The Hainan Province has a total land area (primarily including Hainan Island, Xisha Islands, Zhongsha Islands, and Nansha Islands) of 35,400 km², with a sea area of approximately 2,000,000 km². Hainan Island is China’s special economic zone and pilot free trade zone, covering approximately 34,000 km². It is the southernmost provincial administrative region in China. In Hainan Island, the terrain around the edge is low and flat, but in the middle, it rises high to form dome mountains. Wuzhishan and Yinggeling are located in the center of the uplift. Hainan Island has a tropical oceanic monsoon climate, indicating that it is warm and hot throughout the year with abundant rainfall. From 2003 to 2018, the forestland of Hainan Island was primarily concentrated in the south, and the eastern and northern areas, such as Haikou City, Chengmai County, and Danzhou City. The coastal areas of Hainan Island were mostly non-forest land and construction land (Figure 2).

In this study, 393 forest permanent sample plots were selected from China’s National Forest Continuous Inventory data in 2003, 2008, 2013, and 2018 (Figure 3). Forest permanent sample plots are square, with an area of 666.67 m² and a length and width of 25.82 m. China’s National Forest Continuous Inventory database comprises sub-populations, sample plots number, dominant tree species groups code, average age, latitude, longitude, altitude, slope direction, slope position, slope gradient, the thickness of the overburden soil layer, tree species groups name, DBH of each period, and tree volume of each period ( Zeng et al., 2015 ). The monthly climate datasets of Hainan Island from 2003 to 2018 are provided by the basic ground, and automatic weather stations of 19 national weather stations and meteorological networks in Hainan Island. The daily observational data of 19 national meteorological stations in Hainan Island from 2003 to 2013 comprised of average atmospheric pressure, average maximum temperature, average minimum temperature, sunshine hours, monthly sunshine percentage, average water pressure, average relative humidity, maximum daily precipitation, maximum wind speed, and wind direction of maximum wind speed.

There are approximately 4,800–5,800 species of vascular plants on Hainan Island, among which 397 are endemic. Endemic species are primarily distributed in the southwest region of Hainan Island, followed by the southeast (Zhu et al., 2021). The 393 forest permanent sample plots of China’s National Forest Continuous Inventory in Hainan Island were primarily distributed in the southwest and southeast of Hainan Island. In combination with “Hainan Flora,” 11 major tree species groups were screened out. The remaining trees were divided into other hard broadleaf trees and other soft broadleaf trees (Table 1). Moreover, combined with the meteorological data of the past 20 years, complex information such as the site environment and geographical position, is divided into growth geospatial environmental indicators (e.g., longitude, latitude, altitude, and temperature) and local and regional environmental indicators (e.g., slope gradient, slope direction, slope position, and soil thickness). For the cluster analysis, the DBH growth model of the tropical forest in Hainan Island tree species groups is as follows: Δ Y t + Δ t ( j ) = A j ⋅ ( Y t ( j ) ) 2 ⋅ e − B j ⋅ Y t ( j ) ⋅ e λ G E I ( m ) ⋅ X G E I ( m ) ⋅ e λ L E I ( n ) ⋅ X L E I ( n ) . (1)

In Model 1, j is 13^th major tree species groups; Y t ( j ) is DBH (mm); Δ Y t + Δ t ( j ) is the 5-year growth of DBH (mm); A j is the growth rate of the major tree species groups; B j is the growth acceleration rate of the major tree species groups; and λ G E I ( m ) is the growth geospatial environment influence indicator, consisting of λ L , λ B , λ H , λ T M I N , λ T M A X , and λ R . λ L E I ( n ) is the environmental index of the growing local area, comprising λ α , λ β , λ γ and λ h . The 10 environmental influence indicators are shown in Table 2.

When different feature vectors come together, data with small absolute values, such as the annual average minimum temperature, are vulnerable to data with large absolute values, such as rainfall. We then need to normalize the extracted feature vectors so that the normalized value of the feature vector is between 0–1 to ensure that each feature vector is treated equally by the classifier. The normalization of a feature vector is shown in Table 2: 1) from north to south, Hainan Island stretches from Mulan Bay (northern latitude: 20°9 “32”) to Jinmu Corner (northern latitude: 18°9 “21”); 2) from west to east, it stretches from Beibu Gulf (eastern latitude: 108°37 “15”) to Tonggu Corner (eastern latitude: 111°3 “6”); 3) Wuzhishan is the highest mountain on Hainan Island, with a peak altitude of 1,867.1 m, and the lowest altitude of Hainan Island is 0 m; 4) over approximately 20 years, the lowest annual average rainfall was in Southwest Hainan Island (Changjiang Li Autonomous County, Dongfang City, Ledong Li Autonomous County, Sanya City) from 2009 to 2013 and was 1,045.34 mm. The maximum annual average rainfall was in Central Hainan Island (Baoting Li Miao Autonomous County, Qiongzhong Li Miao Autonomous County, Tunchang County, Wuzhishan City) from 2009 to 2013 and was 2,552.08 mm; 5) the lowest annual average minimum temperature was in Central Hainan Island from 2004 to 2008 and was 19.873°C. The highest annual average minimum temperature was in Southwest Hainan Island from 2014 to 2018 and was 23.360°C; 6) the lowest annual average maximum temperature was in Northeast Hainan Island (Chengmai County, Dingan County, Haikou City, Wenchang City) from 2009 to 2013 and was 28.027°C. The highest annual average maximum temperature was in Northwest Hainan Island (Baisha Li Autonomous County, Danzhou City, Lingao County) from 2014 to 2018 and was 29.432°C; 7) the slope gradient was between 0° and 60°; 8) the slope direction was divided into 0°, 45°, 90°,..., 345°; 9) the slope position was divided into the upper slope position (1), middle slope position (0.625), and lower slope position (0); and 10) according to the forest permanent sample plot data used in this study, the thickness of the overburden soil layer was between 30 and 300 cm ( Qiu et al., 2020 ).

The data of more than 45,000 trees from 393 forest permanent sample plots were randomly divided into five groups to ensure that each group contained all of the tree species information. Four groups of data were fitted with a DBH growth model for the major tree species groups in Hainan Island’s tropical forest. The other group was used for accuracy verification. To verify the accuracy of the model in predicting the DBH growth of the major tree species groups over 5 years, a set of reserved data was used. Bias, relative bias (bias%), root mean square error (RMSE), and relative root mean square error (RRMSE) were calculated (Qiu et al., 2018a; Qiu et al., 2018b), and the prediction of DBH growth was evaluated comprehensively. B i a s = 1 n ∑ i = 1 n e i = 1 n ∑ i = 1 n ( y i − y r i ) , (2) B i a s % = B i a s y r ¯ × 100 % , (3) RMSE = ∑ ( y r − y ri ) 2 n , (4) R R M S E = R M S E y r ¯ × 100 % . (5)

In these models, y i is the i^th estimation, y r i is the ith reference, y r ¯ is the mean of the reference values, and n is the number of estimations.

Tropical forest carbon storage measurement primarily includes height curve model, forest volume model, forest biomass conversion model, and forest carbon content. The parameters of the models are summarized according to our previous results on the forest ecosystem carbon storage in China (Qiu et al., 2020).

DBH measurements are typically fast, convenient, and accurate; however, tree height measurements are time-consuming and laborious. Therefore, in the forest survey, tree height was only measured for some dominant trees. It is typically predicted using the tree height curve model (Sharma and Parton, 2007) for different tree species groups, as follows: H = a j d b j . (6)

In Model 6, H is the tree height (m); d is the DBH (cm); and a j and b j are the tree height curve model parameters ( Qiu et al., 2020 ), as shown in Supplementary Appendix Table SA1.

The estimates of increasing forest volume can also serve as a basis for the estimates of aboveground forest biomass and carbon storage. This forest biomass and carbon storage data have gradually become the basis of international treaties as forest volume can be considered as the accumulation of tree volume ( McRoberts et al., 2013 ). Therefore, volume calculation for tree species groups is the key to this investigation. Currently, most countries in the world use two variable volume equation tables as the basic tree volume tables. In recent years, the two variable volume equation table is widely used in China as a tree volume model for 56 major tree species (35 needles and 21 broad leaves), as compiled by the Chinese Agriculture and Forestry Department. The details are summed into tropical forest tree volume model parameters, as shown in Supplementary Appendix Table SA2. The forest volume M (m³/ha) and forest volume growth Δ M (m³/ha·5 years) of the tropical forest are shown as follows: M = ∑ 1 j c j ⋅ d ¯ j g j ⋅ H ¯ j f j ⋅ N ⋅ k j , (7) Δ M ≈ M ⋅ ( g j ⋅ Δ d ¯ j d ¯ j + f j ⋅ Δ H ¯ j H ¯ j ) . (8)

In Model 7 and Model 8, j represents tree species groups; c j , g j , and f j are accumulation parameters ( Qiu et al., 2020 ); d ¯ j is the average chest diameter of tree species groups j (cm); H ¯ j is average tree height of tree species j (m); N is a forest permanent sample plot forest density (plant/ha); k j is the proportion of tree group j to tree species groups; Δ d ¯ j is the DBH growth of an tree group j (cm); and Δ H ¯ j is the height of an tree group j (m).

Biomass expansion factor (BEF) can be defined as a fixed ratio of forest biomass and forest volume, and multiple reaction monitoring (MRM) is used to estimate the forest carbon storage in different areas. If forest biomass is calculated based on forest inventory data, BEF between forest biomass and volume must be established (Fang et al., 2001). Therefore, the forest biomass (Mg/ha) and forest biomass growth (Mg/ha·5 years) of the tropical forests are shown as follows: B = p j M + q j , (9) Δ B = p j Δ M . (10)

In Model 9 and Model 10, p j and q j are the forest biomass conversion parameters between the forest biomass and volume ( Qiu et al., 2020 ), as shown in Supplementary Appendix Table SA3. Furthermore, the average carbon content of each type of tree forest is 51.09%, which is between 46.75%–54.89% ( Liang et al., 2010 ). In this study, the average carbon content was used to calculate forest carbon storage.

The model-fitting R ² values of Other Pine (1), Casuarina, and Eucalyptus were 0.482, 0.431, and 0.603; the model-fitting R ² of other tree species groups was between 0.844 and 0.957 (Figure 4). The model-fitting results are shown in Table 3. The growth rate parameters of the major tree species groups ranged from 0.003 to 0.054, and the growth acceleration parameters ranged from 0.010 to 0.035. The growth rate and acceleration parameters for the different tree species groups had significant differences. The influence parameter of the latitude was 0.098, which indicates that from Jinmu Corner in the southernmost part of Hainan Island to Mulan Bay in the northernmost part, the higher the latitude, the better the trees grow. The influence coefficient of the longitude was 0.790, which indicates that the longitude can promote tree growth from Western Beibu Gulf to Eastern Tonggu Corner. The annual average minimum temperature was −0.276, which can inhibit tree growth. The annual average maximum temperature was 0.373, which can promote tree growth. The influence parameter of the annual average rainfall was 0.103, which indicates that the annual average rainfall can promote tree growth. The influence parameter of the slope gradient was −0.043, indicating that the slope has an inhibitory effect on tree growth. A parameter value of 0.057 for slope direction indicated that trees on the north slope grew better than those on the south slope. The influence parameter of the slope position was 0.013, which indicates that tree growth on the upper slope is better than that on the lower slope. The influence parameter of soil thickness is −0.083, which indicates that the soil thickness in the range of 15–100 mm can promote tree growth.

The results (Figure 4) showed that the DBH growth prediction’s bias ranged from -0.46 to 0.07 cm, RMSE ranged from 1.50 to 5.29 cm, the bias% ranged from −2.96% to 0.55%, and RRMSE ranged from 12.18% to 34.30%. The predicted values were evenly distributed on both sides of the reference values, indicating that the DBH growth prediction accuracy of major tree species groups was good.

In this study, the “interpolation” function of the “spatial analyst tool” in ArcMap 10.2 was selected, and the “inverse distance weight method” was used to deal with the geospatial environmental influence indicators’ change of the tropical forest in Hainan Island. The geospatial environmental influence indicators for tropical forest growth of Hainan Island were subdivided into a 0.05° × 0.05° grid. The growth pattern of the tropical forest in Hainan Island is shown in Figure 5. From the west to the east of Hainan Island, the growth of geospatial environmental influence indicators of the tropical forest increased gradually. The southwestern part of Hainan Island, such as Dongfang City, Ledong Li Autonomous County, Sanya City, and Changjiang Li Autonomous County had lower geospatial environmental indexes, ranging from 1.0014 to 1.6708. The geospatial environmental index of Baisha Li Autonomous County, Wuzhishan City, Baoting Li, Miao Autonomous County, and Danzhou City ranged from 1.6708 to 1.9433. The geospatial environmental index of Lingshui Li Autonomous County, Qiongzhong Li and Miao Autonomous County, and Lingao County ranged between 1.774 and 1.9433. The cities in the eastern part of Hainan Island, such as Wenchang, Qionghai, Wanning, and Tunchang counties, had higher geospatial environmental indexes, ranging from 1.9433 to 2.5120. Northern cities such as Chengmai County, Ding’an County, and Haikou City had a low to high geospatial environmental index ranging from 1.3272 to 2.0381.

In this study, without considering the total consumption of forest stands, the carbon storage of tropical forests in Hainan Island was statistically predicted according to the National Forest Management Plan (2016–2050) (Figure 6). From 2003 to 2008, tropical forest area decreased slightly, while from 2008 to 2013 it increased slightly. From 2013 to 2018, the tropical forest area significantly increased by 7.66 × 10⁴ ha and is expected to increase by 6.45 × 10⁴ ha in the next 30 years. From 2003 to 2018, the volume of the forest increased slowly, and the forest carbon storage increased slowly after a slight decrease from 2003 to 2008. The tropical forest volume increased by 2,403.11 × 10⁴ m³, and the forest carbon storage increased by 13.07 TgC in these 15 years. It is estimated that in the next 30 years, the tropical forest volume will increase by 23,188.33 × 10⁴ m³, and the forest carbon storage will increase by 106.68 TgC; the forest volume per unit area and forest carbon density increased slowly before 2018, and then increased rapidly after a slight decrease in 2018. It is estimated that in next 30 years, the tropical forest volume per unit area will increase by 119.26 m³/ha changing to an increase by 392.34 m³/ha; the forest carbon density will increase by 69.60 Mg/ha changing to an increase by 194.08 Mg/ha. The annual forest carbon sink was −0.06 TgC/yr from 2003 to 2008. Due to large-scale afforestation, there was a slight increase to 1.78 TgC/yr from 2008 to 2013 and slightly increased to 0.89 TgC/yr from 2013 to 2018. In the next 30 years, the forest carbon sink is expected to grow annually and then will tend to be stable.

Longitudes and latitudes are used as the impact indicators of the regional microclimate in each plot. Hainan Island is located in the Northern hemisphere. Thus, the higher the latitude in the Northern hemisphere, the shorter the day. In areas with higher latitudes, photosynthesis slows down, and tree growth cycles increase, resulting in more organic matter accumulation. Therefore, tropical forest trees grow best in high latitudes. Due to the longitudinal zonality of Hainan Island, the water content in the east to west direction was significantly different. Therefore, the higher the longitude, the better the tropical forest grows. From the perspective of topography, the smaller the slope, the more aboveground biomass (De Castilho et al., 2010), the better the plant root system (Stokes et al., 2009), and consequently, the better the tree growth. The north slope was shady with less evaporation, and the soil moisture content was better there than on the south slope. Hainan Island is located in the tropics; it has a diverse mountain terrain, a large number of microclimates (Jiang et al., 2016), and uneven precipitation, and these important factors limit tree growth. Therefore, the tree growth status on the north slope was better than that on the south slope. The tropical forest in the downhill area was heavily deforested and replaced with economically important tree crops such as rubber; however, the Areca and Cassava will remove the nutrients from the area for tree growth (Wang et al., 2007). Therefore, the growth status of the tree in the uphill area was better than that in the downhill area. From a meteorological and climatic point of view, rainfall through the canopy layer could increase the elements in the water and soil nutrients (Chen et al., 2020). Therefore, the greater the annual average rainfall, the better the tropical forest growth status. Tree growth requires specific temperature conditions, and they generally grow in areas where the average monthly temperatures exceed 6°C (Körner, 1998). In the Hainan tropical forest, the annual average minimum temperature (19.873°C), and the annual average maximum temperature (29.432°C) over the past 20 years were not found to limit tree growth. On the contrary, large temperature differences could improve the photosynthetic efficiency of the tree. Therefore, the trees in the tropical forest could grow better with lower annual average minimum air temperatures and higher annual average maximum air temperatures. The overburden soil layer in Hainan Island was thin, and the differences between the soil layer thicknesses were small, which could explain why the relationship between the tree growth status and soil layer thickness was not evident.

Notably, the higher the altitude on Hainan Island, the better the growth of the tropical forest, as this was different from previous studies (Yuliya et al., 2006; David and Robert, 2007). In general, the vertical growth of the mountain tree was hump curved, and a combination of water and heat was the best at the middle altitudes. This phenomenon can be explained by the water–energy balance hypothesis (O'Brien, 2006). However, most of the altitudes of Hainan Island were below 1,200 m, which did not reach the middle altitude for the mid-domain effect (Syfert et al., 2018). In addition, in the tropical and subtropical mountainous areas, tree lines only appeared in the places where the summer isotherm was as low as 3–6°C (Körner, 1998). In the tropical forest of Hainan Island without tree lines, the vertical zonality of the tree growth was not obvious. On the contrary, the species diversity at higher altitudes of the tropical forest was abundant. Therefore, the tree growth status of the tropical forest in the higher altitude areas of Hainan Island was better.

According to the Ninth Inventory of Forest Resources in Hainan (2003–2018), this study used the annual increase in DBH to accurately estimate the forest carbon sink potential and follow two assumptions. First, this study assumes that the forest area data of the Ninth Inventory of Forest Resources in Hainan represent the distribution of the tropical forest area in Hainan in 2003, 2008, 2013, and 2018. Moreover, there will be no large-scale deforestation and death in the next 30 years. Secondly, this study assumes that the area proportion of existing man-made forests can approximately reflect the area proportion of newly built forests in the future. According to the proportion of existing man-made forests among various forest types of the Hainan tropical forest in the Ninth Inventory of Forest Resources, the total area of new man-made forests in the future will be allocated to various forest types of the Hainan tropical forest in proportion. After deducting the total deforestation of forest stands, total mortality of forest stands and total consumption of forest stands, we calculate the total area of tropical forests in Hainan Island in the next 30 years. The total consumption of forest stands deducted is shown in Table 4. These data include the impact of manual management measures on forest biomass. Therefore, the prediction results given in this study consider the impact of human factors and historical processes on the forest biomass carbon pool in Hainan Island to a certain extent. The prediction results can truly reflect the forest carbon pool and its change.

In addition, the factors affecting the accuracy of the prediction results in this study mainly include the following aspects: First, in the next 30 years, it is assumed that there will be no large-scale deforestation and tropical forest death. According to the model, tropical forest biomass will increase naturally. However, some forests are still dead or cut down in the process of growth. Young growth forests with low biomass density replace mature forests with high biomass density, which will make the estimation of tropical forest carbon storage too large. Assuming that the current mortality ratio of forest stands, and the deforestation of forest stands are maintained for the next 30 years (Table 4), according to the relationship between the current tropical forest volume and carbon storage, the tropical forest in Hainan Island will lose about 24.63 TgC in the next 30 years. Then, the forest carbon storage in 2050 will be reduced from 169.96 TgC to 145.33 TgC. Second, various influencing factors in the future may also reduce the accuracy of newly increased man-made forest estimation. With the change of policies on Hainan Island, human activities, trade, and other factors may also affect the area proportion of newly built forests. Third, factors such as climate and environmental change, natural disasters and CO₂ concentration may also affect the accumulation process of forest biomass density in the future. Fourth, whether China’s forestry sustainable development strategic objectives can be achieved will directly affect the prediction results of this study. Fifth, the method of calculating the tropical forest carbon storage using the Chinese forest average carbon content will cause deviations. Therefore, it is urgent to establish tropical forest carbon content models of different tree species.

Based on the existing defects of the prediction results’ accuracy, we will improve the research from the following two aspects in the future. First, our forest DBH growth prediction model only introduces the relationship between DBH and environmental information. In the future, we want to introduce the relationship between DBH, tree age, and environmental information into the models to predict the DBH change more accurately. Then, we can estimate the forest carbon sequestration potential generated by tree growth more accurately. Second, our existing models have limitations. Our models put forward the aforementioned assumptions under the condition that there will be no extreme climate in the future. However, the future climate is unpredictable. Extreme climate may occur in the future. Therefore, we intend to introduce CMIP6 into the future models. CMIP6 refers to the monthly values of minimum temperature, maximum temperature, and precipitation that were processed for nine global climate models (GCMs): BCC-CSM2-MR, CNRM-CM6-1, CNRM-ESM2-1, CanESM5, GFDL-ESM4, IPSL-CM6A-LR, MIROC-ES2L, MIROC6, MRI-ESM2-0, and for four Shared Socio-economic Pathways (SSPs): 126, 245, 370, and 585. In this way, we can assess the relationship between climate, man-made, trade openness, and forest carbon sequestration potential in different future scenarios more clearly.

The growth process of the tropical forest trees is primarily through respiration and photosynthesis to fix carbon dioxide, which is linked to tree growth. Therefore, accurately predicting the growth process of tropical forests to predict their forest carbon sequestration potential is of huge importance. The forest carbon sequestration potential was different for different forest ecosystems. Compared with temperate forests, the forest carbon sequestration potential of tropical forests was more effective ( Terakunpisut et al., 2007 ). In the case of considering the total consumption of forest stands, the change trend of the tropical forest carbon sequestration potential in Hainan Island from 2003 to 2050 is shown in Figure 7. In 2020, the area of China’s forest was 1.75 × 10⁸ ha, and the forest carbon sink generated by the growth of China’s forest in the next 30 years is estimated to be 4,667.87 TgC ( Qiu et al., 2020 ). The tropical forest area in Hainan Island only accounted for 0.88% of China’s forest area. However, in the next 30 years, the forest carbon sink generated by tree growth in Hainan Island's tropical forest will account for 1.8% of China’s forest carbon sink. Therefore, Hainan Island's tropical forest has huge forest carbon sequestration potential in the next 30 years. From 2020 to 2050, the CO₂ emissions from fossil fuel combustion in China are conservatively estimated to be 91.43 PgC, and China’s forest vegetation (tree, economic, shrub, and bamboo forests) will absorb 22.14% of the CO₂ emission from fossil fuel combustion ( Qiu et al., 2020 ). The growth of tropical forests in Hainan Island will absorb 0.34% of China’s CO₂ emissions. Therefore, although Hainan Island’s tropical forest area is small, its contribution to the absorption of CO₂ emissions is huge. It is roughly estimated that in the next 30 years, the total carbon sink of the tropical forest in Hainan Island will be 83.59 TgC.