal cycle of temperature and its long-term changes, which are crucial for understanding the full spectrum of climate variability and change.

Hainan Island, located in the southernmost part of China’s tropical region, serves as an ideal case study for investigating fine-scale temperature dynamics. Its unique geographical location and climatic characteristics result in significant regional variations in climate change patterns, yet understanding of this spatial variability remains insufficient. Previous studies on Hainan Island have focused on temperature trends, extreme temperature events, and the relationship between interannual temperature variations and sea surface temperatures. However, most analyses have been based on daily or monthly data, lacking in-depth insights into temperature dynamics at finer temporal scales [19,20,21].

To address these knowledge gaps, this study aims to analyze multi-scale temperature trend changes in Hainan Island using hourly scale temperature dynamics from 1940 to 2022 in the Hainan reanalysis data. The research will focus on the following: 1. Analyzing long-term hourly temperature trends across different seasons and diurnal periods. 2. Identifying changes in the frequency, intensity, and duration of extreme temperature events. 3. Exploring spatial variability in temperature trends across the island. 4. Revealing patterns in the timing of maximum and minimum temperatures. This multi-scale, multi-faceted analysis attempts to provide a more comprehensive understanding of the complexity and vulnerability of tropical island climate systems. By utilizing high-resolution hourly data over an extended period, this study aims to contribute valuable insights into the nuanced temperature dynamics of tropical islands in the context of global climate change.

2. Methodology and Datasets

2.1. Study Area

Hainan Island, located at the southernmost tip of China, spans from 18°10′ to 20°10′ N latitude and 108°37′ to 111°03′ E longitude. As China’s only tropical island province, Hainan is surrounded by sea and characterized by a tropical monsoon climate. This climate is distinguished by year-round high temperatures, indistinct seasons, but a clear demarcation between dry and rainy seasons.

The topography of Hainan Island exhibits a distinct concentric layered structure, with elevated central regions and lower surrounding areas, forming a unique dome-shaped mountainous terrain. Centered around Wuzhi Mountain and Yinggeling, the terrain gradually descends outward, creating a diverse landscape of mountains, hills, and plains with a clear tiered structure. This complex topography significantly influences local climate patterns.

In Figure 1, this study uses elevation as the main feature, combined with multi-year average temperatures, and applies the k-means clustering method to divide Hainan Island into three primary climate zones: plains (0–99 m), hills (100–499 m), and mountains (500 m and above). The plains are mainly distributed around the periphery of the island, especially in the northern and eastern coastal areas, where the oceanic influence is more pronounced, and the climate characteristics may be closer to a typical tropical maritime climate. The hilly region is mainly located in the central and eastern parts of the island, forming a transitional zone, with climate features likely falling between those of the plains and mountains. The mountainous areas are concentrated in the south-central part of the island, and due to their higher elevation, they may exhibit distinct mountain climate characteristics, such as a decrease in temperature with altitude and potentially more precipitation.

This topographic diversity, ranging from coastal plains to central mountains, creates a complex climate system within Hainan Island. The interaction between the island’s topography and the prevailing monsoon circulation likely results in significant local variations in temperature and precipitation patterns. Understanding these spatial variations is crucial for comprehensive climate studies and effective regional climate modeling in Hainan Island.

2.2. Datasets

The datasets presented in

Table 1. Overview of climate data resources used: reanalysis, observational, and gridded datasets.

Category	Datasets	Temporal Coverage	Spatial Resolution (°)	Temporal Resolution	Organization
Atmospheric Reanalysis	CERA-20C	1901–2010	2.0	6 h	ECMWF, EU
	NOAA-20CR	1900–2015	2.0	6 h	NOAA, US
	NCEP/NCAR R1	1948–2022	2.5	6 h	NOAA, US
	NCEP/DOE R2	1979–2022	2.5	6 h	NOAA, US
	JRA-55	1979–2017	0.50	6 h	JMA, JP
	MERRA-2	1980–2021	0.50 × 0.313	1 h	NASA, US
	ERA5	1940–2022	0.25	1 h	ECMWF, EU
Land Surface process Reanalysis	GLDAS	1948–2021	0.25	3 h	NASA, US
Extrapolated gridded observation	CRU TS	1901–2022	0.50	Monthly	UEA, UK
	UDEL	1900–2022	0.50	Monthly	UDEL, US
	APHRODITE	1961–2015	0.25	Daily	UT, JP
	CN05	1961–2022	0.25	Daily	NCC, CN
	CMFD	1979–2018	0.10	3 h	ITP, CN
Regional Reanalysis	HNR	1940–2022	0.05	1 h	HNU, CN

Notes: The University of Tsukuba (UT) and the Institute of Tibetan Plateau Research (ITP) are abbreviated as UT and ITP, respectively; The abbreviation for Hainan University is HNU.

can be broadly categorized into four types: atmospheric reanalysis, land surface process reanalysis, interpolated station observations, and regional reanalysis. Each category offers unique advantages and applications in climate research.

Atmospheric reanalysis products, such as CERA-20C, NOAA-20CR, NCEP/NCAR R1, JRA-55, MERRA-2, and ERA5, integrate observational data with numerical model output to provide high-resolution global coverage. These datasets are particularly suitable for large-scale climate studies and long-term trend analyses. ERA5, the latest generation reanalysis, offers hourly temporal resolution and approximately 0.25° spatial resolution, providing enhanced support for regional climate investigations.

Land surface process reanalysis, exemplified by GLDAS, focuses on terrestrial processes by assimilating satellite and ground-based observations. This approach yields critical information on surface temperature, soil moisture, and related variables. Such datasets are invaluable for short-term climate prediction and hydrological modeling.

Extrapolated station datasets, including CRU TS, UDEL, APHRODITE, and CN05, are derived from spatial interpolation of observational station data. CN05, specifically designed for China, achieves 0.25° spatial resolution with daily temporal resolution, potentially offering improved representation of regional climate characteristics. However, these datasets may exhibit increased uncertainty in regions with sparse observational networks.

The Hainan Regional Reanalysis (HNR) represents a state-of-the-art, high-resolution dataset tailored for the Hainan region. Utilizing the Weather Research and Forecasting (WRF) model, HNR performs dynamical downscaling of ERA5 global reanalysis data. The system assimilates daily temperature, relative humidity, and wind speed data from 19 surface stations in Hainan, spanning 1951–2022, thus effectively incorporating local observational information.

HNR’s configuration includes daily 36 h simulations initialized at 00 UTC, employing four-dimensional data assimilation. The final product comprises the latter 24 h of each simulation. The model domain consists of a 200 × 200 × 40 grid with 5 km horizontal resolution, a model top at 5 hPa, and is centered at 19.15° N, 109.75° E. The integration time step is set to 30 s, with simulations covering the period 1940–2022, providing an extensive dataset for long-term climate change studies.

The physical parameterization schemes in HNR are carefully selected for optimal performance in tropical regions. These include the Thompson microphysics scheme, the Kain-Fritsch cumulus parameterization, and the YSU boundary layer scheme. Additionally, the model accounts for subgrid-scale topographic effects (topo_wind) and gravity wave drag to better represent the influence of complex terrain on atmospheric circulation.

A distinguishing feature of HNR is its treatment of land use data. From 1993 to 2022, annual updates of high-resolution land use data from the European Space Agency (ESA) are incorporated, reclassified according to the U.S. Geological Survey (USGS) classification system for compatibility with the WRF model. For the historical period 1940–1992, the original USGS classification is employed. This dynamic approach to land use data enhances the accuracy of simulations by reflecting the evolving landscape of Hainan Island over recent decades. The updated land-use data has been verified to more accurately capture the temperature characteristics and their variations in urban areas.

The high resolution and advanced parameterization schemes of HNR enable improved capture of the effects of Hainan’s complex topography and local circulations on climate. This dataset not only provides crucial support for studying the intricate climate system of Hainan Island but also enhances our capacity to understand and predict climate change and extreme weather events in the region.

The primary objective of this study is to conduct a multi-scale and multi-dimensional analysis of the spatiotemporal variations in temperature on Hainan Island from 1940 to 2022 based on the HNR dataset, complemented by cross-validation using additional mainstream datasets. Because several of the datasets employed for cross-validation (e.g., CRU, CERA-20C, NOAA20C) extend back to 1900, the analysis period is further expanded to include this earlier interval. These longer time series provide valuable information for examining the early characteristics of temperature variability on Hainan Island, offer a more comprehensive long-term perspective, and help enhance the robustness of subsequent analyses derived from the HNR dataset.

2.3. Methodology

In this study, the correlation coefficient (CC) is used to measure the spatial correlation between simulated and observed data. A CC value closer to 1 indicates better simulation performance. Additionally, root mean square error (RMSE) is employed to quantify the deviation between simulated results and observations. A smaller RMSE value suggests that the simulated results are closer to the observed values.

To further evaluate model performance, mean square error (MSE), which is the square of RMSE, is calculated. MSE better reflects the absolute magnitude of errors and is more sensitive to larger errors. For trend analysis, the Theil-Sen trend regression method is adopted. This method is more robust compared to ordinary least squares regression and is insensitive to outliers, making it particularly suitable for long-term trend analysis of climate data.

Given the differences in spatial resolution and temporal scale among the datasets, ensuring their consistency and comparability was a critical step in the analytical process. First, to address the inconsistency in spatial resolution, the low-resolution datasets (e.g., the CRU dataset) were resampled to a 5 km spatial resolution to match the high-resolution HNR dataset, using bilinear interpolation. This procedure ensured uniformity across all datasets at the spatial scale and prevented analytical biases arising from resolution discrepancies.

Second, to address differences in temporal scales, the datasets used in the line charts for multi-dataset comparisons and in the tables presenting temperature change rates over different periods were uniformly converted to annual mean values, thereby ensuring temporal consistency across datasets. In other parts of the analysis where only the HNR dataset was required, hourly data were used, and the corresponding calculations were performed according to the specific analytical needs.

For model performance evaluation, we selected corresponding HNR grid points based on the latitude and longitude of the national meteorological stations on Hainan Island. Using these data, we computed the correlation coefficient, p-value, RMSE, and MSE. Finally, the evaluation results for all stations were averaged to obtain an overall performance assessment.

For the treatment of different seasons, we classified the data based on the average temperature characteristics of different months on Hainan Island. The warm season is defined as April to October, the cold season as December to February of the following year, transition season A as March, and transition season B as November.

3. Results

3.1. Evaluation of HN Reanalysis Datasets

Based on the evaluation results of HNR reanalysis data against 19 ground-based meteorological stations in Hainan from 1951 to 2022 (

Table 2. Evaluation of HNR at multiple temporal Scale.

Scale	Time	CC	RMSE (°C)	MSE (°C)
Year	-	0.82 *	0.31	0.10
Month	-	0.99 *	0.48	0.23
Day	-	0.98 *	0.81	0.65
Warm season	All day	0.85 *	0.27	0.07
	00:00–06:00	0.83 *	0.50	0.25
	06:00–12:00	0.72 *	0.54	0.29
	12:00–18:00	0.91 *	0.26	0.07
	18:00–00:00	0.76 *	0.61	0.38
Cold season	All day	0.90 *	0.60	0.36
	00:00–06:00	0.85 *	0.73	0.54
	06:00–12:00	0.92 *	0.41	0.17
	12:00–18:00	0.86 *	0.56	0.31
	18:00–00:00	0.81 *	1.09	1.18

Notes: CC = Correlation Coefficient; RMSE = Root Mean Square Error; MSE = Mean Square Error. Significance: * p < 0.001. Time periods are given in Coordinated Universal Time (UTC).

), HNR demonstrates strong correlations across various temporal scales. The correlation coefficients (CCs) for annual, monthly, and daily scales reach 0.82, 0.99, and 0.98, respectively, all statistically significant (p < 0.001). This indicates HNR’s capability to effectively capture long-term temperature trends and seasonal variations. The warm season performance slightly outperforms the cold season, with root-mean-square errors (RMSEs) of 0.27 °C and 0.60 °C, respectively. During the warm season, HNR performs best from 1200 to 1800 UTC, with a CC of 0.91 and an RMSE of only 0.26 °C. This may be attributed to the WRF model’s proficiency in simulating daytime convective processes and land–sea thermal contrasts in Hainan Island. For the cold season, the 0600 to 1200 UTC period shows the best performance, with a CC of 0.92 and an RMSE of 0.41 °C.

Overall, HNR’s CCs exceed 0.7 across all temporal scales, often surpassing 0.8, indicating strong correlations with observational data. RMSEs are mostly confined within 1 °C, suggesting high temperature simulation accuracy and demonstrating HNR reanalysis data’s good applicability in the Hainan region. Despite generally good performance, HNR exhibits larger errors in certain periods. For instance, during the cold season, the 1800 to 0000 UTC (0200 to 0800 local time) period shows an RMSE of 1.09 °C and a mean square error (MSE) of 1.18 °C². This may result from challenges in the WRF model’s simulation of nocturnal boundary layer processes, particularly in complex terrain. The greater uncertainty in cold season temperature variations might be associated with the interannual variability of large-scale circulation systems (e.g., the East Asian winter monsoon). Based on this, for a more detailed evaluation of the Hainan regional reanalysis data (HNR) at the annual, monthly, and daily scales, in comparison with observational data from 19 different meteorological stations on Hainan Island, please refer to

Table A1. Evaluation of HNR in Hainan Island from 1951 to 2022.

	CC				RMSE				MAE
Station Name	Year	Season	Month	Day	Year	Season	Month	Day	Year	Season	Month	Day
Qiongshan	0.89 ***	0.95 ***	0.99 ***	0.97 ***	0.47	0.34	0.72	1.10	0.22	0.11	0.53	1.21
Haikou	0.76 ***	0.89 ***	0.99 ***	0.97 ***	0.49	0.54	0.77	1.14	0.24	0.29	0.59	1.31
Dongfang	0.58 ***	0.91 ***	0.99 ***	0.97 ***	0.60	0.61	0.65	1.08	0.36	0.37	0.43	1.16
Lingao	0.90 ***	0.96 ***	0.99 ***	0.98 ***	0.34	0.34	0.64	1.05	0.06	0.12	0.41	1.10
Chengmai	0.54 ***	0.97 ***	0.99 ***	0.98 ***	0.72	0.34	0.48	0.96	0.52	0.11	0.19	0.91
Danzhou	0.55 ***	0.91 ***	0.99 ***	0.97 ***	0.56	0.48	0.63	1.08	0.34	0.23	0.40	1.16
Changjiang	0.91 ***	0.96 ***	0.99 ***	0.97 ***	0.70	1.17	0.86	1.26	0.49	1.38	0.74	1.58
Baisha	0.50 ***	0.95 ***	0.99 ***	0.97 ***	0.89	0.63	0.56	1.11	0.78	0.40	0.31	1.22
Qiongzhong	0.67 ***	0.94 ***	0.99 ***	0.97 ***	0.68	0.97	0.69	1.10	0.47	0.95	0.48	1.21
Ding’an	0.89 ***	0.97 ***	0.99 ***	0.98 ***	0.27	0.41	0.44	0.95	0.07	0.17	0.20	0.91
Tunchang	0.60 ***	0.96 ***	0.99 ***	0.98 ***	0.80	0.75	0.57	1.02	0.65	0.57	0.32	1.05
Qionghai	0.29 *	0.93 ***	0.99 ***	0.98 ***	1.02	0.51	0.44	0.95	1.03	0.26	0.19	0.91
Wenchang	0.82 ***	0.97 ***	0.99 ***	0.98 ***	0.29	0.26	0.34	0.85	0.09	0.07	0.12	0.72
Ledong	0.96 ***	0.93 ***	0.99 ***	0.96 ***	0.47	0.80	0.59	1.02	0.22	0.63	0.34	1.04
Wuzhishan	0.73 ***	0.96 ***	0.99 ***	0.96 ***	0.80	1.00	0.78	1.14	0.63	0.90	0.61	1.31
Baoting	0.88 ***	0.96 ***	0.99 ***	0.95 ***	0.37	0.95	0.60	1.08	0.14	0.91	0.36	1.17
Sanya	0.24 ***	0.51 ***	0.93 ***	0.92 ***	1.67	2.19	1.70	1.86	2.76	4.81	2.90	3.46
Wanning	0.62 ***	0.96 ***	0.99 ***	0.98 ***	0.79	0.71	0.58	0.98	0.62	0.50	0.34	0.96
Lingshui	0.74 ***	0.96 ***	0.99 ***	0.97 ***	0.47	0.60	0.43	0.88	0.22	0.36	0.19	0.78

Notes: *** p < 0.001, * p < 0.05.

in the Appendix A.

3.2. Multiscale Temperature Trends and Variability

3.2.1. Decadal-Scale Temperature Trends

Figure 2 compares the annual temperature in Hainan Island from 1940 to 2022 using the Hainan reanalysis data (HNR) and other datasets. The long-term historical reanalysis (LTR) shows an annual mean temperature of approximately 24.3 °C during 1900–1950, about 0.2–0.6 °C higher than the gridded interpolated data (GDEP). This overestimation may stem from the limitations of NOAA 20C and ERA CERA 20C/20CM reanalyses, which assimilate only pressure data, potentially failing to fully capture Hainan’s complex tropical maritime climate. LTR data indicate a warming trend of about 0.10 °C/50 yr, reflecting a relatively slow warming in the first half of the 20th century, consistent with globally observed trends [22]. GDEP estimates for the same period are slightly lower, averaging around 24.1 °C, with a 50-year warming trend of about 0.20 °C/50 yr. While GDEP uses GHCN data for interpolation, its 0.5° resolution may be too coarse to accurately reflect the local complex island mountain climate of Hainan. Short-term regional extrapolation products begin in the 1960s, showing better consistency with other datasets after 1980, possibly reflecting the use of common observational data. Regional reanalysis data (RRD) emerges in the late 1940s, with a mean temperature of about 24.5 °C, overestimating in situ observations by approximately 0.7 °C.

HNR data starts from 1940, showing a relatively stable mean temperature of about 23.7 °C during 1950–1980. In situ observations, beginning in 1951, indicate a mean temperature of approximately 24.0 °C for the same period, though early observational data sparsity may affect reliability. Notably, temperature trends during this period are relatively flat. ERA5 and GLDAS show mean temperatures of about 23.5 °C and 23.3 °C, respectively, for 1950–1980, underestimating in situ observations by 0.5 °C and 0.7 °C.

From 2000 onwards, global warming trends become more pronounced, with increasing complexity in dataset differences. In situ observations show a further increase in mean temperature to about 25.0 °C, with a 20-year warming rate of 0.50 °C/20 yr. RRD exhibits significant overestimation during this period, with a mean temperature of about 25.3 °C, 0.3 °C higher than in situ observations, and a higher warming rate of approximately 0.70 °C/20 yr. LTR remains relatively stable, with a mean temperature of about 25.1 °C. ERA5 and GLDAS estimates improve, rising to about 24.7 °C and 24.5 °C, respectively, but still underestimate in situ observations by 0.3 °C and 0.5 °C.

Overall, temperature changes were relatively modest from 1900 to 1950, with slight warming in most datasets from 1950 to 1980, and all datasets showing significant warming trends after 1980. The 20-year warming rate reached 0.50 °C/20 yr during 1980–2000, further accelerating to 0.50–0.70 °C/20 yr after 2000. This long-term warming trend, particularly the accelerated warming after 1980, aligns with globally observed changes. Differences between datasets reflect uncertainties in climate data, which were larger in the early period (1900–1950) and gradually decreased with improvements in observational networks and assimilation techniques. However, even in recent periods, differences of 0.3–0.7 °C persist between datasets.

Analysis of temperature trends in Hainan Island from 1900 to 2022 reveals complex warming patterns (

Table 3. Temperature trends in Hainan Island from multiple datasets across 1900–2022.

Datasets	Periods (°C/Decade)
	1900–1949	1950–1979	1979–2022	All Period
I Long-term Series Extrapolated Gridded Observation Data
1.1 Ensemble (1900–2022)	0.06 ***	−0.10 ***	0.16 ***	0.04 ***
1.2 CRU (1901–2022)	0.08 ***	−0.11 ***	0.15 ***	0.03 ***
1.3 UDEL WM (1900–2022)	0.04 *	−0.10 **	0.000	0.01
II Long-term Series Atmospheric Reanalysis Data
2.1 Ensemble (1900–2010)	0.02	0.04 *	0.20 ***	0.07 ***
2.2 NOAA 20C (1900–2015)	0.03	0.12 ***	0.17 ***	0.14 ***
2.3 ERA 20C (1901–2010)	0.07 ***	0.13 ***	0.20 ***	0.09 ***
2.4 CERA20C (1901–2010)	−0.04 *	−0.06 *	0.15 ***	0.000
2.5 ERA20CM (1901–2010)	0.06 *	−0.01	0.24 ***	0.05 ***
III Short-term Series Atmospheric Reanalysis Data
3.1 Ensemble (1948–2022)	-	−0.04	0.21 **	0.08 ***
3.2 JRA55 (1979–2017)	-	-	0.19 ***	0.19 ***
3.3 MERRA2 (1980–2021)	-	-	0.12 ***	0.12 ***
3.4 NCEP1 (1948–2022)	-	−0.05	0.11 **	0.09 ***
3.5 NCEP2 (1979–2022)	-	-	0.13 ***	0.14 ***
IV Short-term gridded observation
4.1 Ensemble (1961–2022)	-	-	0.16 ***	0.24 ***
4.2 APHRO (1961–2015)	-	-	0.14 ***	0.20 ***
4.3 CN05 (1961–2022) China Region	-	-	0.28 ***	0.27 ***
4.4 CMFD (1979–2018)	-	-	0.11 **	0.12 **
V Hainan Regional Atmospheric Reanalysis
5 surface observation (1940–2022)	-	−0.8	0.14 **	0.09 ***
VI Atmospheric Forcing Data for HNR
6 ERA5 (1979–2022)	-	-	0.10 ***	0.10 ***
VII In situ observation
7 In situ obs (1951–2019)	-	0.42 *	0.22 ***	0.27 ***
VIII In situ observation in differed regions
8.2 CN05.1 South China Region	-	-	0.26 ***	0.21 ***
8.1 CN05.2 (1961–2022) Hainan Island	-	-	0.28 ***	0.27 ***
IX Land Surface Reanalysis
9 GLDAS (1948–2021)	-	−0.15 ***	0.02	0.06 ***

Notes: *** p < 0.001, ** p < 0.01, * p < 0.05.

). Long-term series analyses show weak but significant warming trends, with Ensemble and CRU data indicating warming rates of 0.04 °C/decade and 0.03 °C/decade, respectively (p < 0.001). However, NOAA 20C reanalysis data shows a more pronounced warming trend [0.14 °C/decade, p < 0.001]. For the 1900–1949 period, most datasets show slight warming [0.02–0.08 °C/decade], but ERA CERA20C indicates weak cooling [−0.04 °C/decade, p < 0.05], possibly due to early observational data uncertainties and reanalysis system differences. From 1950 to 1979, several datasets show slight cooling trends, such as HNR [−0.8 °C/decade] and CRU [−0.11 °C/decade, p < 0.001]. This may reflect the “global warming hiatus” phenomenon attributed to increased global aerosol emissions in the 1950s–1970s. However, in situ observations show a strong warming trend [0.42 °C/decade, p < 0.05]. Short-term atmospheric reanalysis data like NCEP1 also indicate a slight cooling trend [−0.05 °C/decade]. The 1979–2022 analysis reveals more significant and consistent warming trends. CN05 data shows the strongest warming [0.28 °C/decade, p < 0.001], while HNR indicates warming of 0.14 °C/decade (p < 0.001). JRA55 and ERA5 show warming trends of 0.19 °C/decade (p < 0.001) and 0.10 °C/decade (p < 0.001), respectively. GLDAS data shows a notably lower warming trend [0.02 °C/decade] compared to other datasets, possibly due to its primary reliance on land surface process simulations rather than direct assimilation of atmospheric observations.

Regionally, Hainan Island’s warming trend is similar to that of South China [0.26 °C/decade, p < 0.001] but lower than China overall [0.28 °C/decade, p < 0.001], potentially reflecting the influence of oceanic modulation. The CN05.1 dataset reveals this spatial difference, suggesting unique climate response characteristics for Hainan as a tropical island.

The Hainan regional reanalysis data (HNR) shows a warming trend [0.14 °C/decade, p < 0.001] for 1979–2022 that closely matches in situ observations. However, CN05 data indicates a notably higher trend [0.28 °C/decade, p < 0.001] than HNR, possibly due to differences in terrain consideration during interpolation and spatial distribution of observation stations. Differences also exist among reanalysis datasets, with NCEP1, NCEP2, and MERRA2 showing lower warming trends [0.12 °C/decade, p < 0.001] than in situ observations for 1979–2022, while JRA55 data indicates a slightly higher trend [0.19 °C/decade, p < 0.001].

This figure presents a comparison of various climate data products, including recent reanalysis datasets (JRA-55, MERRA-2, NCEP-DOE R2, and NCEP/NCAR R1), gridded extrapolated products (UDEL and CRU), short-term regional extrapolation products (CN05, CMFD, and APHRODITE), and long-term historical reanalysis products (NOAA-20CR, ERA-20C, ERA-20CM, and CERA-20C). The shaded areas represent the 25th and 75th percentiles of the multi-year averages for each category of datasets.

3.2.2. Seasonal Temperature Patterns

The transition from the cold season to the warm season (March, Figure 3a) shows a significant warming trend, with HNR data indicating a warming rate of 0.11 °C/decade, while the CN05.1 HN data reveals a higher rate of 0.39 °C/decade. Two distinct cold periods occurred in the early 1960s and mid-1970s, with temperatures dropping to approximately 21.5 °C. In contrast, warm periods emerged in the early 1980s and 2010s, with temperatures approaching 24 °C.

The warm season (April–October, Figure 3b) exhibits relatively stable temperature fluctuations, with HNR data showing a warming rate of 0.10 °C/decade, while in situ observations indicate a rate of 0.14 °C/decade. Between 1940 and 1960, the average temperature of the warm season fluctuated between 26.5 °C and 27 °C. Subsequently, through 2020, all three datasets demonstrate a gradual rise in temperatures, stabilizing around 27.5 °C with slight fluctuations.

The cold season (December–February, Figure 3d) shows the most pronounced temperature changes, with HNR data indicating a warming rate of 0.11 °C/decade, while in situ observations report a significantly higher rate of 0.30 °C/decade. Between 1940 and 1960, the average temperature during the cold season fluctuated between 18 °C and 19 °C. A cold period occurred from the 1960s to the mid-1970s, followed by a marked upward trend in temperatures starting in the late 1970s.

The transition from the warm season to the cold season (November, Figure 3c) shows temperature trends intermediate between those of the warm and cold seasons. HNR data reveals a warming rate of 0.20 °C/decade, while CN05.1 HN data indicates an even higher rate of 0.47 °C/decade. Between 1940 and 1960, November temperatures fluctuated significantly, ranging from 20.5 °C to 22 °C. A notable cold period occurred in the late 1970s, with temperatures dropping to around 20.5 °C. Following this, temperatures showed a general upward trend, albeit with significant interannual fluctuations.

3.2.3. Seasonal Diurnal Temperature Patterns

Figure 4a presents temperature changes during March (the transition period from the cold season to the warm season). The warming trends show notable differences across different time periods. Compared to the early morning (0–5 h, 0.08 °C/decade) and morning (6–11 h, 0.07 °C/decade) periods, the nighttime (18–23 h) and afternoon (12–17 h) periods exhibit higher warming rates of 0.10 °C/decade and 0.10 °C/decade, respectively.

Figure 4b illustrates the temperature trends during the warm season (April–October). Similarly to March, the nighttime period shows the most pronounced warming trend (0.11 °C/decade), followed by the afternoon period (0.11 °C/decade), while the early morning (0.11 °C/decade) and morning (0.09 °C/decade) periods display relatively lower warming rates.

Figure 4c,d present temperature trends for November (the transition period from the warm season to the cold season) and the cold season (December–February), respectively. A comparison reveals that, in all four seasons, the warming trends are most significant during the afternoon and nighttime periods. Specifically, for November, the nighttime warming rate is 0.18 °C/decade and the afternoon warming rate is 0.17 °C/decade. However, during the cold season, the warming trends in all periods are lower compared to other seasons, particularly due to the decrease in average temperatures towards the end of the 2010s. Nonetheless, the nighttime period (0.07 °C/decade) still exhibits a higher warming rate than the other periods within the cold season. This suggests that the warming phenomenon in Hainan Island is more pronounced during the nighttime.

3.3. Spatial Heterogeneity in Temperature Dynamics

3.3.1. Decadal Patterns

Figure 5a illustrates the most pronounced warming trend in the coastal plain region (elevation 0–99 m), with a linear regression equation of y = 0.0086x + 7.18, indicating an annual warming rate of 0.086 °C/decade. The average temperature in this region increased from approximately 23.8 °C in the early 1940s to about 24.8 °C in 2020, resulting in a total warming of approximately 1 °C over 80 years. Notably, two significant temperature peaks occurred in the mid-1950s and around 2020.

Figure 5b depicts a slightly lower warming trend in the hilly and plateau regions (elevation 100–499 m), with an annual warming rate of 0.08 °C/decade. The average temperature in this region rose from about 22.6 °C in the early 1940s to approximately 23.4 °C in 2020, showing a total warming of approximately 0.80 °C over 80 years. The temperature variation curve in this area exhibits similar fluctuations to the coastal plain region, but with smaller amplitudes, possibly reflecting the buffering effect of topography on temperature changes.

Figure 5c shows the warming trend in the central mountainous region (elevation 500 m and above), with an annual warming rate of 0.08 °C/decade, similar to that of the hilly and plateau regions. The average temperature increased from approximately 20.8 °C in the early 1940s to around 21.5 °C in 2020, resulting in a total warming of approximately 0.70 °C over 80 years. Notably, while the warming rate in the central mountainous area is similar to that in other regions, its absolute temperature values are significantly lower, reflecting the substantial impact of elevation on temperature.

Spatially, the contour map in the figure illustrates variations in warming rates across Hainan Island. The southern and northern coastal areas show relatively higher warming rates, approximately 0.13–0.15 °C/decade, while the central and western regions exhibit lower warming rates of about 0.07–0.09 °C/decade.

Figure 6a illustrates a significant warming trend across Hainan Island during the warm season, with marked differences among various topographic regions. The plain areas (outlined in gray) exhibit the most pronounced warming trends, with rates generally ranging between 0.09 and 0.12 °C/decade, especially in the northern and southern coastal regions, where rates reach 0.12 to 0.15 °C/decade. The hilly areas (outlined in red) display relatively lower warming rates, mostly between 0.09 and 0.12 °C/decade. The mountainous areas (outlined in green) show the lowest warming rates, primarily between 0.08 and 0.11 °C/decade, consistent with the influence of elevation on temperature, reflecting the buffering effect of topography on temperature changes. A small region in the southern and northern parts of Hainan Island exhibits the highest warming rates, averaging 0.12 °C/decade, with a peak rate of 0.17 °C/decade. All these warming rates have been confirmed through statistical significance testing. Overall, the warming trend in the warm season is consistent with observations from other regions in China, though the rate is slightly lower than the national average (0.22 °C/decade). This discrepancy may be attributed to Hainan’s unique tropical maritime climate, where oceanic moderation likely mitigates the warming rate.

Figure 6b shows that the warming trend in the cold season is as significant as that in the warm season. The warming rates in plain areas typically range from 0.08 to 0.12 °C/decade, with some regions also reaching 0.14 to 0.17 °C/decade. The warming rates in the hilly areas range from 0.06 to 0.12 °C/decade, while the rates in mountainous areas are relatively lower, mostly between 0.08 and 0.11 °C/decade. All these warming trends have been confirmed through statistical significance testing. The spatial distribution pattern of cold-season warming is generally similar to that of the warm season, with only slight differences. This indicates that in response to climate change, topography has a more significant influence on Hainan Island compared to other factors, though oceanic factors must also be considered in temperature distribution. For example, the northwestern region exhibits the lowest warming trends in both the warm and cold seasons.

3.3.2. Seasonal Patterns

Figure 7a,b illustrate the temperature trends during the early morning hours (00:00–05:00 UTC+8). The warm season shows a pronounced warming trend, with most regions experiencing warming rates of 0.08–0.12 °C/decade. In contrast, the warming during the cold season is weaker, with fewer areas exhibiting warming rates of 0.10 °C/decade or higher; most regions show rates of 0.06 °C/decade. Notably, mountainous areas (outlined in green) exhibit relatively lower warming rates in both seasons. Figure 7g,h depict the temperature variations during the nighttime (18:00–23:00 UTC+8). The warming trends display a generally consistent spatial distribution across both seasons. During warm season nights, most areas maintain warming rates of 0.08–0.12 °C/decade. Mountainous regions consistently show lower warming rates, approximately 0.08–0.10 °C/decade in both seasons. Haikou (20°02′ N, 110°20′ E) and Sanya (18°15′ N, 109°30′ E) stand out as notable high-temperature zones, demonstrating clear urban heat island effects.

Figure 7c,d present the temperature trends during the morning hours (06:00–11:00 UTC+8). Compared to the early morning period, the warming trends in both seasons are more pronounced. In the warm season, eastern and southern coastal areas experience warming rates of 0.14–0.18 °C/decade, with similar rates observed during the cold season. Hilly regions exhibit relatively lower warming rates in both seasons, with rates of 0.08–0.10 °C/decade in the warm season, and slightly lower rates during the cold season. Figure 7e,f show the temperature changes during the afternoon (12:00–17:00 UTC+8). The warming in the warm season afternoons is most significant, particularly in the southern and eastern coastal areas, where warming rates reach 0.16–0.18 °C/decade. In the cold season, afternoon warming is slightly lower, with most areas showing warming rates of 0.14–0.16 °C/decade. Mountainous regions consistently exhibit lower warming rates: 0.10–0.16 °C/decade in the warm season and 0.06–0.14 °C/decade in the cold season. All warming rates mentioned have been subjected to statistical significance testing.

3.4. Temporal Evolution of Temperature Probability Density Functions

Figure 8a shows that the mean temperature in March increased from 22.16 °C in 1940–1978 to 22.99 °C in 1979–2022, with an overall mean of 22.70 °C for the period 1940–2022, resulting in a warming of 0.83 °C. The probability density functions (PDFs) indicate that the distributions for all three periods are approximately Gaussian, with the 1979–2022 curve showing a distinct rightward shift. These results demonstrate a clear warming trend in the mean temperature of March over the past 80 years, which has become particularly pronounced since the 1980s.

In Figure 8b, the mean temperature for April–October increased from 26.59 °C in 1940–1978 to 27.14 °C in 1979–2022, with a long-term average of 26.94 °C for 1940–2022. The warming amplitude of 0.55 °C is smaller than that observed in March. The PDFs are similarly near-normal, but the 1979–2022 curve exhibits a slight rightward shift and is narrower than those of the other periods, indicating that temperature variability during this period has been smaller, with a moderate warming trend over the past 80 years.

Figure 8c shows November as another transitional month, with a more pronounced warming trend than in March. The mean temperature increased from 21.84 °C in 1940–1978 to 22.82 °C in 1979–2022, with a total mean of 22.47 °C for the period 1940–2022. The warming amplitude of nearly 1 °C is the highest among all seasons. The PDF curve exhibits a noticeable rightward shift and widening over time, signifying substantial temperature fluctuations in November during the past 80 years, with a very evident warming trend, and an increased frequency of high-temperature events since the 1980s.

The cold season (December–February) trend, shown in Figure 8d, differs from those of the warm season. The mean temperature increased from 18.87 °C in 1940–1978 to 19.89 °C in 1979–2022, with an overall mean of 19.53 °C for the period 1940–2022. The warming amplitude of nearly 1 °C is comparable to that of November and substantially exceeds the warming rate observed during the warm season. The PDFs also show a significant rightward shift and widening, with a higher peak for the period 1979–2022, indicating an increasing probability of mild winter events and greater temperature variability during the boreal winter.

3.5. Diurnal Temperature Variation: Timing and Magnitude

3.5.1. Diurnal Timing of Maximum and Minimum Temperatures

In Figure 9a,c, the timing of maximum temperature exhibits significant spatiotemporal differences. Maximum temperatures primarily occur in the afternoon, around 13:00–14:00 (Beijing Time) during the warm season and slightly later during the cold season, generally consistent with diurnal solar radiation patterns and seasonal variations. However, several distinct regions deviate from this dominant pattern: in the northeastern coastal Wenchang area (110.8° E), maximum temperatures occur earlier at approximately 12:00–13:00, attributable to its geographical position receiving solar radiation heating first, combined with flat terrain and sparse vegetation cover conducive to rapid warming; during the warm season, the eastern edge of Diaoluoshan extending to Shimei Bay and Boundary Island coastal areas also exhibit 12:00 maximum temperature characteristics, possibly related to windward slope topography accelerating heat accumulation, as steep terrain enhances solar radiation absorption efficiency, especially under clear weather conditions, causing more rapid surface warming than in plain regions. The 12:00 maximum temperature phenomenon observed in the western coastal Qizi Bay of Changjiang (108.5° E, 19.3° N) demonstrates the importance of local microclimates, with three surrounding small mountains creating a unique thermal environment through topographic enclosure effects. Additionally, industrial activities significantly influence maximum temperature timing: the Yangpu area (109.3° E, 19.8° N) during the warm season and Haikou urban center and New Port industrial zone (110.4° E, 20.2° N) during the cold season both exhibit 12:00 maximum temperature characteristics, potentially resulting from anthropogenic heat emissions from industrial activities and altered thermal characteristics due to modified urban underlying surfaces.

As shown in Figure 9b,d, the timing of minimum temperature across Hainan Island displays significant spatiotemporal differences, spanning from 6:00 to 10:00 (Beijing Time). Spatial differentiation during the warm season is primarily controlled by dual regulation of topography and oceanic influences: coastal plain areas, moderated by the ocean’s large heat capacity, experience minimum temperatures typically at 7:00–8:00; while central mountainous regions (such as Jianfengling, Diaoluoshan, and other high-elevation areas) exhibit notably delayed minimum temperatures until 8:00–9:00 due to combined effects of nocturnal inversion layer formation and cold air pooling. Spatial differences during the cold season are more pronounced, forming an east–west division approximately along 109.6° E. Eastern plains and coastal areas, under oceanic thermal regulation and relatively weaker radiative cooling, experience delayed minimum temperatures until 9:00–10:00; while western mountainous and hilly regions, influenced by topographic lifting effects and stronger radiative cooling, experience earlier minimum temperatures at 8:00–9:00. This significant spatiotemporal differentiation can be explained through multiple physical mechanisms: first, complex terrain systems form cold air pools, causing continuous cold air accumulation in low-lying areas, affecting near-surface temperature diurnal variation characteristics; second, cold air invasion pathways demonstrate marked differences between eastern and western regions, with eastern coastal areas benefiting from oceanic thermal buffering that delays minimum temperature occurrence, while western regions directly affected by cold air intrusions without oceanic moderation experience earlier minimum temperatures. From a microclimatological perspective, this complex spatiotemporal distribution pattern not only reflects Hainan Island’s unique geographical location and topographical features but also reveals the integrated mechanisms of local circulation, land–sea thermal differences, and large-scale weather systems across different seasons.

Based on the analysis of Figure 10a,c, changes in the timing of maximum temperature exhibit distinct seasonal differences. During the warm season, most regions show an advancing trend, particularly pronounced in coastal plain areas (−0.03 to −0.08 h/decade), equivalent to an advance of 2–5 min every decade. In contrast, the central mountainous region shows relatively weaker trends, with some localized areas even exhibiting delayed trends (0–0.03 h/decade). The pattern of change during the cold season is more complex, with coastal areas maintaining an advancing trend, but with slightly reduced magnitude (−0.02 to −0.05 h/decade), while the northern coastal areas demonstrate a strong delayed trend (0.03–0.105 h/decade). This delay may be attributed to slower sea surface temperature increases in the northern coastal regions compared to other coastal waters, with lower sea surface temperatures playing a more significant moderating role (the spatial map of annual average sea surface temperature and its multi-year variation trend is shown in Figure A1 of the Appendix A).

Trends in the timing of minimum temperature exhibit more significant spatial heterogeneity and seasonal differences. During the warm season, eastern and northeastern mountainous regions generally show delayed trends (0.015–0.075 h/decade), corresponding to a delay of approximately 1–5 min/decade. Western mountainous areas exhibit relatively weaker trends, with some localized areas even showing advancing trends (−0.03 to −0.09 h/decade), potentially reflecting changes in the dynamics of nocturnal inversion layers in mountainous terrain. Changes during the cold season are more complex, with northern coastal areas showing advancing trends (−0.03 to −0.1 h/decade), while inland and mountainous regions display significant spatial variations. This complex spatiotemporal evolution pattern likely results from multiple interacting factors. Changes in the frequency and intensity of cold air activities also significantly influence diurnal temperature variations, particularly evident in the mountainous regions and leeward coastal and hilly areas of Hainan Island, which exhibit delayed trends (0.03–0.105 h/decade).

3.5.2. Long-Term Changes in Diurnal Temperature Range

Based on the analysis of the spatial distribution characteristics of diurnal temperature range (DTR) in Hainan Island, distinct geographical differentiation patterns are observed during warm and cold seasons. During the warm season (Figure 11a), the highest DTR values (8–9 °C) are mainly concentrated in the north–central region (Nandujiang River Valley area). The central mountainous area, benefiting from the dual influence of vegetation cover and elevation effects, exhibits relatively smaller DTR (approximately 5 °C); while coastal areas (especially the coastal zone bounded by 19° N) under strong oceanic modulation show DTR of only 1–2 °C. The spatial distribution during the cold season displays a pronounced southwest–northeast gradient, with maximum values (approximately 9 °C) occurring in the narrow zone between Jianfengling and the central mountains, possibly related to the interaction between prevailing northeast monsoon winds in winter and topographic lifting effects. During the cold season (Figure 11b), the northeastern coastal areas maintain low DTR levels in both cold and warm seasons, reflecting the persistent influence of oceanic modulation.

4. Discussion

Through analysis of temperature change characteristics in Hainan Island from 1900 to 2022, significant spatiotemporal heterogeneity and periodic features were revealed. During 1900–1949, a slight warming trend (0.02–0.08 °C/decade) was observed, consistent with the global-scale gradual warming phenomenon in the early 20th century [1]. The period of 1950–1979 exhibited temperature fluctuations, with some datasets showing a slight cooling trend (−0.8 °C/decade), corresponding to the global “warming hiatus” potentially influenced by increased global aerosol emissions during the 1950s–1970s [23]. From 1979 to 2022, a more pronounced and sustained warming trend (0.14–0.28 °C/decade) was observed. Regarding the switch in land use data for the HNR dataset after 1993, sensitivity tests have been conducted to rule out the possibility of trend exaggeration. The warming trend during this period is consistent with the overall warming trend in southern China [24,25]. From an atmospheric process perspective, the spatial heterogeneity of temperature changes in Hainan Island is controlled by multiple factors. During the warm season, strong air–sea interactions modulate temperature changes by altering boundary layer structure [26]. Research indicates that cold season warming (0.3 °C/decade) significantly exceeds warm season warming (0.136 °C/decade), reflecting the global winter warming amplification effect [27].

The temperature and temperature changes on Hainan Island exhibit significant spatiotemporal variability. Temporally, this study found that in all four seasons, the warming rate during the night is the highest. This may be due to the increase in atmospheric water vapor, cloud cover, and absorbing aerosols in recent decades, which enhances downward longwave radiation [28], weakening the surface’s effective radiation and reducing nighttime cooling efficiency. As a result, the boundary layer suppresses heat dissipation, leading to a higher warming rate at night than during the day. Spatially, the study found that the warming rate is lower in the central mountainous region, likely due to the dual regulation of topographic uplift and dense vegetation coverage. In contrast, coastal plain areas show higher warming rates, particularly in some regions in the south and north. This is primarily attributed to these areas being part of the two major urban regions in Hainan (the southern and northern cities), indicating the significant impact of the urban heat island effect on temperature changes. However, the reasons for the higher warming rates in the non-urban areas of the southern coastal plain require further investigation, as the contributions of topography and ocean influences to this phenomenon remain unclear. Additionally, the generally lower warming rate in the northwest, as mentioned earlier, is notably influenced by the lower average sea surface temperature and warming rate in the Beibu Gulf, which plays a significant regulatory role (detailed sea surface temperature and its change rates are shown in Appendix A Figure A1).

The daily temperature range (DTR) shows significant spatiotemporal differences. During the warm season, coastal plain areas exhibit a trend toward earlier maximum temperatures [−0.05 to −0.1 h/decade], while the timing of minimum temperatures is significantly delayed, confirming the notable rise in global average temperatures over recent decades. The spatial distribution of DTR shows clear geographical heterogeneity, with Hainan Island’s unique topography and monsoon climate leading to significant seasonal variations in DTR. In the warm season, the northern region exhibits the maximum DTR values (8–9 °C), while during the cold season, the maximum DTR (approximately 9 °C) occurs between Jianfengling and the central mountainous region. The interaction between the northeast monsoon and complex terrain is likely the primary driver of this phenomenon [29,30,31,32,33].

Probability density function (PDF) analysis reveals systematic characteristics of temperature changes. The PDF curves for the transitional months of March and November are wider and exhibit greater fluctuation, whereas the curves for the warm season are narrower. This seasonal differentiation distinguishes Hainan Island from other tropical islands in terms of climate change characteristics [34,35]. The expansion of the PDF distribution indicates an increased probability of extreme temperature events, a common feature observed in global tropical islands [36,37,38].

Overall, this study demonstrates that climate change on Hainan Island is reflected not only in the general rise in temperature but also in the reshaping of diurnal thermal rhythms and the spatial patterns of warming. The use of hourly data reveals systematic shifts in the timing of temperature extremes, indicating that climate warming is altering the “diurnal thermal structure” upon which human activities and natural processes depend. These findings have direct implications for climate adaptation strategies, such as adjusting outdoor work schedules in agriculture and construction.

At the same time, the markedly different warming rates between the coastal plains and the central mountainous regions highlight the strong modulation of regional climate change by local factors such as topography, vegetation, and urbanization. The rapid warming observed in coastal cities reinforces the influence of the urban heat island effect and suggests the need to prioritize urban ventilation planning and green infrastructure. In contrast, the relatively slow warming in the central mountainous regions indicates their potential role as “climate refugia” under increasing climate risks, underscoring the importance of protecting mountain ecosystems.

However, there are some uncertainties in this study. ERA5 reanalysis data shows significant uncertainty during the 1940s to 1950s due to sparse assimilation data [39]. These uncertainties propagate through the model system, ultimately affecting the reliability of the climate data for Hainan Island. In particular, in regions with complex terrain, uncertainties in the downscaling process are more pronounced due to the limited accuracy of early topographical data.

5. Conclusions

This study, utilizing the HNR dataset, investigates the spatiotemporal characteristics of temperature variations in Hainan Island from 1940 to 2022. Furthermore, by incorporating multiple datasets, it analyzes the early characteristics of temperature changes in Hainan Island from 1900 to 1940. The key findings are as follows:

• Temperature changes in Hainan Island during 1900–2022 exhibited significant periodic characteristics: slight warming during 1900–1949 (0.02–0.08 °C/decade), warming hiatus or slight cooling during 1950–1979 (−0.8 °C/decade), and accelerated warming during 1979–2022 (0.14–0.28 °C/decade). The overall warming rate was lower than that of mainland China, reflecting the unique response pattern of tropical island climate to global warming.
• Temperature variations exhibit significant spatial heterogeneity, with coastal plains (0.11 °C/decade) warming faster than mountainous inland regions (0.08 °C/decade). This disparity is primarily influenced by a combination of factors, with the lower warming rate in mountainous areas largely attributed to topographic regulation and high vegetation cover.
• Diurnal temperature range (DTR) variations show distinct topographic dependence. During the warm season, the northern region exhibits the maximum DTR (8–9 °C), while coastal areas display the minimum values (1–2 °C). In the cold season, a significant southwest–northeast gradient is observed, with the maximum values occurring between Jianfengling and the central mountainous regions (approximately 9 °C). This phenomenon is closely related to the interaction between oceanic regulation, the northeast monsoon, and topographic lifting effects.
• Probability density function (PDF) analysis reveals that the PDFs for transitional and cold seasons are wider, with greater fluctuations, and show a distinct rightward skew, most notably in the 1979–2022 period. This indicates that, while the average temperature on Hainan Island is rising, the frequency of extreme temperature events is also expected to increase.
• The timing of maximum and minimum temperatures showed systematic changes. During the warm season, the occurrence of maximum temperature in coastal plains advanced (−0.05 to −0.1 h/decade), while mountainous regions exhibited a delay trend (0–0.03 h/decade). This spatial disparity indicates that topography and land–sea thermal contrast are the primary regulatory mechanisms of diurnal temperature variations.