Recent pace of change in human impact on the world’s ocean

Humans interact with the oceans in diverse and profound ways. The scope, magnitude, footprint and ultimate cumulative impacts of human activities can threaten ocean ecosystems and have changed over time, resulting in new challenges and threats to marine ecosystems. A fundamental gap in understanding how humanity is affecting the oceans is our limited knowledge about the pace of change in cumulative impact on ocean ecosystems from expanding human activities – and the patterns, locations and drivers of most significant change. To help address this, we combined high resolution, annual data on the intensity of 14 human stressors and their impact on 21 marine ecosystems over 11 years (2003–2013) to assess pace of change in cumulative impacts on global oceans, where and how much that pace differs across the ocean, and which stressors and their impacts contribute most to those changes. We found that most of the ocean (59%) is experiencing significantly increasing cumulative impact, in particular due to climate change but also from fishing, land-based pollution and shipping. Nearly all countries saw increases in cumulative impacts in their coastal waters, as did all ecosystems, with coral reefs, seagrasses and mangroves at most risk. Mitigation of stressors most contributing to increases in overall cumulative impacts is urgently needed to sustain healthy oceans.

Impacts of human activities on the ocean have been shown to be substantial, ubiquitous 1 and changing 2 . The resulting cumulative impact of these activities often leads to ecosystem degradation or even collapse [3][4][5][6][7] , and studies of individual marine ecosystems (e.g., coral reefs, kelp forests, seagrasses) have shown declines in condition globally due to increasing anthropogenic stressors [8][9][10][11][12][13] . Ongoing and emerging policy around managing for cumulative impacts to the oceans creates a pressing need to understand how, and how fast, cumulative impacts are changing. Expansion of existing uses of the ocean and emerging new ones -including offshore energy, ocean farming, and ocean mining -requires an understanding of what else is impacting those locations, how those new uses will add to existing impacts, and critically whether the cumulative impact of these ocean uses is changing, and how quickly. Both the European Union's Marine Strategy Framework Directive and the United Nation's Sustainable Development Goal 14 focus on assessing and reducing cumulative pressures to the oceans, and the upcoming renegotiation of the Aichi Biodiversity Targets in 2020 will benefit from a deeper understanding of the pace and pattern of change in cumulative impacts. Furthermore, the accelerating rate of creating marine protected areas (MPAs) to meet Convention on Biological Diversity (CBD) targets of 10% of the ocean within protected area by 2020 (ref. 14 ), and the push globally to create very large MPAs 15 , could similarly benefit from detailed maps of where and how fast cumulative impacts are changing, as this information is critical to siting and managing effective protected areas.
To assess the pace of change in cumulative human impacts (CHI) we calculated and mapped the cumulative impact of 14 stressors related to human activities (including climate change, fishing, land-based pressures, and other commercial activities) on 21 different marine ecosystems globally for each of eleven years spanning 2003-2013 (Fig. S8), building on previous methods developed to calculate and map CHI 12,13 . The intensity of each stressor is mapped at 1 km resolution and rescaled to values between 0 and 1, using either known or estimated ecosystem thresholds or upper quantile values from the distribution of global stressor intensity values across years. The intensity of each stressor is then converted to an estimate of impact on each ecosystem by multiplying the stressor's intensity by the corresponding ecosystem vulnerability where the ecosystem occurs 16  www.nature.com/scientificreports www.nature.com/scientificreports/ deeper ecosystem types. Even for coastal ecosystems, climate stressors were dominant drivers of change (Fig. 4b), although land-based pressures and shipping also increased notably for many ecosystems (Fig. 4b).
Mapping the pace of change in cumulative human impacts on the ocean provides a fundamentally novel understanding of current and potential future risks to marine ecosystems and biodiversity. The vast majority of the ocean is experiencing significantly increasing impacts from multiple human stressors (Fig. 1a); much of this area currently remains at relatively modest impact, such that a snapshot view of impact gives a false sense   www.nature.com/scientificreports www.nature.com/scientificreports/ of condition (Fig. 1B). More critically, if current trajectories of change persist, the global cumulative impact of humans on the ocean will be profound and may rapidly push many ocean regions past critical tipping points of sustainability 1,2,4,20 .
Despite these sobering results, messages of hope remain. During this time period, many countries, particularly in Europe, Asia, and parts of Africa, saw notable declines in impacts from commercial fishing, and many countries saw reduced impacts from land-based pollution (Fig. 3). In a few cases, these declines were larger than increases in climate change and other stressors, leading to overall decreases in CHI, and in all cases the declines helped mitigate increases in CHI. Coordinated, comprehensive management that accounts for multiple stressors can leverage decreases in single stressors to accommodate potential increases in others when making strategic development and conservation decisions. Results also highlight that spatial variability in the local manifestation of climate change may offer local refugia that can be targeted for protection and management to 'buy time' in efforts to mitigate and adapt to a changing climate 21 . Despite major challenges in reducing greenhouse gas emissions, these results indicate that climate mitigation to meet targets of the Paris Agreement would have major www.nature.com/scientificreports www.nature.com/scientificreports/ positive impact on the condition of marine ecosystems and would significantly slow or halt increasing trends in CHI in vast areas of the ocean.
Our results are robust to many methodological decisions because our focus was on change in impact over time, as data processing and analytical decisions remain consistent across all years. Furthermore, we have found previously that global patterns and results are robust to model parameters, including stressor vulnerability weights 12,22 . However, for several reasons our results are likely conservative 22 . First, many human activities with known stressors to marine ecosystems could not be included (e.g., deep sea mining, plastic pollution, offshore energy, aquaculture, noise pollution, terrestrial mining, logging, oil spills), primarily because of limited or nonexistent data on the spatial distribution or temporal change in their intensity. Many of these excluded activities have been expanding in geographic extent and intensity over the past decade. Second, our analysis did not include the most recent 5 years of impact because many datasets are not yet available for these years, during which time many are expected or known to have further increased, most notably climate related stressors 23 . Third, multiple interacting stressors often produce synergistic rather than simply additive outcomes 24 , such that increasing intensity of individual stressors within the context of multiple stressors is likely to accelerate CHI faster than we modeled here. Finally, we expect nonlinear relationships to exist between ecosystem condition and the intensity of both individual stressors and CHI [25][26][27] . These nonlinearities would lead to faster than linear increases in ecosystem impact with increasing stressor intensity that would necessitate lower thresholds for rescaling individual stressors. Very few data or known thresholds currently exist to allow inclusion of these nonlinearities in our assessment.
Previous snapshot views of cumulative human impacts on the ocean 12,13 have already been widely used to inform where to locate new MPAs [28][29][30] , new ocean uses within a spatial planning framework 31 , and new conservation or restoration strategies 32 ; to assess if existing MPAs are working 33 ; and, combined with biodiversity data, to assess species risk to inform Aichi targets and other conservation goals 14 . Understanding the pace of change in human impacts provides a much richer understanding of how, where, and critically how quickly, human activities are affecting marine ecosystems and ultimately the services they provide humanity, thus offering a much more informed baseline to guide strategic conservation actions and assessments.
Looking forward, with human dependencies on land expanding and increasingly leading to conflict, countries around the world are progressively pushing into the ocean, intensifying past uses and adding additional onesincluding offshore energy, marine aquaculture, and even human settlements. Such expansions are driven by the need to feed and support a rapidly growing global human population, but come with yet greater impacts on the ocean. This reality requires humanity to face difficult decisions ahead. To help support the global human population and mitigate the impacts we are having on our landscapes, we are shifting our impacts into the sea. How much more change can these ecosystems endure?

Methods
We calculated change in the intensity of 14 stressors from human-based activities during an 11 year period from 2003 to 2013 at a ~1 km resolution and estimated the impact on the global ocean based on the magnitude of the stressor as well as the vulnerability of 21 marine ecosystems to each stressor. The cumulative impact of all 14 stressors was then calculated for each km 2 for each year. To determine annual change in stressor and cumulative impacts, we applied a linear regression model to each raster cell and calculate whether the slope is significantly different from 0. General model. We calculated stressor and cumulative impact I c at a ~1 km resolution, based on previously developed methods 12,13 , using the following information: 1. Stressor intensity rasters describing the magnitude of 14 stressors on a scale of 0-1 (1 is highest relative stress). 2. Ecosystem rasters describing the location (1 if present, otherwise NA) of 21 global marine ecosystem types. 3. Vulnerability matrix describing the vulnerability of each ecosystem to each stressor. Vulnerability is a value from 0-4.
Stressor impacts, I s , were calculated for each of the 14 stressors by first multiplying the stressor intensity raster, D j , by each ecosystem raster, E i , and the corresponding vulnerability value, μ ij . The stressor × ecosystem × vulnerability rasters are summed for each stressor and then divided by the number of habitats (m) in each raster cell: Nearly all source data used to derive the stressor layers had native coarser resolution (Table S2), and were therefore resampled/reprojected using nearest neighbor estimates of cell values. Using the nearest neighbor approach preserves the values of the original data, and assumes the coarse-scale value is evenly distributed across all 1 km cells within that region. The coarser scale pattern is essentially recreated and finer resolution information is preserved where and when it is appropriate. www.nature.com/scientificreports www.nature.com/scientificreports/ We used the WGS84 Mollweide projection because it preserves area so data towards the poles are not visually over-represented.

Stressors. We include stressors from 4 primary categories:
• Fishing: commercial demersal destructive, commercial demersal nondestructive high bycatch, commercial demersal nondestructive low bycatch, pelagic high bycatch, pelagic low bycatch, artisanal • Climate change: sea surface temperature, ocean acidification, sea level rise • Ocean: shipping • Land-based: nutrient pollution, organic chemical pollution, direct human, light Stressors included in our analyses are listed in Tables S1 and S2 and described in detail within the Supplementary Information.
Given our focus of describing trends in human impact on marine ecosystems, global datasets reporting results at regular intervals were critical. Given this constraint, we were unable to include some stressors from previous impact analyses 12,13 because they did not include enough information for us to estimate annual change from 2003 to 2013. These excluded stressors include invasive species, ocean pollution, UV intensity, and benthic structures.
Other anthropogenic drivers we considered, but could not be included due to incomplete spatial or temporal coverage, were: hypoxic zones, coastal engineering (piers, rock walls, etc.), non-cargo shipping (ferries, cruise ships, etc), aquaculture, disease, changes in sedimentation and freshwater input, and tourism.
Given discrepancies in how different data layers define the global coastline, we resolved differences by extending all stressor rasters to a common coastline. In some cases the gaps occurred because the monitoring that produced the stressor data missed some regions; however, it was most often due to converting a coarser resolution raster to a finer resolution raster, resulting in zig-zags of missing data along the coast. This made gapfilling necessary. Details for how spatial gapfilling was done for each stressor are provided in the Supplementary Information.
Stressors are rescaled to have values between 0-1. Rescaling allows for direct comparison among drivers with dramatically different units of measurement. With the exception of ocean acidification, we rescaled the data by dividing by the 99.99th quantile across all global raster cells and years (values are capped to a maximum value of 1). We used all years of data to rescale the data to ensure comparability across time periods. The 99.99th quantile was used to minimize the influence of outliers. This approach assumes a linear relationship between the magnitude of the stressor and the impact on the ecosystem. This assumption ignores thresholds that likely exist but are known for very few stressors. For ocean acidification we used known information about biological thresholds to rescale the data.
For many stressors, the distribution of values was highly skewed such that rescaling relative to the highest values resulted in intermediate values of the stressor that were underestimated. In these cases, we log transformed stressor values prior to rescaling. We indicate when data were transformed in descriptions of each stressor in the Supplementary Information.