Spatial and temporal distributions of temperature and rainfall on tropical islands of Vietnam

With an increase in frequency and intensity across the globe, climate change has posed high risks to any region in the world, including the islands. In recent years, along with climate change, sea-level rise, extreme weather events and coastal erosion have also substantially affected biodiversity, vegetation, and economies of the islands, especially small islands, small island developing states (Thomas et al. 2020), tropical and subtropical islands (Singh & Bainsla 2015) as well as Pacific islands (Field et al. 2014). In this regard, small island states located mainly in the tropics and the subtropics share similar features that increase their vulnerability to the presence of climate change (Hay et al. 2001). Besides that, socio-economic activities such as pollution, overfishing, and unsustainable development have also exacerbated the ecosystem degradation in the islands (Mcleod et al. 2019). As a result, studies on the islands in the context of climate change have gained more attention in recent literature.

Past research has examined climate trends on islands in various regions by different methods. Using observation data at 27 tidal measuring points to assess the sea-level change in Western Tropical Pacific islands from 1950 to 2009, Becker confirms that Niño-Southern Oscillation (ENSO) events affected the change of sea level in the area, by ±20 to 30 cm in comparison with average sea level (Becker et al. 2012). By employing RClimDex software and the Mann–Kendall test, another study suggested unceasing warming trends, both at the regional and the global level, using data in Trinidad islands from 1961 to 2010 with temperature and precipitation observation data (Beharry et al. 2014). Taking Puerto Rico as a test case with results from 12 General Circulation Models, Khalyani determines a warming pattern together with a decline in precipitation in tropical islands by projecting a decrease of 130–1,397 mm in precipitation (from 1960–1990 mean to 2071–2099 mean) (Khalyani et al. 2016). Using high-resolution data of monthly and annual rainfall from 1920 to 2012, another study concluded that more than 90% of the state has drying tendencies, especially the western side of the Hawaii island which experiences the highest long-term rainfall decrease in the annual and dry season (Frazier & Giambelluca 2016). Regarding island vulnerability and resilience, climate change and environmental stress had direct effects on communities including demographic and social changes (Ratter 2017). Potential risks to islands were estimated to be high due to tropical cyclones and sea-level rise (Thomas et al. 2020).

To understand historical climate change and forecast future trends, the Global Climate Model (GCM) is one of the most powerful tools to simulate physical, chemical, and biological characteristics of the atmosphere, land, as well as oceans and generate climate projections. GCMs provide useful data to assess future climate changes and were developed by multiple organizations, one of which is the Coupled Model Intercomparison Project (CMIP) with CMIP6 as the latest version. Up until now, CMIP6 has published the results of 23 models for some countries in the world with the latest four emission scenarios including SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5. Studies have used the result data from the CMIP6 to project and assess climate factors for regions over the world such as in South America (Almazroui et al. 2021a; Diaz et al. 2020), the United States, Central America and the Caribbean (Almazroui et al. 2021b), Central and South America (Ortega et al. 2021), Amazon (Parsons 2020), and Southwestern South America (Rivera & Arnould 2020). However, it is notable that CMIP6 models have the tendency to overestimate precipitation based on calculations for Xinjiang during 1995–2014 (Zhang et al. 2022) or for the Tibetan Plateau during 1961–2012 (Cui et al. 2021) and might underestimate average temperature trends (Cui et al. 2021).

According to the World Bank, Vietnam stands among the most vulnerable countries to climate change impacts in the tropical region. Among 4,000 islands with a total area of islands of about 1,600 km², only 100 islands have an area ≥1 km². The remaining islands are small in size and considered to have higher exposure to disaster risks under climate change. Nevertheless, research on climate change in Vietnam faces great challenges due to the sparseness of the monitoring network and the lack of monitoring data which is also a common issue in other countries (Nurse et al. 2014). Hence, previous research only analyzed climate change either for a specific island (An et al. 2014) or at the provincial level (MONRE 2021). Limited literature that focused on coastal zones and islands, on the other hand, only examined temperature elements (air temperature, sea water temperature, sea level and the number of typhoons) up to 2012 (Vu 2017) or projected flood risk due to sea-level rise (MONRE 2021).

To our best knowledge, no study on climate change assessment of the islands in Vietnam has been conducted by using the most recent data and applying the results of the CMIP6 project (WorldClim v2.1). Thus, this study uses observation data up to 2020 at 10 meteorological stations on the islands to evaluate the spatial and temporal distribution as well as the trends of climate change using the Mann–Kendall test. The ACCESS (the Australian Community Climate and Earth System Simulator) is one of 23 models that was developed by Australian research. The use of extracted results from ACCESS-ESM1.5 over 20-year periods (2021–2040, 2041–2060, 2061–2080, and 2081–2100) at 30 s resolution also enables this study to project rainfall and temperature trends for the islands of Vietnam based on the latest version of the global model under four emission scenarios of SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5. Based on the result of this research, Mann–Kendall test can be employed to calculate historical temperature and precipitation trends while future climate trends can be predicted using downscaling method and CMIP6 scenarios. These methods can be applied to similar tropical islands with areas above 1 km² in other countries and regions. In addition, this research also provides a scientific basis to build detailed climate change adaptation scenarios for each island in substitution for large-scale scenarios which are more challenging to implement.

The study area consists of 10 islands in Vietnam as shown in Table 1. Regarding geologic origin, there are volcanic islands (Ly Son, Phu Quy) which are located in the South Central Coast region, and islands of mixed geologic type (Bach Long Vi, Co To, Con Co, Truong Sa Lon, Song Tu Tay, Con Dao, Phu Quoc, and Tho Chu) which are located in the Gulf of Tonkin, and Southern coastal region.

The terrain of islands in Vietnam is characterized by low mountains. The highest peak mountain is Nui Chua on the Phu Quoc island at 547 m. Because of the low terrain, the amount of rainfall on the islands is close to that of the surrounding ocean. The distributions of temperature and rainfall across the islands vary a great deal as they stretch over 12 latitudes. While islands in the Northern region such as Co To and Bach Long Vi often have cold winters, the Southern islands do not.

The landscape of the islands is not favorable to the formation of freshwater streams. Most islands have no surface flows and scarce freshwater, especially in the dry season, except in Con Dao and Phu Quoc islands. Despite being small and short, freshwater in these two islands is preserved due to well-protected forests as well as large evergreen broadleaf forest areas. The islands in the study area are all inhabited, some of which have high population density such as the Ly Son island with approximately 2,200 people/km² or the Phu Quy island with approximately 1,700 people/km². As a result, climate change has a large impact on freshwater supply and sustainable socio-economic development.

This study uses two main data sources. The first one is the daily rainfall and temperature data of 10 meteorological stations provided by the Vietnam Meteorological and Hydrological Administration. The commencement years of monitoring are not the same for all stations, but the monitoring is still ongoing in 10 islands. The list of meteorological stations as well as the observation periods and their locations are shown in Table 1.

The second consists of the historical and future climate data extracted from the ACCESS-ESM1.5 model of WorldClim v2.1 under four emission scenarios of SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5 by ArcGIS 10.5 software for tropical islands of Vietnam.

In this study, the Mann–Kendall test is applied to identify the trend of climate elements (annual rainfall – RA, maximum 1-day rainfall – RX, annual temperature – TA, maximum temperature – TX, and minimum temperature – TN) on 10 islands of Vietnam. The Mann–Kendall non-parametric statistical test method (Kendall 1975), developed by the U.S. Geological Survey (Helsel et al. 2005), has widely been used in the climatic and hydrological sciences to identify trends in time series data. The application of the Mann–Kendall test to assess climate change trends has been carried out in research on river basins of Vietnam such as the Kone river basin (Phan et al. 2010), Ba river basin (Phan 2014) as well as some basins over the world (Kundu et al. 2015).

With a high spatial resolution at 30 s, 2.5 min, 5 min, and 10 min, Worldclim v2.1 produced the data set of historical climate and future climate data. The Worldclim v2.1 built two sets of historical data including data for the period 1970–2000 and monthly data for 1960–2018 including temperature, precipitation, solar radiation, vapor pressure, and wind speed. To ensure the validity and reliability of the projected result of WorldClim v2.1, this study also compared the simulated results of the CMIP6 to the observation data at meteorological stations. Among 10 meteorological stations, only two stations (Co To and Bach Long Vi) have data for the period 1970–2000. The remaining stations have shorter observation periods.

The CMIP6 projections of Worldclimv2.1 are based on the Shared Socio-economic Pathways (SSPs) scenarios with different radiative forcing levels at the top of the atmosphere. The simulated radiative forcing levels around the year 2100 are 2.6, 4.5, 7.0, and 8.5 W m⁻² in SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5 scenarios, respectively. The future climate data available in WorldClim v2.1 program was also CMIP6 downscaled under four scenarios (SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5) in 23 climate models. Compared to CMIP5 models, the forecast results of climate change in CMIP6 models are slightly higher because of the increase in CO₂ (Meehl et al. 2020).

In this study, the results from the ACCESS-ESM1.5 model, which is one of 23 models in Worldclim v2.1, at a resolution of 30 s by ArcGIS 10.5 software for tropical islands of Vietnam are extracted. However, CMIP6 with resolutions from 30 s to 10 min cannot cover all islands because the areas of some islands are smaller than the size of the smallest grid cells (1 km × 1 km). Therefore, it is only possible to get future data for eight of 10 islands in the periods of 2021–2040, 2041–2060, 2061–2080, and 2081–2100. Climate data including the minimum monthly temperature (°C), the maximum monthly temperature (°C), and monthly rainfall (mm) are used to estimate the climate changes in the future for the islands of Vietnam.

Rainfall on islands in Vietnam has a heterogeneous distribution and complex fluctuations. The average annual rainfall ranged from 1,147 mm (Bach Long Vi) to 2,855 mm (Phu Quoc) (Figure 3(c)). Bach Long Vi and Phu Quy stations had an average annual rainfall of less than 1,500 mm. Only the Co To station had an average annual rainfall of 1,500–2,000 mm. Half of the stations had an annual rainfall of 2,000–2,500 mm and only two stations had an average annual rainfall of above 2,500 mm. Nine of 10 islands in the study area are small islands (except for the Phu Quoc island), and surface water resources are limited. Thus, low rainfall is one of the difficulties for economic development.

The islands have a distinct rainy season division which is similar to other areas in the tropical region but with a more complicated distribution. The rainy season lasts from five to 10 months depending on the island. While the rainy season lasts up to 10 months on the Song Tu Tay Island, it only lasts for five months on Bach Long Vi, Con Co, and Ly Son islands (Figure 3(d)). The starting and ending times of the rainy season on all islands are also not uniform. Rainfall volume concentrates mainly in the rainy season months, accounting for more than 90% of the annual rainfall in most islands. The rainy season in the Con Dao island lasts for seven months from May to November with an amount of 1,896mm, accounting for 93.6% of the annual rainfall. The rainy season in Truong Sa Lon, Song Tu Tay, Phu Quy, Phu Quoc, and Tho Chu islands, on the other hand, usually lasts from eight to 10 months with the amount accounting for over 90% of annual rainfall. In other stations, the rainy season lasts only for 5–6 months and accounts for less than 90% of the annual rainfall. The lowest monthly rainfall on the islands accounts for 1–3% of the annual rainfall. The driest month of Con Co, Truong Sa Lon, and Song Tu Tay islands is April while that of the remaining is February. The lowest monthly rainfall is usually below 50mm per month except for Phu Quy and Con Dao islands, the average lowest monthly rainfall is below 10mm. The uneven distribution in both space and time causes issues for water exploitation as well as water use on the islands because most islands of Vietnam are small islands and not suitable for water storage.

An assessment of past and present climate and climate change trends indicates that temperature has increased as much as 0.1 °C per decade in regions with small island states (Hay et al. 2001). In another study (Vu 2017), the trend of average annual temperature increase in the islands was determined with the observation data up to 2012. In this study, the trend of temperature and rainfall on the islands are determined using the observation data up to 2020 by the Mann–Kendall test. The parameters in the Mann–Kendall test such as S (Mann–Kendall test statistic), Tau (Kendall's Tau), and Z (The normalized test statistic) are calculated as described in Equations (1), (2), and (4), respectively.

The calculated results of Tau, S, and Z values of the maximum temperature at Bach Long Vi, Ly Son, Truong Sa Lon, Song Tu Tay, Phu Quy, Con Dao, Phu Quoc, and Tho Chu stations were >0 (Table 2). The p-values of Bach Long Vi, Truong Sa Lon, Song Tu Tay, and Tho Chu stations were <0.05, indicating a significant increasing trend of the maximum temperature. The calculated results of Tau, S, and Z values of the maximum temperature at Co To and Con Co stations were <0 (Table 2). The p-value of the Co To station was <0.05, indicating a significant decreasing trend of the maximum temperature. The increasing trend of the maximum temperature at four stations was identified as 0.59–0.73 °C per decade, the decreasing trend at the Co To station by 0.22 °C per decade.

The calculated results of Tau, S, and Z values of the minimum temperature at Co To, Con Co, Truong Sa Lon, Song Tu Tay, Phu Quy, Con Dao, Phu Quoc, and Tho Chu stations were >0 (Table 2). The significance was identified at Co To, Con Co, Truong Sa Lon, Con Dao, and Phu Quoc stations. On the other hand, the calculated results of Tau, S, and Z values of the minimum temperature at Bach Long Vi and Ly Son stations were <0 and the significance was identified only at the Bach Long Vi station (Table 2). The increasing trend of minimum temperature at five stations was identified by 0.2–0.81 °C per decade, decreasing trend at the Bach Long Vi station by 0.59 °C per decade, and stable at the remaining stations.

The calculated results of Tau, S, and Z values of the maximum 1-day rainfall at Co To, Truong Sa Lon, Phu Quy and Phu Quoc stations were >0 while those at Bach Long Vi, Con Co, Ly Son, Song Tu Tay, Con Dao, and Tho Chu stations were <0 (Table 2). However, only the Co To station had a significant increasing trend of the maximum 1-day rainfall. The increase of 0.81mm in the maximum 1-day rainfall per decade was identified based on the trend equation.

There are only 3/10 and 1/10 stations that have statistically significant temporal trends in annual rainfall and the maximum 1-day rainfall, respectively. On the contrary, there are statistically significant temporal trends in annual, maximum, and minimum temperature at 10/10, 5/10, and 6/10 stations, respectively. Overall, the global warming trend can be identified on the islands in Vietnam. In particular, the maximum temperature in the period 2010–2020 is aligned with the assessment by the World Meteorological Organization in 2020. Therefore, it implies that climate change is very evident on the islands. These results are also aligned with previous literature that there were no statistically significant temporal trends in precipitation (Beharry et al. 2014).

According to the above results, the changes in temperature and rainfall factors on the islands of Vietnam's sea were very complicated. Moreover, Vietnam is one of the five countries that was strongly affected by climate change (Hay et al. 2001). Hence, forecasting the changes in climate factors in the future is deemed important.

Future data in CMIP6 are built for 23 models under four emission levels of SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5 under the periods 2021–2040, 2041–2060, 2061–2080, and 2081–2100. Among 10 meteorological stations on 10 islands of Vietnam, it is not possible to get future data for two islands with areas of less than 1 km², namely Truong Sa Lon and Song Tu Tay, as the highest resolution of the CMIP6 model is 30 s.

In contrast to the maximum monthly temperature, the minimum monthly temperature tends to change in a complex way. The minimum monthly temperature will experience change by 1.9–5.4, 2.0–6.0, −2.0 to 6.3, and −1.8 to 7.4 °C by 2100 in comparison with the baseline period, under the SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5 scenarios, respectively. The increasing trend of the minimum monthly temperature under four emission scenarios occurs on the islands located in the Northern sea of Vietnam. In contrast, the minimum monthly temperature on islands in the Southern sea of Vietnam mostly decreases under all scenarios.

The average annual rainfall will face the change by −20.2 to 47.3, −23.1 to 52.4, 28.4–36.4, and 25.3–40.4% by 2100 in comparison with the baseline period, under the SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5 scenarios, respectively. Ly Son and Phu Quy islands have a decreasing trend in average annual rainfall in comparison with the baseline period. The low average annual rainfall on the Phu Quy island as well as the decreasing trend of rainfall might cause difficulties for water use on the island. Despite being one of the two islands with low annual rainfall, the Bach Long Vi island has the highest increase under all scenarios by 30.6–52.4% in comparison with the baseline period.

Forecast results showed that the maximum and minimum monthly temperature tend to increase until the end of the 21st century. On the contrary, trends of average annual rainfall are inconsistent and vary across islands. Average annual rainfall has a decreasing trend in most islands except for those in the Thailand Gulf, which is Phu Quoc and Tho Chu islands. The seasonal division of rainfall in the scenarios is aligned with observed data. In addition, climate trends calculated based on SSP scenarios in this research are aligned with previous research (Almazroui et al. 2021a; Diaz et al. 2020; Ortega et al. 2021; Parsons 2020; Rivera & Arnould 2020).

To our best knowledge, this is the first research to predict temperature and precipitation change on the tropical islands of Vietnam. Due to data limitation, validation was carried out only for monthly data in the period of 1970–2000. Additionally, the process of downscaling climate models for small islands is another cause of uncertainty. On the other hand, uncertainties in future climate projections are primarily caused by the future climatic forcing scenario, model structure, and intrinsic variability of the climate system (Hawkins & Sutton 2009). Forecast data are calculated on a 20-year basis and are only suitable to build long-term development strategies. Due to the large water demand on the islands of Vietnam, short-term socio-economic development plans or water resource use plans require a more detailed forecast on an annual, monthly or daily basis.

Overall, the distribution of rainfall and temperature on the tropical islands of Vietnam is complex in both space and time as well as both in the past and future. The annual rainfall on the islands of Vietnam ranges from 1,147 to 2,855 mm with a rainy season lasting from five to 10 months depending on each island. The annual temperature on the islands tends to increase from North to South and is usually lower than in the mainland. The amplitude of temperature fluctuation at the stations also tends to decrease from North to South. In particular, only the annual temperature of the observation period tends to increase. The average annual temperature at all monitoring stations tends to increase by 0.22 °C per decade with the range from 0.11 to 0.33 °C. Other climate trends are inconsistent across islands and only significant in meteorological stations that have long-observed data.

A 20-year forecast until the end of 21st century showed that only maximum and minimum monthly temperature have increasing trends. Average annual rainfall does not have a clear pattern but tends to have decreasing trends except for those in the Gulf of Thailand. The seasonal division of rainfall in the scenarios is aligned with observed data. However, this research also poses several limitations. First, reference data could only be used for comparison for a 30-year period (1970–2000). Second, results from climate models are not responsive to two islands with an area of less than 1 km² in this research because of the maximum resolution of CMIP6 at 30 s. Nevertheless, the results from climate models will not be responsive to islands with an area of less than 1 km² because of the maximum resolution of CMIP6 at 30 s. According to the research results, it can be concluded that extracting the results from the ACCESS-ESM1.5 can calculate future data, even for small areas like the tropical islands of Vietnam. The results of this study provide important implications for climate change adaptation scenarios and territorial planning of tropical islands.

The authors would like to thank anonymous reviewers for their helpful comments. The authors also express their gratitude to the project on ‘Senior Researcher Support – Code NVCC10.05/23-23’ in support of this research. We acknowledge the World Climate Research Programme, which, through its Working Group on Coupled Modelling, coordinated and promoted CMIP6. We thank the climate modeling groups for producing and making available their model output, the Earth System Grid Federation (ESGF) for archiving the data and providing access, and the multiple funding agencies that support CMIP6 and ESGF.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.