Tropical cyclone risk for global ecosystems in a changing climate
Coastal ecosystems provide a range of services including erosion prevention, clean water provision and carbon sequestration. With climate change, the rapid change in frequency and intensity of tropical cyclones may alter the composition of the ecosystems themselves potentially degrading the services they provide.
Here we classify global ecoregions into dependent, resilient and vulnerable and show that a combined 9.4% of the surface of all terrestrial ecosystems is susceptible to transformation due to cyclone pattern changes between 1980–2017 and 2015–2050 under climate scenario SSP5-8.5 using the STORM model.
Even for the most resilient ecosystems already experiencing winds >60 m s−1 regularly, the average interval between two storms is projected to decrease from 19 to 12 years which is potentially close to their recovery time. Our study advocates for a shift in the consideration of the tropical cyclone impact from immediate damage to effects on long-term natural recovery cycles.
Species composing ecosystems have co-evolved with the geoclimatic conditions at their location. Hence, they are not only adapted to the average climatic conditions, such as the temperature or precipitation regimes, but also to extreme events such as floods, wildfires or tropical cyclones. Climate change is rapidly modifying the climatic conditions and this has been shown to lead to a shift in suitable conditions for a large number of species among insects, vertebrates and plants1. Shifting abiotic conditions may alter the compositions of larger ecosystems and lead to profound modifications of their functioning. Complementing shifts in means, climate change is rapidly modifying the regional and temporal patterns of extreme events2, which might further impact global ecosystems. A change in frequency and intensity of extreme weather events might even cause more rapid and durable shifts in ecosystems than the gradual increase of the means of climate values3,4,5,6.
Climate change is expected to modify the intensity and frequency of tropical cyclones and their geographical impacts in non-trivial patterns7,8,9. Tropical cyclones are intense low-pressure systems that form over warm ocean waters which can produce strong winds, heavy rainfall and storm surges. Together with the possible ensuing flooding and landslides, they can cause wide-scale disturbances for the ecosystem in the path of the storm and beyond10,11.
Tropical cyclones may affect all dimensions, including the biogeochemistry, hydrography, mobile biota, sedentary biota or microbiota of ecosystems5,12. As the climate warms, the surface waters of the oceans become warmer, providing more fuel for tropical storms to grow in strength13. Further, the amount of moisture in the atmosphere is expected to increase, which can lead to more intense rainfall during tropical storms, and the rising sea level will expose new areas to storm surges14. Overall, the frequency of high-intensity storms is expected to increase, while low-intensity storms are expected to decrease8. Large regional deviations are however expected, such as an overall frequency increase in the northeast Pacific8. Beyond the changes in cyclone severity, changes in regional patterns are likely to expose areas that have not historically experienced these types of storms8,15,16, yielding unprecedented consequences for not only human societies9,17 but also natural ecosystems12,18,19.
Tropical cyclones have been shown to impact ecosystems through various processes, such as wind throw, causing defoliation and canopy damage, which can lead to reduced photosynthesis, growth and productivity of trees, as well as increased susceptibility to pest and disease outbreaks18,20. Heavy rain can cause landslides, soil erosion and nutrient loss21 potentially causing decreased growth and productivity of vegetation, but also direct vegetation mortality22, especially when already disturbed by anthropogenic land use23. Storms can add large amounts of saltwater to coastal freshwater and terrestrial ecosystems disrupting habitats24.
Together, disturbances from tropical storms can alter the structure and composition of ecosystems, change the habitat quality for animal species and lead to mortality or displacement of biota25,26, causing local loss of biodiversity27,28. For instance, extreme weather events account for 11% of global mangrove forest loss between 2000 and 202029.
Previous methods, such as remote sensing and field studies have sought to quantify the immediate impact of tropical storms on forests30, but not how they may modify ecosystems over longer periods. Several studies on the effect of storms on ecosystems have been conducted, most focused on explicit relation for specific species31, part of the vegetation19,32 or for regional ecosystems12. Estimating the potential effect of tropical cyclones on global ecosystems with a long-term perspective is limited by methodological challenges19,33,34,35,36,37 and the absence of systematic measurements of ecological response across ecosystems38.
Ecosystems adapt to abiotic conditions over a timescale from decades to centuries and most ecosystems are expected to be in equilibrium with pre-anthropocene climate39. The statistical historical exposure of ecosystems to tropical storms can be quantified using modelled wind fields, which offer an indirect measure of how much they can accommodate disturbance (Fig. 1). Depending on the historical frequency of tropical cyclones and the potential adaptation of ecosystems to these disturbances, ecosystems can be classified as resilient if historically high levels of disturbance primed them to be able to recover quickly from frequent and intense tropical storms.
In contrast, ecosystems rarely exposed to tropical cyclones in the past are likely to have difficulty recovering from increased disturbances and can be classified as vulnerable. For some ecosystems, the disturbance regime is such that it is an integral force in structuring the ecosystem that is then dependent on this force40,41. While tropical cyclones may appear as purely destructive extreme events, they generate regular disturbances that naturally structure the dynamics of ecosystems42.
The impacts of tropical cyclones are immediately followed by recovery and therefore those ecosystems are in a state of dynamic equilibrium12,18,19,33,43. Furthermore, tropical cyclones may support the regulation of certain ecological processes such as water provision, soil enrichment with nutrients, restoration of ecological niches, support seed dispersion or temporarily mix the salinity and phytoplankton content in estuaries44. Thus, tropical cyclones can also play a functional role, shaping specific ecosystems33,45,46 and being a necessary perturbation to maintain their functionality, thereby supporting biodiversity following the intermediate disturbance hypothesis.
Coastal ecosystems are known to provide a large number of essential services to human societies such as nutrient cycling, erosion control, hatchery for fishes and crustaceans, water purification, cultural importance, tourism, carbon sequestration and circularly, protection from cyclones52,53,54. Thus, it is timely to assess the risk of tropical cyclones under climate change to ecosystems55 and integrate this knowledge into the socioeconomic risk assessment methodologies established in the context of the Intergovernmental Panel on Climate Change (IPCC)2 and Sendai disaster risk reduction56 frameworks. We propose here using the open-source CLIMADA risk modelling platform57 to investigate the vulnerability of global coastal ecosystems to tropical cyclones at a global scale, inspired also by the panarchy principles58. More precisely, using the STORM model datasets STORM-B and STORM-C15,59 we will study the spatial frequency distributions of the storms under current climatic conditions (1980–2017) to then derive where climate change under the shared socioeconomic pathway SSP5-8.5 in 2015–2050 might put ecosystems at risk of modification.
Tropical cyclone impact on ecosystems
For modelling the tropical cyclone disturbance, we use the STORM (synthetic tropical cyclone generation model) synthetic probabilistic set of tracks generated statistically using as baseline the historical records from 1980 to 201759. Furthermore, we use the ecoregions introduced by ref. 60 as the definition of global ecosystems (or bioregions) for which tropical cyclone disturbance is assessed. As shown in Fig. 2, tropical cyclones are regionally confined phenomena. Hence, ecosystems described on the level of regionally defined ecoregions (indicated by polygon shapes), which combine both biological composition and geographic information, are well suited to study the relation with tropical cyclones. In contrast, ecosystems defined on the level of biomes (indicated in colours in Fig. 2) would have too large geographical extents across continents. For instance, the biome tropical and subtropical moist broadleaf forests can be found in the Amazon and Congo basins, which are storm-free, as well as in the Yucatan Peninsula in Mexico and the Philippines, which experience some of the highest tropical cyclone activity worldwide. For each cyclone track, we derive a corresponding wind field using the parametric Holland 2008 model as implemented in CLIMADA61,62, which is then used as a measure of their severity.
Since it is expected that the impact at different intensities differs vastly, we separate wind speeds into three categories: low (L), middle (M) and high (H) intensity storms (Fig. 2; Methods). For each grid point on land and each storm category, we derive the expected annual frequency as shown in Extended Data Fig. 1. We find that 12.3% of the area of all terrestrial coastal ecosystems experience a tropical cyclone of at least low intensity with a frequency of 1 in 250 years over their territory on average. This is the basis for determining the vulnerability of ecosystems to tropical cyclones.
Impact categories
We classify ecoregions as being vulnerable, resilient or dependent63 with regards to tropical cyclones for each of the three intensity categories (L, M and H) (Fig. 1), inspired also by the panarchy principles58 and ecosystems dynamics model assumptions33. A vulnerable ecosystem has rarely been exposed to the disturbance, and the damages of tropical cyclones are high since it has probably not evolved resilience and recovery mechanisms (at most 20% of the area is affected with a probability of 1 in 20 years). In contrast, a resilient ecosystem has been exposed to tropical cyclone disturbance historically frequently (>1 in 20 years) on a large portion of its area (>20%) and thus has evolved to cope with the destructive effects and recovers on timescales faster than the historical disturbance frequency. We classify ecosystems as dependent when the disturbance is naturally present at high frequency (>1 in 20 years) over most of its geographical extent (>80%) and thus its natural ecological dynamics may depend on these regular disturbances64. We thus implicitly assume that inland areas are relevant for the recovery of the coastal areas if they belong to the same ecological system of interlinked species27,65,66. On the basis of these definitions and using the tropical cyclone frequency maps (Extended Data Fig. 1), we determine the vulnerability of all ecoregions as shown in Fig. 3. The exact parametrization is summarized in Fig.
1 and described in the section ‘Vulnerability of ecosystems’. While there are clear distinctions in the frequency, intensity and area of exposure to tropical cyclones among ecoregions (Extended Data Fig. 2a,b), the distributions are continuous and the response of ecoregions close to the threshold is less certain. To account for this uncertainty, we run a global uncertainty and sensitivity analysis67,68,69 (Supplementary Information) considering variation in the position of all thresholds for classifying ecoregions into dependent, vulnerable and resilient (Supplementary Table 1). We find that the selection of the thresholds does result in shifting resilience and dependence for certain ecoregions (Extended Data Figs. 3a and 4a), but is not the dominant factor for the uncertainty in the risk under climate change
Using this classification (summarized in Fig. 1), we find that resilient ecoregions cover an area of 2.5 million km2, the dependent ecoregions cover 0.8 million km2 and the vulnerable ones affected by cyclones cover 16.5 million km2. Together, the resilient and dependent ecosystems represent 2.3% of the surface of all terrestrial ecosystems and 17% of those affected by storms at least once in 250 years. Of the considered 844 ecoregions worldwide—of which 290 are affected by tropical cyclones—200 are affected and vulnerable. Moreover, 26 ecoregions were classified as resilient and 64 as dependent. We remark that all ecoregions resilient or dependent on a higher-intensity storm, turn out to be dependent on the corresponding lower-intensity storms (that is, are affected more often than 1 in 20 years over 80% of their area).
Most of the resilient and dependent ecoregions are concentrated in the larger East Asian regions, the Caribbeans and the Central American peninsula, the Pacific islands and Madagascar as shown in Fig. 3. Among those, we found that only five (Ogasawara subtropical moist forests, Luzon montane rain forests, Luzon rain forests, Mariana’s tropical dry forests and Nansei Islands subtropical evergreen forests) are fully adapted to high (H) intensity storms (winds >58 m s−1). For the resilient ecosystems, the average return period of tropical cyclones is 13 (L), 18 (M) and 19 (H) years, for dependent ecosystems 9 (L), 11 (M) and 17 (H) years and for vulnerable (and affected) ecosystems 87 (L), 102 (M) and 117 (H) years. These return periods can be interpreted as average upper ends to the recovery time of the ecosystems. A detailed list of all vulnerabilities, including the average return period, is given in Supplementary Table 2, and the vulnerabilities per category are shown in Extended Data Fig. 5.
Projected effects of climate change
We compare the regime of tropical cyclones between the STORM-B baseline 1980–2017 and the STORM-C future period 2015–2050 under the high-emission scenario SSP5-8.5. While there remain large uncertainties, SSP5-8.5 probably represents a higher bound on climate change for the near future and may be seen as a probable scenario given the ongoing emission trajectories (compare ref. 70). For the future climate, we consider all four global climate models (GCMs) from the STORM-C dataset15. We always report results corresponding to the median return period at each grid point and show the results for the individual GCMs in the Extended Data Fig. 7. For each ecoregion, we compare the change in the tropical cyclone frequency for each intensity category separately. We then compare the present status of ecoregions classified as dependent, vulnerable or resilient and identify ecoregions that will suffer a substantial increase in the frequency of tropical cyclones (over at least 20% of their area, increase above 10% (20%) for vulnerable (resilient) ecoregions) or a substantial decrease (at least 5% over 80% of the area for dependent ecoregions). The thresholds are summarized in Fig. 1. With this approach, we assume that a change in frequency can lead to a change in ecosystem composition as the dynamical processes required for recovery are no longer synchronized with the disturbance regime5,20,27,42,58. If the average frequency of tropical cyclones in a given region increases substantially, the system would in turn be disturbed in a way that does not allow a return to the initial state as illustrated in Fig. 4. For dependent regions, a substantial decrease changes the composition of the ecosystem as the vital disturbance is reduced.
We find that 194 ecoregions see a substantial increase in frequency, 5 are confronted with a substantial decrease in frequency and 13 are newly affected. While the substantial decrease in frequency is confined to the Caribbeans and the larger Antilles, the increase is found all over the world (Fig. 5): in Oceania, Madagascar and the East African coast, southeast and northwest North America, the Caribbeans, northern South America, the Pacific islands, East Asia, Australia and the Philippines. Interesting regional pattern also arise, such as in the Caribbeans where parts are at risk from a decrease in low-intensity storms, while other parts are at risk from an increase in high-intensity storms (Extended Data Fig. 6). Overall, the combined surface of all ecosystems at risk from tropical cyclones in a changing climate amounts to 9.4% of the combined area of all land ecosystems worldwide. Of these, the newly affected ones contribute 0.9% of the total area. A detailed list of all values for all ecoregions is given in Supplementary Table 3.
On average, ecoregions at risk from tropical cyclones in a changing climate are subject to a 55%, 99% and 149% increase in frequency for the low-, middle- and high-intensity storms, respectively. This results for instance in a reduction of the average time for recovery of resilient systems to high-intensity storms from 19 to 12 years (Table 1). Regarding their geographic distribution, for the high-intensity storms, only ecoregions around China, Japan, Korea and the Philippines are projected to be at risk from a substantial frequency increase. Newly affected and at-risk regions are found in Madagascar, some Pacific islands and in the regions of China, Japan and Korea. For middle-intensity storms, the same regions are potentially affected. Furthermore, regions in Oceania are newly affected, while the Pacific islands and the Antilles are at risk from an increase in frequency. For low-intensity storms, large areas are newly affected or subject to a substantial increase in frequency. Moreover, few dependent ecoregions in the Caribbeans are confronted with a substantial decrease in frequency, which is consistent with findings in ref. 15. Detailed maps for low-, middle- and high-intensity storms are shown in Extended Data Fig. 6. Overall maps for all four GCMs are shown in Extended Data Fig. 7.
The identification of at-risk ecoregions depends nonlinearly on the combination of chosen parameters (Fig. 1) for the classification into resilient, dependent and vulnerable, as well as on the risk threshold levels and the GCMs. Thus, to test the model we performed a full global uncertainty and sensitivity analysis varying all input parameters in reasonable ranges as described in the Supplementary Information and shown in Extended Data Figs. 3b, 4b and 8. We find that the number of regions at risk varies more for low-intensity storms than for high-intensity storms. In the most optimistic choice of input parameters, we still identify 29, 55 and 134 ecoregions at risk from low-, middle- and high-intensity storms, respectively. The number of ecoregions at risk is most sensitive to the choice of risk threshold and the choice of GCM. Out of the four GCMs, three (CMCC-CM2-VHR4, CNRM-CM6-1-HR and HadGEM3-GC31-HM) generally agree on the geographical distribution (Extended Data Fig. 8). The model EC-Earth3P-HR instead shows a large number of ecoregions facing a substantial decrease in frequency, as well as an increase in the Indian Ocean15. The number of newly affected ecoregions by low-, middle- and high-intensity storms in optimistic case are 9, 10 and 9, respectively, and these values are most sensitive to the affected area and the affected frequency thresholds.