How The Ocean Cleanup Mapped the World's Rivers | Research | The Ocean Cleanup

How The Ocean Cleanup Mapped the World's Rivers | Research | The Ocean Cleanup

To substantially decrease riverine plastic emissions into the ocean, in how many rivers do we need to implement mitigation measures, where are these rivers located, and what are the characteristics of these rivers? For the last three years, researchers at The Ocean Cleanup have been working on answering these questions to effectively reduce oceanic plastic pollution. Our study – published today in  Science Advances  – estimates that 1000 rivers account for nearly 80% of global riverine plastic emissions into the ocean, made up of a high number of small and medium-sized rivers. While this number is much higher than previous estimations (100 times), it is only 1% of rivers worldwide, which means solving the problem is feasible. By collectively taking a global approach with various technologies to target these most polluting rivers, we can drastically reduce the influx of plastic into the ocean.

These high-resolution data allow for The Ocean Cleanup and other organizations worldwide to develop focused mitigation strategies and technologies to reduce riverine plastic emissions. In 2019, The Ocean Cleanup launched our river cleanup technology, the Interceptor, and have been using this data as a compass to find suitable locations to deploy Interceptor solutions. With this study now accepted, The Ocean Cleanup can use its findings to further support and accelerate global plastic extraction initiatives, policy, and consumer changes and awareness.


The Ocean Cleanup’s mission is to rid the world’s oceans of plastic. Decreasing the inflow of plastics into the ocean is one part of our solution to solve the problem of plastic accumulating in our oceans. But how are plastic emissions distributed over the 100,000 rivers, creeks, and canals that exist worldwide? Two previous studies (2017) gathered data on mismanaged plastic waste generated within a river basin and applied a correction factor to estimate the river’s plastic emission. These studies predicted that as little as 5 or 47 rivers account for 80% of riverine plastic emissions.

Since 2017 more data on macroplastics in rivers became available from several research groups, including The Ocean Cleanup. These data, and new insights, suggested that global riverine plastic emissions could be more widespread than previously thought. Therefore, to effectively impact pollution in rivers and oceans, we needed to update our global river plastic emissions model by including more details on plastic transport dynamics and data at a higher spatial resolution.

In collaboration with researchers from various institutions and organizations, we created a new model framework and included data from 136 field measurements, representing 67 rivers in 14 countries of 3 different continents, collected between 2017 and 2020 to calibrate and validate our model. We used analogous research and insights from field campaigns to conceptualize plastic transport and worked with 60 earth science and plastic experts to parametrize our probabilistic model. Our updated model now suggests that  1000 rivers account for nearly 80% of global annual emissions, ranging between 0.8 million and 2.7 million metric tons per year , with small urban rivers among the most polluting.

To build a model that simulates global riverine plastic transport, we needed global data sets. By utilizing data on terrain slope, elevation (DEM), flow direction (HydroSHEDS), runoff data (GRUN), land use data (GLC2000), precipitation and wind data (WorldClim2), we calculated a global river network, river classes, terrain characteristics, the distance from every location on Earth towards the nearest river and ocean, and the corresponding mobilization and transport probabilities. We then figured the probability for mismanaged plastic waste to reach the ocean for each grid cell. This plastic emission probability is then multiplied with the corresponding mismanaged plastic waste and accumulation within a river basin, resulting in the annual plastic emissions per river mouth.


Plastic waste increasingly accumulates in the marine environment, but data on the distribution and quantification of riverine sources required for development of effective mitigation are limited. Our model approach includes geographically distributed data on plastic waste, land use, wind, precipitation, and rivers and calculates the probability for plastic waste to reach a river and subsequently the ocean. This probabilistic approach highlights regions that are likely to emit plastic into the ocean. We calibrated our model using recent field observations and show that emissions are distributed over more rivers than previously thought by up to two orders of magnitude. We estimate that more than 1000 rivers account for 80% of global annual emissions, which range between 0.8 million and 2.7 million metric tons per year, with small urban rivers among the most polluting. These high-resolution data allow for the focused development of mitigation strategies and technologies to reduce riverine plastic emissions.


Plastic pollution in oceans and rivers is an emerging environmental hazard ( 1 ), and accumulation on riverbanks, deltas, coastlines ( 2 ), and the ocean surface ( 3 ) is rapidly increasing. Of all the plastics ever made to date, it was estimated that 60% has been discarded in landfills or in the natural environment ( 4 ). Plastic pollution poses threats on aquatic life, ecosystems, and human health ( 5 6 ). Plastic litter also causes severe economic losses through damage to vessels and fishing gear, negative effects on the tourism industry, and increased shoreline cleaning efforts, adding up to US$1.26 billion per year for the Asian-Pacific Rim alone ( 7 ). Work on the origin and fate of plastic pollution in aquatic environments suggests that land-based plastics are one of the main sources of marine plastic pollution ( 8 ), either by direct emission from coastal zones ( 9 ) or by transport through rivers ( 10 11 ). Riverine plastic transport remains understudied, especially in areas that are expected to contribute most to global plastic emissions into the ocean ( 12 ). A better understanding of pathways and transport mechanisms of plastic waste to and within rivers and the global distribution of riverine plastic emissions into the ocean is a prerequisite to developing effective prevention and collection strategies.

Previous attempts to estimate the distribution of global riverine emissions of plastic into the ocean ( 10 11 ) relied on empirical indicators representative of waste generation inside a river basin. These assessments demonstrated a significant correlation between (micro)plastic concentration data collected by surface trawls in rivers, national statistics on mismanaged plastic waste (MPW) generation, and population density. For both studies, an empirical formulation was presented on the basis of this correlation, which was extrapolated to other rivers where data were not available. This resulted in predicted plastic (micro- and macroplastics combined) emissions of 1.15 million to 2.41 million metric tons (MT) per year ( 10 ) and 0.41 million to 4 million MT year–1 ( 11 ). These studies did not account for spatial distribution of plastic waste generation or climatological or geographical differences within river basins. According to these studies, the 10 largest emitting rivers contribute 50 to 61% and 88 to 94% to the total river emissions. Both models agreed on a disproportional contribution of Asian rivers to global plastic emissions. While these modeling efforts have provided a first approximation of the magnitude and spatial distribution of global riverine plastic emissions, they emphasized the scarcity of data on macroplastic contamination in freshwater ecosystems. Available measurements used for calibration of emission predictions were not always collected directly at the river mouths, and studies reported data on plastic contamination using varying units and methods, including surface trawling from boats or bridges ( 13 15 ).

Sampling methods using surface net trawls for freshwater plastic contamination may be well suited for monitoring microplastic concentrations (size, <0.5 cm). However, insufficient sampled volumes limited by net opening width or pump outlet dimensions may result in the underestimation of macroplastics (several centimeters in size) ( 16 ) that account for most of the mass of plastic emissions ( 17 ). Instead, visual observations from bridges provide more consistent results for the quantification of floating macroplastic in rivers ( 18 ). In recent years, results from long-term visual counting campaigns for the quantification of floating macroplastic emissions from rivers of different continents have been made available ( 19 ). At a global scale, these studies provided observational evidence for the disproportional contribution of Asian rivers in plastic emissions predicted by numerical models ( 20 24 ). Nevertheless, at a local scale, the studies reported discrepancies between observations and theoretical formulation ( 23 ), emphasizing the limitation of current models and the need for a revised formulation accounting for basin-scale geography, land use, and climate to more accurately estimate floating macroplastic emissions.

Here, we present a revised estimate of global riverine macroplastic emissions into the ocean using the most recent field observations on macroplastics and a newly developed, distributed probabilistic model to more accurately represent driving mechanisms of plastic transport (e.g., wind, runoff, and river discharge), differentiating between areas with different land use and terrain slope, and including plastic retention on land and within rivers. Microplastic transport is not included in this study. We derived probabilities for plastic waste to be transported from land to river and from river to sea from six different geographical indicators and generated a high-resolution (3 × 3–arc sec cells) global map of the probability for waste discarded on land to reach the ocean within a given year. This information combined with the most recent estimates of MPW generation on land ( 25 ) allowed us to estimate the annual emissions of plastic from rivers into the ocean. We calibrated and validated our model against 136 recent field observation data points ( n  = 52 for calibration and  n  = 84 for validation) of monthly riverine macroplastic transport from more than 67 rivers in 14 countries. We show how the consideration of transport probability for plastic within a river basin can highly increase or decrease the estimated emissions of the corresponding river into the ocean. At a global scale, this results in a considerably wider distribution of source points with large rivers contributing less to the total than expected, while urban rivers in South East Asia and West Africa are identified as the main hot spots for plastic emissions. We classified macroplastic-emitting rivers according to size, providing insight into which river class contains the highest number of rivers and the largest accumulative emissions. The classification and distribution of emission points provide a basis for development of mitigation strategies and technologies as well as a road map for upscaling existing mitigation technologies.


Study design

In this study, we calculate the probability for MPW generated inside a river basin to leak into aquatic environments. When combined with spatial data on MPW generation ( 25 ), our framework (Fig. 1) allows for the prediction of riverine plastic mass emissions  ME  into the ocean. Probabilities are derived from physical and environmental characteristics including precipitation, wind, terrain slope, land use, distance to river, river discharge, and distance to the ocean. Precipitation and wind are included as mobilizing forces and differentiate between climate zones, while terrain slope and land use are included to reflect the probability to reach a river and account for differences in landscapes from which MPW is generated. Distance to river and distance to ocean include the geographical location of MPW generation in relation to the nearest river and ocean. The probability of MPW to reach the nearest river depends on the landscape and the distance between the location of generation and the river. We conducted an expert elicitation to constrain model parameters and explored the impact of parameters with sensitivity analyses. We calibrated and validated our model against 136 field measurements of monthly emissions of floating macroplastics from 67 different rivers, collected between 2017 and 2020. A Monte Carlo and one-at-a-time (OAT) sensitivity analyses were performed, showing correlations of individual parameters with model output and field observations. On the basis of the ratio of residuals between 125 observed and modeled locations, a confidence interval was constructed. Model predicted emission points could range within a factor of 4 with a 68% confidence interval and a factor of 10 with a 95% confidence interval.

Fig. 1 Model framework.

Plastic emission in a river mouth  ME  is computed by accumulating of MPW multiplied with the probability of waste leaking into the ocean,  P ( E ) within a river basin.  P ( E ) is constructed with  P ( M ),  P ( R ), and  P ( O ), which contain physical processes accountable for MPW transport.

Comparison with observations

A dataset of monthly averaged plastic transport near the river mouth was constructed from literature case studies and observational reports. This dataset was divided into a dataset used for calibration and one for validation (tables S1 and S2, respectively). Data that were collected and published before March 2019 were used for the calibration. Data published or made available to us after March 2019 were used for the validation. These studies use standardized methods to observe and quantify macroplastic transport according or comparable to published approaches ( 18 26 ); see table S3 for details on observational and model-predicted riverine plastic emissions per month or per year.

Calibrated model results were compared with field observations, and a good order-of-magnitude relationship was demonstrated (coefficient of determination,  r 2 = 0.71,  n  = 51). All model predictions are within one order of magnitude from observations (the Pasig River is on the border of one order of magnitude) (Fig. 2 and fig. S1), except for the Kuantan River. The Kuantan River is considered an outlier, with observed concentrations an order of magnitude lower than estimated by the model; when the Kuantan River is included in the model, the coefficient of determination  r 2 is 0.61 (table S4).

Fig. 2 Observations compared with modeled data for floating macrolitter emissions per river.

Regression analysis carried out with 136 records from 67 different rivers of different sizes spread across the globe. The dataset was split into a calibration ( n  = 52) and a validation ( n  = 84) dataset. The coefficient of determination of the logarithmic regression,  r 2, is 0.71 for the calibration and 0.74 for the validation dataset. Symbols indicate midpoints of extrapolated measurements (MT month−1) on the  x  axis versus our best calibrated model prediction on the  y  axis. The horizontal whiskers indicate the upper and lower values reported for observational data (if published), and the vertical whiskers indicate the upper and lower value of the 68% confidence interval of model predictions. The dark blue symbols correspond to data points used for calibration, and light blue symbols represent the validation data points, while the symbol (triangle, circle, and square) indicates the continent from where the location originates. The logarithm of both the measurements and the model results is presented here. The dotted gray lines represent one-order-of-magnitude deviation from the  x  =  y  line in the middle. The Kuantan and Besos rivers (indicated in red) are outliers with more than one order of magnitude difference compared with observational results.

We validated (Fig. 2 and fig. S2) our model against 84 independent data points collected from literature on macroplastic observations or macroplastic correlated to microplastic observations. These data points originate from 51 rivers in six countries. We consider the Besos River (10 data points) an outlier because there are four weirs, which may act as a sink for plastics ( 27 ) directly upstream from the observation point ( 28 ). The remaining 74 validation data points are within one order of magnitude from the observed values, and our model predictions demonstrated a better correlation than during the calibration exercise (coefficient of determination,  r 2 = 0.74,  n  = 74). A separate validation graph (fig. S3) shows the results where 39 Japanese prefectures are merged into one single data point. This illustrates that the validation results are not biased by the inclusion of many points from one single country ( r 2 = 0.84).