Improving urban trash reduction tracking with spatially distributed Bayesian uncertainty estimates

https://doi.org/10.1016/j.compenvurbsys.2019.05.001Get rights and content

Highlights

  • An approach is presented to quantify uncertainty associated with assessments of urban trash loading to stormwater systems.

  • Outputs can be used to track trash mitigation effectiveness and prioritize areas for action.

  • The approach has been incorporated into web-based geospatial stormwater management platform.

Abstract

Urban stormwater runoff is among the most significant sources of trash delivery to waterways, degrading aquatic habitats and contributing to oceanic trash gyres across the globe. Municipal water quality permits that require elimination of trash inputs to stormwater systems employ visual trash assessments on city streets to demonstrate litter reduction progress. We present a novel method to increase the utility of these assessments by quantifying their degree of certainty at a granular spatial scale via Bayesian credibility intervals. Using data collected in the City of Salinas, California, we illustrate how the outputs can be used to determine effective trash controls and prioritize areas for management actions. Spatial dependence was incorporated to the uncertainty estimates within individual stormwater drainages and results were interpolated to adjacent parcels. After 3–6 observation periods over 20 months, we found approximately 30% of the city area showed minimal litter accumulation at an 80% certainty level. The outputs provide a practical alternative for cities to determine compliance with stormwater trash regulations, update understanding of trash accumulation patterns, and iteratively adjust sampling designs in response to new observations. The methods described have been implemented as a web-based geospatial decision support tool to help stormwater managers target implementation actions and report progress to regulators.

Introduction

With growing understanding of the magnitude of trash delivery to oceans, the need for action to prevent further aquatic habitat degradation is now widely recognized (Day, Shaw, & Ignell, 1989; Hammer, Kraak, & Parsons, 2012; Law et al., 2010; Moore, 2008). Urban trash (alternatively termed anthropogenic litter) is a water pollutant that impairs beneficial uses, degrades aquatic habitats and causes entanglement, death from ingestion, and transport of invasive species (Sigler, 2014). While the relative contributions of various sources to ocean trash gyres like the Great Pacific Garbage Patch (Dautel, 2009) are not fully understood, urban stormwater contributes to the overall marine debris problem (EPA, 2011; Wheeler & Knight, 2017). Many communities throughout the United States, including City of New York (NYSDEC, 2015), City of Los Angeles (SWRCB, 2015a), San Francisco Bay Area (SFRWQCB, 2015), and the City and County of Honolulu (Hawaii Department of Health, 2012), have implemented water quality permit regulations to reduce or eliminate trash from urban stormwater. Recent amendments to the California Ocean Plan (SWRCB, 2015a) and the Water Quality Control Plan for Inland Surface Waters, Enclosed Bays, and Estuaries of California (SWRCB, 2015b) require California cities to reduce trash inputs to storm drain systems by 2030 to levels that do not adversely impact aquatic habitats.

Stormwater programs require a means to prioritize locations for mitigation actions, quantify urban trash reduction effectiveness, and provide meaningful reporting of annual progress to regulators. California cities have two options to meet National Pollutant Discharge Elimination System (NPDES) permit compliance: 1) install trash ‘full capture systems’ that separate or prevent downstream movement of all trash particles >5 mm in diameter or 2) implement a combination of institutional controls (e.g. street sweeping, trash pickups, education and outreach) to eliminate trash sources to the stormwater system. Areas not served by full capture systems require monitoring to demonstrate that they are clean enough to achieve ‘full capture equivalency’ with minimal trash available for transport into the storm drain system (SWRCB, 2015a).

Measuring the effectiveness of urban trash mitigation measures requires a reliable means to characterize patterns of trash accumulation within a city (e.g. Marais & Armitage, 2004; Marais, Armitage, & Wise, 2004). Various qualitative and quantitative field protocols have been used to characterize trash accumulation and impacts in waterways, with methods oriented to different study objectives such as understanding trash sources and types (Rosevelt, Los Huertos, Garza, & Nevins, 2013), transport dynamics and delivery to receiving waters (Moore, Cover, & Senter, 2007; Moore, Sutula, Bitner, Lattin, & Schiff, 2016; San Diego Bay Debris Study Work Group, 2016), habitat impacts (Hoellein, Rojas, Pink, Gasior, & Kelly, 2014), and community impacts (Muñoz-Cadena, Lina-Manjarrez, Estrada, & Ramon-Gallegos, 2012). Methods employed have included stream bank surveys (City of Los Angeles, 2016; SFRWQCB, 2004; Moore et al., 2007), flux estimates in rivers (BASMAA, 2016b), repeated roadway surveys (City of Los Angeles, 2017), and drone-based imagery analysis (Deidun, Gauci, Lagorio, & Galgani, 2018; Hengstmann, Gräwe, Tamminga, & Fischer, 2017). Since the amount of trash entering the stormwater system is dependent on the levels of trash that accumulate on streets, sidewalks, and other impervious surfaces (Wheeler & Knight, 2017) and other factors that influence transport (Moore et al., 2007), a measure of trash accumulation can provide an appropriate metric to estimate loading to receiving waters.

Visual assessments of trash accumulation on roadways are a rapid, qualitative method to measure the accumulated trash available for transport into the storm drain system within a certain area, which we refer to hereafter as trash condition. For this study, we employ a variation of a previously developed visual assessment protocol, termed the On-Land Visual Trash Assessment (OVTA), which has shown empirical association with measured trash loads (volume/area) (BASMAA, 2014), and has been accepted by the California State Water Resources Control Board (SWRCB) as a means to comply with water quality permit requirements (SWRCB, 2018). Compliance is achieved via OVTA results by demonstration that areas are in the ‘Low’ OVTA trash condition category (SWRCB, 2015a). The Low trash condition category is described as having a maximum of “a few small pieces of trash” within a city block length and has been estimated to have equivalent trash loading to the stormwater system as installation of full capture systems (BASMAA, 2014).

While the field protocol for OVTA is well-developed, sampling recommendations from (SWRCB, 2017) are based on very limited data and analysis (BASMAA, 2016a). Temporal and spatial variability of trash conditions create a substantial challenge for cities to determine appropriate frequency and spatial density of observations to achieve adequate levels of confidence or power for detecting changes. While important considerations for sampling design and spatial analysis have been addressed (Wheeler & Knight, 2017), they have not been incorporated to monitoring requirements from SWRCB. Moreover, sampling recommendations provided by a post-hoc sample size and power analysis are not responsive to new information as monitoring data are acquired. Since the variance of trash condition estimates may be non-stationary over both time and space (Lippiatt, Opfer, & Arthur, 2013; Moore et al., 2016; Ryan, Moore, Van Franeker, & Moloney, 2009), patterns of uncertainty are likely to shift as more data are collected. Since the implementation period for cities to achieve trash compliance will span at least a decade (SWRCB, 2015a), a dynamic approach to the problem can help cities efficiently allocate monitoring and implementation resources as their understanding of municipal trash patterns improves.

Uncertainty should be a fundamental dimension of quantifying the status of environmental conditions to support resource management decision making, since it has direct bearing on our capacity to use data and models to test hypotheses about patterns or changes (e,g. Beven, 2001; Beck, 1982). While use of models, both numeric and statistical, can facilitate wise decisions, dealing with uncertainty in the outputs is among the greatest challenges facing practitioners (Barton et al., 2012). Understanding of the applicability of Bayesian methods to environmental decision-making has grown in recent years (Barton et al., 2012; Ellison, 1996; Wikle, 2003), partly driven by development of new computational tools that make them available to a wider audience of researchers and development of more efficient implementation methods (e.g. Blangiardo, Cameletti, Baio, & Rue, 2013; Brown, 2015; Lunn, Spiegelhalter, Thomas, & Best, 2009). Bayesian methods have been widely applied to the problems of measuring uncertainty in environmental variables (Clark & Gelfand, 2006; Cressie, Calder, Clark, Ver Hoef, & Wikle, 2009; Pulkkinen, 2015) and been shown to have several advantages over frequentist counterparts for experimental design elements such as sample size determination (De Santis, 2007; Joseph, Du Berger, & Bélisle, 1997; Sahu & Smith, 2006).

The problem of ongoing trash condition assessment lends itself to an iterative sampling design that can incorporate new information as it becomes available. Bayes Theorem provides a formal method to update prior knowledge with new evidence wherein a prior expectation (previous belief) is combined with a likelihood function (new data) resulting in a posterior distribution (updated belief) that is used for statistical inference. The less information available in any given year to define trash conditions, the more influence the prior expectation has on the updated belief, and as more data are collected, the importance of the prior belief is diminished. The Bayesian approach allows use of diverse information types in the form of the prior and can be used to make probabilistic statements about trash conditions for specific years—which is important for regulatory compliance, and discrete locations—which is useful for targeting trash management actions and evaluating the effectiveness of those actions. A primary advantage is that it provides a direct accounting of uncertainty associated with parameter values (e.g. mean trash condition) owing to the fact that these parameters are treated as probability distributions rather than point values as they are in a frequentist framework. Accessibility to Bayesian methods has been facilitated by tools built in the R programming language to sample the posterior distributions via Markov Chain Monte-Carlo (MCMC) or estimate them via Integrated Nested Laplacian Approximations (INLA) (Blangiardo & Cameletti, 2015; Martins, Simpson, Lindgren, & Rue, 2013). In this study we provide a new application of Bayesian methods to quantify uncertainty associated with visual trash assessment data for the purpose of iteratively informing regulatory compliance, sampling design, and management actions.

Section snippets

Visual trash assessments

Visual trash assessment data were collected throughout the City of Salinas from spring 2017 to winter 2018 to estimate trash condition on city streets and sidewalks. Salinas is located on California's Central Coast with a population of 160,000 and an area of approximately 60 km2 (Fig. 1). The City is mostly surrounded by agricultural fields and stormwater flow to receiving waters is governed by eight sub-drainages primarily defined by the stormwater infrastructure. Three of the streams flowing

Urban trash condition and certainty patterns

Mean trash condition for the study period ranged from 2.5 to 935 L/ha. and the mean trash condition for all of Salinas was 70 L/ha, which falls near the median of the Moderate trash condition category. The majority of the 30-m grid cell values were in either the Low (42%) or Moderate (44%) categories and were low in the High (12%) and Very High (2%) categories (Fig. 3). Areas of the city with the most trash (High and Very High trash condition) were concentrated in commercial areas of the city

Discussion

The method presented employs a simple Bayesian approach to quantifying the certainty in trash condition estimates, which is intended to align with resources available to make such tools widely available to cities. Understanding patterns of trash condition uncertainty is critical for decision making since there is often little initially known about the variance of trash over space or time and the cost of ongoing assessments limits data collection. While spatial autocorrelation patterns of trash

Conclusions

Given the growing understanding of urban trash impacts and related water quality regulations, there is a critical need for cities to efficiently determine compliance and prioritize actions. This study demonstrated a methodology in the City of Salinas that is responsive to regulatory requirements and includes spatially distributed uncertainty estimates that provide valuable context for stormwater visual trash assessment results. Low levels of certainty were mostly prevalent in trashy areas, but

Acknowledgements

The authors acknowledge support from the City of Salinas Stormwater Program and contributions from their Stormwater Program Manager, Heidi Niggemeyer. Kevin Butler at the Environmental Systems Research Institute (ESRI) provided valuable suggestions to guide the analysis and interpretation of results along with detailed review of the manuscript.

References (66)

  • J.J. Barber et al.

    Hierarchical spatial modeling for estimation of population size

    Environmental and Ecological Statistics

    (2007)
  • D.N. Barton et al.

    Bayesian networks in environmental and resource management

    Integrated Environmental Assessment and Management

    (2012)
  • Bay Area Stormwater Management Agencies Association (BASMAA)

    San Francisco Bay area Stormwater trash generation rates. Final technical report

  • Bay Area Stormwater Management Agencies Association (BASMAA)

    Tracking California's trash project: Evaluation of the on-land visual assessment protocol as a method to establish baseline levels of trash and detect improvements in Stormwater quality. State water resources control board grant agreement no. 12-420-550

    (2016)
  • Bay Area Stormwater Management Agencies Association (BASMAA)

    Tracking California's trash project: Testing trash “flux” monitoring methods in flowing water bodies

  • K.J. Beven

    Rainfall-runoff modelling: The primer

    (2001)
  • M. Blangiardo et al.

    Spatial and spatio-temporal Bayesian models with R-INLA

    (2015)
  • P. Brown

    Model-based geostatistics the easy way

    Journal of Statistical Software

    (2015)
  • Sam Francisco Regional Water Quality Control Board (SFRWQCB)

    Rapid trash assessment protocol: Surface water ambient monitoring program

  • City of Los Angeles

    Trash monitoring and reporting plan: Los Angeles River and Ballona Creek Watersheds

    (2016)
  • City of Los Angeles

    Clean streets LA clean streets index

  • J. Clark et al.

    Hierarchical modeling for the environmental sciences: Statistical methods and applications

    (2006)
  • N. Cressie

    Statistics for spatial data: Revised edition

    (1993)
  • N. Cressie et al.

    Accounting for uncertainty in ecological analysis: The strengths and limitations of hierarchical statistical modeling

    Ecological Applications

    (2009)
  • R. Day et al.

    The quantitative distribution and characteristics of Marine Debris in the North Pacific Ocean, 1984–1988

  • Susan L. Dautel

    Transoceanic trash: international and United States strategies for the great Pacific Garbage

    Patch. Golden Gate U. Envtl. LJ

    (2009)
  • F. De Santis

    Using historical data for Bayesian sample size determination

    Journal of the Royal Statistical Society: Series A (Statistics in Society)

    (2007)
  • C. DiMaggio

    Bayesian hierarchical approaches to spatial analysis of injury and disaster data

  • A. Ellison

    An introduction to Bayesian inference for ecological research and environmental decision-making

    Ecological Applications

    (1996)
  • A. Getis et al.

    The analysis of spatial association by use of distance statistics

    Geographical Analysis

    (1992)
  • J. Hammer et al.

    Plastics in the Marine environment: The dark side of a modern gift

    Reviews of Environmental Contamination and Toxicology

    (2012)
  • Hawaii Department of Health

    National Pollutant Discharge Elimination System (NPDES) permit

  • R.J. Hijmans et al.

    Raster: Geographic analysis and modeling with raster data. R Package Version 2.0–12. Package used August 2018

  • Cited by (6)

    • Using a deep learning model to quantify trash accumulation for cleaner urban stormwater

      2022, Computers, Environment and Urban Systems
      Citation Excerpt :

      Cost trade-offs associated with different experimental monitoring designs should be explored determine approaches that can best inform management questions (e.g., Wheeler & Knight, 2017). The greater frequency and spatial coverage of observations afforded by imagery interpreted the trash detection model is likely to result in greater statistical power for detecting changes over time and spatial patterns (Conley et al., 2019), but this should be tested explicitly. Future research should also include testing additional trash detection models across a wider array of urban environments, including the impact of parked cars, which obstruct trash from view of the camera.

    • Quantifying clogging patterns of infiltration systems to improve urban stormwater pollution reduction estimates

      2020, Water Research X
      Citation Excerpt :

      Of course, the sources of uncertainty already discussed propagate into the model outputs, which must be considered, if they are to be used as the basis for decision making. As the relationships identified in this work are further refined with additional measurements, they will be incorporated web-based stormwater planning software previously reported in Beck et al. (2017) and Conley et al. (2019) (www.2nform.com). The revised algorithms will allow the relevant parameters to be handled dynamically with local sediment delivery to BMPs modeled based on site specific characteristics, and BMP sizing and drainage defined as inputs via a user interface.

    • Modernizing stormwater programs through digital transformation

      2020, 93rd Water Environment Federation Technical Exhibition and Conference 2020, WEFTEC 2020
    View full text