New elevation data show that by midcentury frequent coastal flooding will rise higher than areas currently home to hundreds of millions of people. As a result of heat-trapping pollution from human activities, rising sea levels could within three decades push chronic floods higher than land currently home to million people.
By , areas now home to million people could fall permanently below the high tide line. The new figures are the result of an improved global elevation dataset produced by Climate Central using machine learning, and revealing that coastal elevations are significantly lower than previously understood across wide areas.
The threat is concentrated in coastal Asia and could have profound economic and political consequences within the lifetimes of people alive today. Findings are documented in a new peer-reviewed paper in the journal Nature Communications. Box 1. Related Resources. As humanity pollutes the atmosphere with greenhouse gases, the planet warms. The consequences range from near-term increases in coastal flooding that can damage infrastructure and crops to the permanent displacement of coastal communities.
Over the course of the twenty-first century, global sea levels are projected to rise between about 2 and 7 feet, and possibly more. The key variables will be how much warming pollution humanity dumps into the atmosphere and how quickly the land-based ice sheets in Greenland and especially Antarctica destabilize. Projecting flood risk involves not only estimating future sea level rise but also comparing it against land elevations.
However, sufficiently accurate elevation data are either unavailable or inaccessible to the public, or prohibitively expensive in most of the world outside the United States, Australia, and parts of Europe.
This clouds understanding of where and when sea level rise could affect coastal communities in the most vulnerable parts of the world.
A new digital elevation model produced by Climate Central helps fill the gap. Based on sea level projections for , land currently home to million people will fall below the elevation of an average annual coastal flood. By , land now home to million people could sit permanently below the high tide line. Adaptive measures such as construction of levees and other defenses or relocation to higher ground could lessen these threats.
In fact, based on CoastalDEM, roughly million people currently live on land below high tide line. This population is almost certainly protected to some degree by existing coastal defenses, which may or may not be adequate for future sea levels. Despite these existing defenses, increasing ocean flooding, permanent submergence, and coastal defense costs are likely to deliver profound humanitarian, economic, and political consequences. This will happen not just in the distant future, but also within the lifetimes of most people alive today.
That factor is coastal elevation. In the absence of coastal defenses such as levees, elevation determines the extent to which ocean floods can wash over the land. Accurately measuring coastal elevation over large areas is neither easy nor cheap. Some countries, such as the United States, use a remote-sensing technology called lidar to reliably map the heights of their coastlines, and publicly release the results.
Lidar is relatively expensive, however, typically requiring plane, helicopter, or drone overflights, as well as laser-based equipment. Although SRTM data are freely available online, they are less reliable than lidar. SRTM data measure the tops of features that protrude from the ground—such as buildings and trees—as well as the ground itself. As a result, SRTM data generally overestimate elevation, particularly in densely forested and built-up areas.
In low-lying parts of coastal Australia, for instance, SRTM data overestimate elevation by an average of 8. Globally, the average overestimate appears to be roughly six feet two meters. These values match or exceed most of the highest sea level rise projections for the entire century.
In coastal regions, overestimates of elevation produce underestimates of future inundation driven by sea level rise. Understanding the real threat posed by future sea level rise requires a better view of the ground beneath our feet. That is the purpose of CoastalDEM. Developed using machine learning working with more than 51 million data samples see methodology , the new dataset is substantially more accurate than SRTM, particularly in densely populated areas—precisely those places where the most people and structures are threatened by rising seas.
CoastalDEM cuts the average error to less than 2. Combining CoastalDEM with sea-level-rise and coastal-flood models produces new estimates of exposure to rising seas around the world box 2. Those estimates reveal that far more land—and more people—will be vulnerable to sea level rise during this century than previously believed chart 1.
Sea level rise is a global story, and it affects every coastal nation. Mainland China, Bangladesh, India, Vietnam, Indonesia, and Thailand are home to the most people on land projected to be below average annual coastal flood levels by table 2. Together, those six nations account for roughly 75 percent of the million people on land facing the same vulnerability at midcentury.
Chart 1. Current population below the elevation of an average annual flood in , top six countries Moderate emissions cuts. A closer look at the cases of mainland China, India, Bangladesh, and Vietnam sheds light on the scope of the problem. Start with mainland China. By , land now home to 93 million people could be lower than the height of the local average annual coastal flood.
Low-lying Jiangsu Province, which abuts Shanghai, is also vulnerable. So are Tianjin, the main port for the capital city of Beijing, and the Pearl River Delta region, an urban agglomeration comprising several major mainland cities and the special administrative regions of Hong Kong and Macau explore map at coastal. The national government has already decentralized many aspects of water management: flood protection is the responsibility of regional water management boards.
Public authorities have also bolstered hard defences including a 3,km network of dikes, dams and seawalls, including the famous Maeslant Barrier. Cities like Rotterdam offer a model for how to manage sea-level rise. Rotterdam is one of the safest delta cities in the world precisely because it has learned to live with water. This attitude can be traced back to the 13th century, when local merchants and city administrators erected a metre dam to keep high waters at bay, but also to facilitate drainage.
New canals were built in the s to improve water quality and reduce epidemic outbreaks of cholera. Several decades after catastrophic floods killed over 1, people in , the Maesland Barrier was constructed. Across the Netherlands, cities like Rotterdam are converting ponds, garages, parks and plazas into part-time reservoirs. Chinese cities are also taking action to mitigate and adapt to sea-level rise.
As in the case of the Netherlands, the Chinese were motivated in part by disaster. In , floods killed roughly 4, people when the Yangtze River basin overflowed. A growing number of big cities such as Beijing — which more than doubled its total land coverage in the last decade — are also suffering a rise in floods.
The Chinese government has responded with a combination of hard engineering, environmental and people-based strategies, together with the relocation of millions of citizens. In , China launched the so-called sponge city initiative. The term actually originated in Hyderabad when the city authorities started collecting storm-water to offset water demand during planting season. More than 30 cities are currently part of the initiative including Shanghai — one of the most flood-prone cities in the world.
The Chinese expect that at least another cities will join in the coming decade. The city is already rocked by two to three typhoons every year. Shanghai is also sinking, albeit less slowly than Jakarta.
To reduce its exposure to rising seas, Shanghai has constructed km of protective seawalls that stretch across the Hangzhou Bay and encircle the islands of Chongming, Hengsha and Changxing. As in the case of Rotterdam, Shanghai has also installed massive mechanical gates to regulate overflowing rivers. South East Asian cities are busily building defences against sea level rise.
For example, Jakarta is building a massive sea wall with Dutch support, and is planning to relocate , people from threatened riverbanks and reservoirs. Critics, however, fear that the city is not doing enough to address groundwater issues that are causing the city to sink. Bangkok, which faces similar challenges to Jakarta, has also laid out a 2,km canal network and central park with a capacity to drain 4 million litres into underground containers.
Some Thai parliamentarians fear that the only way forward is a managed retreat, moving the capital further inland. But arguably the most dramatic responses to rising sea levels are occurring in those parts of the world that are most acutely at risk.
Essentially all estimates are below the vertical bias of SRTM. Of these, we consider two representative sea-level projections for this assessment, labeled here as K14 3 and K17 4. K14 is a probabilistic projection that is closely aligned with IPCC findings 10 , 30 , while K17 is not probabilistic and emphasizes the possibility of more rapid sea-level rise because of unstable ice-sheet dynamics Further details of these models are discussed in the methods section.
Both sets of projections are conditional on global carbon emissions; RCPs 2. These models use as the baseline year zero sea-level rise , which we treat as present-day with respect to sea level for relevant vulnerability estimates. Because higher and more frequent coastal flooding is a direct impact of sea-level rise 33 , 34 , we also assess potential exposure to ECWLs resulting from annual floods added on top of rising seas.
These return levels vary spatially from a 5th percentile of 0. Given each sea level scenario analyzed Supplementary Table 1 , and alternately using SRTM and CoastalDEM, we estimate the number of people on land that may be exposed to coastal inundation—either by permanently falling below MHHW, or temporarily falling below the local annual flood height Table 1 , Supplementary Data 1.
Coastal defenses are not considered, but hydrologic connectivity to the ocean is otherwise enforced using connected components analysis. Future population growth and migration are also not considered; rather, we use essentially current population density data from Landscan 13 to indicate threats relative to present development patterns.
Low-lying areas isolated from the ocean are removed from the inundation surface using connected components analysis.
Gray areas represent dry land. Axis labels denote latitude and longitude. Population exposure to projected sea level or coastal flooding is most commonly expressed as the total estimated exposure below a particular water level total exposure 14 , 16 , 17 , 19 , 21 , 36 , but is increasingly also presented as the difference in exposure above a contemporary baseline marginal exposure 16 , 21 , Each approach has complementary strengths and limitations, discussed later.
Here, we include marginal exposure values for key findings, while focusing more on total exposure. These values form the basis of the difference between total and marginal exposure estimates. For one moderate future scenario, sea levels projected by are high enough to threaten land currently home to a total of — million people to a future permanently below the high tide line, or a marginal increase of 40 30—60 million.
Total and marginal exposure each rise by another 50 20—90 million people by end of century. A total of — million people are on land threatened by annual flood events in , or an extra 60— million beyond the contemporary baseline. In the case of Antarctic instability, a total of — million people today live on land indicated as vulnerable to an annual flood event by mid-century, rising to as many as — million by These values represent marginal increases of 50 20—90 and — million from the present, respectively.
More broadly, the effect on estimated ECWL exposure from changing the elevation data used exceeds the combined effects of emissions level, Antarctic behavior, and incorporation of annual flooding, as assessed using SRTM. For example, based on CoastalDEM, the total median current population on land falling below the projected mean higher high water line in under low emissions and a fairly stable Antarctica RCP 2.
The marginal increases in exposure from baseline are 20 6—41 million and 34 7—77 million , respectively. Under the same emissions scenario and either sea-level model, annual flood events at least double the corresponding estimates, threatening land occupied by over 60 million additional people.
Total populations on vulnerable land. Estimates based on CoastalDEM. Countries wholly north of 60 degrees N are excluded because CoastalDEM is undefined at those latitudes. Source data are provided as a Source Data file. National boundaries based on public domain vector map data by Natural Earth naturalearthdata. Bangladesh, India, Indonesia, and the Philippines see a 5-fold to fold change in estimated current populations below the projected high tide line after applying CoastalDEM.
Globally, application of CoastalDEM leads to increased exposure estimates for the great majority of nations Fig. The total global value is designated with the red point. Source data are in Supplementary Data 1. Percentage rather than absolute exposure serves as a normalized metric of threat Supplementary Data 4. It follows that some coastal municipalities within these nations will see even larger proportions of their populations threatened with displacement.
This count is up from two using SRTM. Supplementary Data 1 and 4 provide results for the present, mid-century, and The aspirational outcome of applying CoastalDEM to ECWL exposure analysis is to, as closely as possible, estimate the same amount of coastal vulnerability that a DEM derived from airborne lidar data would. We validate our results by first performing three representative ECWL exposure analyses using lidar-derived data in the US and Australia. In Fig. Values of nearly zero imply a close match between exposure computed using both lidar and the target DEM, while larger absolute values suggest under-estimation or over-estimation of vulnerability.
We note that while the neural network that generated CoastalDEM was trained on lidar-derived data in the US, Australian lidar data is used only to validate the results, meaning strong results seen here mitigate fears that the model has been overfitted. The neural network that generated CoastalDEM did not fully correct this large error. SRTM error is strongly correlated with factors such as land slope 39 , dense vegetation 24 , and high population density 40 , which themselves exhibit natural spatial autocorrelation.
These features could manifest at any number of spatial scales some towns may be only a few kilometers wide, while some urban agglomerations and forests are far larger. We therefore conduct a sensitivity analysis to explore the potential effects of error in CoastalDEM on our population exposure estimates, including the effects of autocorrelated error. Monte Carlo simulations are regularly used to model DEM error and generate distributions of flood exposure estimates, from which uncertainty may be evaluated 38 , 42 , The wide range of autocorrelation scale present here makes the second option unsuitable, and with no ground-control-point data available globally, the third is not possible.
Because of our expectations around the importance of spatial autocorrelation, we apply a modified, multi-scale approach to the first of these three methods. Assuming a normal distribution of error centered on zero and using a fixed global standard deviation, we generate error fields using each of 6 different block sizes within which uniform error applies, ranging from 1 pixel 3 arcseconds to 1 degree. We add the blocked errors to the original CoastalDEM to produce new simulated 3 arcsecond DEMs for computing exposure; the resulting exposure distributions are then evaluated separately for each block resolution.
While vertical error will inevitably vary some from place to place, the similarity in error between the US and Australia increases our confidence in the value we employ. As in the main study, connected components analysis is used to remove isolated areas under the inundation surface before computing exposure. Table 2 and Supplementary Data 5, respectively, provide global and country-level results for this sensitivity analysis. This bias may be caused by higher spatial frequency DEM alterations cutting off some low-lying inland areas connected to the ocean through narrow pathways in the original CoastalDEM.
Consistent with this mechanism, bias dissipates at larger error-block sizes. In general, larger areas of analysis and smaller error blocks lead to less sensitivity in ECWL exposure estimates, because each of these factors leads to larger random samples, making errors more likely to cancel out.
Conversely, smaller areas and larger blocks each lead to smaller samples and more sensitivity. These results suggest that CoastalDEM error exerts little influence on our global estimates, but reasonable caution should be applied when interpreting national scale assessments, particularly for smaller countries such as the SIDS.
Given the known factors at many spatial scales that contribute to DEM error, this assumption is unrealistic. Assessing characteristic error autocorrelation scales is beyond the scope of this study, but realistic CIs will be considerably narrower than implied by the 1-degree scale.
Despite improvements, elevation dataset error remains an important limitation in this study. Their use for research has faded in comparison with SRTM, given its higher horizontal resolution and order-of-magnitude lower error. AW3D30 is a digital surface model primarily derived from stereo optical satellite imagery, and does not specifically attempt to improve vertical bias in either urban or forested areas.
It uses regression analysis to remove vertical error correlated with a number of vegetation metrics. For sake of comparison, the analyses described in this article were repeated for these DEMs, and included in Supplementary Data 1 and 4. Future modeling efforts may improve estimation of terrain elevations in tall-building districts and areas affected by SRTM striping.
Ultimately, the most accurate assessments of vulnerability to rising seas, especially for smaller areas, will require development and public release of improved coastal area elevation datasets building directly off of new high resolution observations increasingly collected by satellites today. While Landscan is widely used in the research literature, it cannot capture any bias toward or away from development within the lowest-lying coastal areas at sub-kilometer spatial scales.
GRUMP is another population dataset with the same horizontal resolution, though it involves less sophisticated spatial modeling and is available only through It models nighttime rather than ambient population density 51 , and has been shown to produce notably higher predictions of exposure to ECWL Gridded Population of the World 52 is another alternative, based directly on census data without further modeling.
Newer datasets, such as Worldpop 53 and the High Resolution Settlement Layer 54 , are anticipated to model population densities with higher accuracy at finer resolution, but are not yet available globally.
We emphasize that this analysis combines future water level projections with contemporary population densities. Results should therefore not be taken as projected impacts. Rather, they reflect the portion of presently developed land at risk in the future, which we interpret as a threat indicator. Efforts to integrate projected population growth, migration, economic development and coastal defenses into ECWL exposure projections have begun 19 , 36 , However, the spatial scales of socioeconomic projections remain very coarse compared to the scales at which elevation and current development data are available, posing a stiff challenge to their meaningful integration into analyses where fine-scale detail is critical.
In addition, behavioral and economic responses to rising seas are likely to be unpredictable, due to the largely unprecedented nature and scale of the problem. The vulnerability model employed in this analysis, a bathtub model where we classify all land below a given water height and hydrologically connected to the ocean as exposed to extreme coastal water levels, presents another partial limitation of the study. While this approach is reasonable in indicating land threatened with permanent inundation due to higher sea levels, it tends to overestimate exposure from episodic flooding, especially at small spatial scales 56 , It is likely that hydrodynamic models would predict less vulnerability to one-year floods than we estimate here.
Areas accordingly misclassified as exposed to annual flooding would nonetheless likely face relatively frequent inundation risks. Furthermore, this analysis assumes a static coastal topography, with the exception of a linear model of vertical land motion implicit in the sea-level projections used.
It is difficult to predict how these factors affect the uncertainty of our results, especially since sea-level change may trigger complex process responses. However, we note that armored, developed, and maintained shorelines in urban areas, where vulnerable populations are concentrated, may generally be less susceptible to such factors than undeveloped land.
Several explanations are possible. What we showed in this recent paper is that when you look at the modern data on rotation and you correct for ice age, you have a leftover, and that leftover is precisely what it should be if it were due to the kind of melting that global change scientists believe happened in the 20th century. This is an entirely different way to show that ice sheets are melting.
And so rotation provides what a scientist would call a really elegant integrated measure of the mass balance of polar ice sheets. In my family, we had more discussions about Renaissance history than we ever did about science.
Some said it was some ice effect, that ice volumes had changed. More often people thought that it was linked to changes in the rate at which tectonic plates were created. There are so many interesting problems in our science that you can see with your eyes. But your eyes can fool you. Richard Feynman, the great physicist, used to start his physics lectures by showing students their intuition could take them a long way.
They could do things just through intuition that would get them roughly the right answer. Then he used to throw some counterintuitive examples at them. You need to understand when your intuition might go wrong. I think some scientists would disagree with me, but I think you really do have to give yourself time to think.
And I strongly encourage my graduate students to have other interests, because the best way to have that time is to take a break from science. Nautilus uses cookies to manage your digital subscription and show you your reading progress. It's just not the same without them. Please sign in to Nautilus Prime or turn your cookies on to continue reading. Thank you! Global melting: Though it may seem counterintuitive, melting glaciers in one area may cause local sea levels to drop—while causing a rise in sea levels farther away.
What happens when a big glacier like the Greenland Ice Sheet melts? So what is the combined impact of the ice-sheet melt, water flow, and diminished gravity? What happens with melting in Antarctica?
Why is the source of the melt important? Diminished gravitational attraction lowers the sea near an ice sheet. At the same time, water flowing into the ocean raises it.
0コメント