search
Contact Us
HOME

  • Crossref logo
Journal Search Engine

to

Search Advanced Search Adode Reader(link)
Download PDF Export Citaion korean bibliography PMC previewer
ISSN : 1225-1011(Print)
ISSN : 2288-1727(Online)
The Journal of Fisheries Business Administration Vol.57 No.1 pp.23-32
DOI : https://doi.org/10.12939/FBA.2026.57.1.023

The Impact of the Inland Total Maximum Daily Load Program: Evidence from South Korea

Tae-Hyun Kim*
*Associate Research Fellow, Department of Food & Agriculture Economic Research, Korea Rural Economic Institute, Naju-si, Jellanam-do, 58321, Rep. of Korea
* Corresponding author : https://orcid.org/0000-0003-2142-6791, +82-61-820-2396, taekim@krei.re.kr
10/03/2026 ; 18/03/2026 ; 19/03/2026

Abstract


Nutrient loads in inland rivers can propagate downstream to estuaries and near-coastal waters, where water-quality conditions may affect fisheries and aquaculture. This study examines the effect of a nutrient load control program on total phosphorus in South Korea. Using a difference-in-differences approach with fixed effect models, the study constructs monitoring stations in the Nakdong, Geum, and Yeongsan rivers as the treatment group and monitoring stations in the Han River as the control group. The study leverages total phosphorus (TP) control as the policy intervention. Using monitor-by-time data from the national Total Maximum Daily Load network from 2007 to 2012, the study found that TP levels decreased by approximately 3.6% after the policy intervention. The study further examines flexible and honest DIDs and finds limited but suggestive evidence of this improvement. The findings provide suggestive evidence of load-based regulation in reducing TP and provide policy-relevant evidence on upstream nutrient control that may inform broader downstream water-quality management in connected river-estuary systems.


Total Maximum Daily Load Program , Total Phosphorus , Difference-in-differences , South Korea

수질오염총량제의 영향 분석

김태현*
*한국농촌경제연구원 부연구위원

초록

    I. Introduction

    Water quality policy protects ecosystems and supports the sustainable development of regional economies. In 2004, South Korea initiated an inland Total Maximum Daily Load (TMDL) program in the Nakdong, Geum, and Yeongsan rivers. The program sets target water quality levels for each river and then calculates the total pollutant load that can be discharged into the water without exceeding those targets. The program does not focus solely on reducing discharges; rather, it allows development that remains within the assimilative capacity set up by the targets. The program has evolved over time: Phase 1 emphasized point-source control of biochemical oxygen demand (BOD) in the three rivers, and Phase 2 added total phosphorus (TP) control and later extended coverage to the Han river. This inland regulation also has economic relevance for fisheries and aquaculture, as production and revenues in inland and coastal areas depend on nutrient loads carried downstream through river networks (Baek et al., 2025).

    Phosphorus has increasingly emerged as the binding nutrient in Korean rivers (Cho et al., 2024, 2023; Ly et al., 2021). TP is not only a primary pollutant responsible for eutrophication and harmful algal blooms in upstream areas, but also a contaminant that flows downstream, affecting estuaries and coastal waters (Lee and Kim, 2007;Lefcheck et al., 2018). This downstream transport is further intensified as monsoonal and typhoon rains mobilize phosphorus from soils and sediments, and the subsequent low-flow season limits dilution (Baek et al., 2025), amplifying nonpoint inputs. Point-source oriented BOD controls have limitations in addressing these dynamics. In response, phase 2 incorporated explicit TP load limits alongside BOD to mitigate algal bloom and hypoxia risks and to better align inland management with downstream concerns in estuaries. From a fisheries-management perspective, these downstream stressors can translate into production losses and higher operating costs in aquaculture-intensive embayments, making upstream phosphorus control potentially relevant for coastal fisheries management (Huang et al., 2010;Martino et al., 2020).

    Government reports and a large number of studies in environmental-engineering fields (Jung et al., 2016;Kim et al., 2014;Park et al., 2011) document the water quality improvements under the program. Much of this evidence relies on pre-post comparisons - simply comparing water quality measures before and after the policy within a single river or a small set of monitoring stations. While the pre-post comparison has helped establish an initial picture of water quality improvement, it lacks an explicit counterfactual and can be sensitive to contemporaneous external shocks - for example, expansions in treatment capacity, land-use change, or macroeconomic conditions - making the estimates vulnerable to bias and confounding. For fisheries and aquaculture policy, this distinction matters because biased evaluation of upstream controls can misstate expected reductions in downstream eutrophication and hypoxia risk. That, in turn, can distort where agencies allocate scarce resources-for example, across upstream nonpoint investments versus estuarine monitoring, emergency response, or aquaculture management.

    Thus, this study estimates the effect of inland TMDL policy on TP levels by comparing a treatment group with a plausibly counterfactual control. The study uses a monitor-by-time panel from January 2007 to May 2013 and estimates difference-in-differences (DID) models. To mitigate confounding noted in prior engineering studies, this study exploits the panel structure with a suite of fixed effects that absorb time-invariant heterogeneity and common shocks. This study further examines the timing of effects using a flexible DID and reports robustness using honest DID, which remains valid under limited pre-trend deviations (Rambachan and Roth, 2023;Roth, 2022). The results provide limited but suggestive evidence that TP levels fell by approximately 3.6% in the treated rivers relative to the Han river. Flexible DID estimates indicate effects become statistically distinguishable from zero roughly six months after the policy intervention, and honest DID confidence bands remain broadly consistent with modest reductions under relaxed parallel-trend assumptions.

    This study contributes to several strands of literature. First, this study provides nation-wide, monitor-level evidence on inland TMDL effects on TP. While previous studies in South Korea typically investigated a single river or a small number of monitoring stations (Kim et al., 2014;Park et al., 2011), this study examines the policy's effect on water quality across all four major rivers. Second, this study adapts DID methods widely used in environmental and resource economics for high-frequency, multi-period settings (Callaway and Sant'Anna, 2021; Christensen et al., 2023;Keiser and Shapiro, 2019). Rather than relying on simple pre-post comparisons, it leverages cross-river variation and policy timing to construct a plausible counterfactual. While there are limited studies applying quasi-experimental approaches to water quality policy in South Korea (Kim, 2024;Ma, 2018), Ma (2018) relied on annual-level data and did not report the parallel-trend assumption, limiting causal interpretation. This study builds on Kim (2024), who applied DID to assess a coastal water quality management program, and extends the approach to inland TMDL. Third, building on concerns raised in government reports and engineering studies that simple comparisons overlook key hydrological factors, this study incorporates covariates such as water temperature and streamflow, along with a suite of fixed effects, to improve estimation reliability. These adjustments help isolate the average treatment effect by accounting for confounding hydrological variation and improving model robustness.

    The remainder of the paper is organized as follows. Section 2 presents the institutional background and outlines empirical strategy. Section 3 describes the data and reports the main findings. Section 4 discusses policy implications and concludes.

    Ⅱ. Policy Background1)

    The inland TMDL program caps total daily pollutant loads to protect water quality. It addresses a well-documented limitation of concentration standards: when wastewater volumes rise with economic activity, total loads can increase and degrade receiving waters even if effluent concentrations satisfy discharge limits. The TMDL program addresses this limitation by setting river-specific targets, estimating allowable loads with hydrologic and water quality models, and allocating those loads to unit watersheds and municipalities along each river. The government implemented the program in stages. Phase 1 (2004-2010) emphasized point-source control of BOD. In Phase 2 (2011-2015), policymakers added TP controls and later extended mandatory TMDL coverage to the Han river in June 2013 after earlier pilot efforts near Paldang Lake (Hankyoreh Media Group, 2010).

    The TMDL program is administered by the central government. Ministry of Climate, Energy and Environment sets river-specific water quality targets. Then, local governments prepare a ten-year masterplan that estimates allowable and assigned loads combining with development and reduction plans. Next, municipalities develop annual implementation plans that allocate assigned loads across major source categories - e.g., residential, livestock, land, industry, landfill, and aquaculture. Finally, the ministry performs annual implementation reviews; when targets are missed or assigned loads are exceeded, it can restrict development approvals, issue compliance orders, or adjust fiscal support (Ministry of Climate, Energy and Environment 2012; Chungbuk Daily, 2012).

    Program evaluations span multiple dimensions: trends in pollutant loads, compliance with assigned loads, observed shifts in water quality, and the effectiveness of point- and nonpoint-source management. This paper only focuses on observed shifts in ambient quality because the pre-post comparisons commonly used in practice can yield biased estimates without an explicit counterfactual.

    Ⅲ. Empirical strategy

    This study constructs a counterfactual for treated monitoring stations-the TP levels they would have been absent the Phase 2 TP regulation-by using contemporaneous changes at control monitoring stations (Angrist and Pischke 2009; Gertler et al., 2016). Equation (1) represents the difference-in-differences (DID) model.

    arc ( T P n t ) = β 0 + β 1 ( T r e a t n × P o s t t ) + γ n + δ y + δ m + Z n t ′ θ + ϵ n t
    (1)

    T p n t is total phosphorus at monitor n in period t. This study applies the inverse hyperbolic sine to accommodate a skewed distribution with many small values while preserving a log-like interpretation for positive values (Kim, 2024).

    As noted above, this study utilizes variation in the timing of TP regulation implementation under Korea's Phase II TMDL as a quasi-experimental source of identification. TP controls tightened in the Nakdong, Geum, and Yeongsan rivers starting in 2011, whereas the Han River adopted mandatory TP-based TMDL in 2013. Accordingly, this paper defines monitoring stations in Nakdong, Geum, and Yeongsan as the treated group and stations in Han as the control group over the study period.

    Thus, T r e a t n = 1 represents monitoring stations in the Nakdong, Geum, and Yeongsan rivers, and 0 represents monitoring stations in the Han river. P o s t t = 1 represents periods after the inland TP regulation begins (January 2011) and 0 otherwise. The parameter of interest, β 1 , measures the average treatment effect relative to the Han river control.

    Additionally, this study includes covariates that plausibly affect TP levels: water temperature and streamflow, which relate to eutrophication dynamics, nonpoint runoff, and algal activity (Kim et al., 2014;Park et al., 2011). Monitor fixed effects ( γ n ) absorb time-invariant monitor heterogeneity (e.g., geomorphology, longitudinal position, long-run land cover). Year ( δ y ) and month ( δ m ) fixed effects absorb common macro, policy, and seasonal shocks. The following table reports specifications that vary the fixed-effects structure as sensitivity checks. The standard errors are clustered at the monitor level to allow arbitrary serial correlation and heteroskedasticity within monitoring stations.

    To verify the parallel-trends assumption and the timing of the policy response (Christensen et al., 2023), this study estimates a flexible DID that replaces P o s t t with leads and lags relative to the policy intervention date. 1 [ t = τ ] indicate month τ (with January 2011 as the reference). This flexible DID traces pre-policy differences and post-policy adjustment over time as shown in Equation (2).

    arc ( T P n t ) = ∑ τ μ τ 1 [ t = τ ] + ∑ n = τ γ τ ( 1 [ t = τ ] × T r e a t n ) + Z n t ′ θ + δ n + ϵ n t
    (2)

    When the parallel-trends assumption is violated, the conventional difference-in-differences estimator may yield biased results. To address this limitation, the study employs the honest difference-in-differences framework. This approach constructs confidence intervals that remain robust even under modest violations of the identifying assumptions, thereby enhancing the credibility of causal estimates. By incorporating honest DID alongside the event-study specification, the analysis ensures that the estimated treatment effects are not only flexible across time but also statistically reliable under weaker assumptions (Rambachan and Roth, 2023).

    Ⅳ. Data and results

    1. Data

    This study uses a publicly available dataset from Korea's Water Environment Information System. Each monitor reports approximately 35 observations per year. <Table 1> presents descriptive statistics. This study excludes missing values and extreme upper-tail outliers, which account for 1.52% of TP, 6.31% of streamflow, and 1.50% of water temperature readings. After these exclusions, the dataset contains 47,224 observations for TP, 45,672 observations for streamflow, and 47,222 observations for water temperature.

    <Table 2> shows the number of water quality monitoring stations. Specifically, the data covers four rivers with independent TMDL monitoring stations: Han (62), Nakdong (93), Geum (59), and Yeongsan (57). The final sample includes 14,874 control observations and 33,057 treated observations.

    2. Results

    <Table 3> reports estimated coefficients across specifications that vary covariates and fixed effects. Models 1-3 exclude covariates; Models 4-6 add temperature and streamflow. Models 1 and 4 include monitoring station, year, and month fixed effects, which absorb time-invariant monitoring station characteristics and shocks common to all monitoring stations within a given year or month (e.g. macro conditions and seasonality). Models 2 and 5 replace separate year and month effects with year-by-month fixed effects alongside monitoring station fixed effects, thereby absorbing shocks specific to a particular year-by-month (e.g., July 2011). Models 3 and 6 include monitoring station-by-month fixed effects, which absorb monitoring station-specific seasonal patterns (e.g., monitoring station A's typical July conditions) and thus most tightly control for spatially heterogeneous seasonality among the three blocks of models.

    Only Models 3 and 6 found a statistically significant average treatment effect of the TMDL program on TP. Specifically, Phase 2 TMDL is associated with an average reduction in TP of approximately 3.6% in treated rivers relative to the control river. Estimates in the less demanding fixed-effect designs are smaller in magnitude and statistically indistinguishable from zero.

    This pattern reflects how fixed-effects structures shape identifying variation: station-by-month effects absorb station-specific seasonal cycles while preserving cross-basin policy contrasts, whereas coarser time effects risk residual seasonal confounding or over-absorbing common pre-post variation, reducing precision in treatment estimates. Accordingly, this paper interprets the 3.6% estimate as suggestive evidence rather than a definitive causal magnitude.

    Difference-in-differences model requires the parallel-trend assumption: treated and control groups should follow similar pre-policy trend (Bueno et al., 2019;Roth, 2022). This study utilizes equation 2 and plots on the left panel of <Fig. 1>, providing point estimates (dots connected by a blue line) with 95% confidence intervals (gray dashed bands). Although several pre-treatment coefficients differ from zero, indicating imperfect parallel trends, effects become statistically distinguishable from zero roughly six months after the policy intervention, which suggests a short lag before TP responds.

    To further address the violation of parallel trend assumption, the study estimates honest DID, which delivers valid confidence bands under bounded deviations from parallel trends (Rambachan and Roth, 2023;Christensen et al., 2023). The right panel of <Fig. 1> overlays the flexible DID point estimates (black line with dots) and 95% confidence intervals (blue dashed) with honest DID bands in red. Although some post-period upper bounds overlap zero, most post-period intervals fall on the improvement side, indicating that - under plausible relaxations - the direction of the TP effect remains broadly consistent.

    Ⅴ. Conclusion

    This paper estimates the effect of Korea's TMDL policy on total phosphorus. To mitigate bias in simple pre-post comparisons, this study compares treated rivers with a plausibly counterfactual control river using a difference-in-differences design with fixed effects. The results suggest that the Phase 2 TMDL program improves TP levels by approximately 3.6% relative to the control. A flexible DID indicates a policy lag of about six months, and honest DID confidence bands remain broadly consistent with post-policy improvements. Because inland nutrient conditions can translate into downstream water-quality risks that affect fisheries and aquaculture operations, estimating upstream policy impacts provides relevant evidence for fisheries management and related economic decisions.

    The results suggest that load-based regulation can deliver short-run reductions in TP. Policymakers could maintain the load-based framework while prioritizing non-point sources, which have become increasingly important in watershed pollution control (US EPA, 2007). Because non-point loads rise with rainfall and discharge - through stormwater, agricultural runoff, and urban impervious surface - effective implementation typically requires targeted measures such as infiltration and retention infrastructure and agricultural nutrient BMPs (Muhammetoglu et al., 2025). Given that upstream nutrient loads propagate to estuaries and coastal waters, aligning indicators, evaluation protocols, and operations across inland and coastal TMDL programs could strengthen integrated watershed-coastal management, particularly in aquaculture-intensive, semi-closed embayments where long residence times and stratification heighten eutrophication risks. Prior work links eutrophication-related hypoxia and harmful algal blooms to economic impacts in fisheries and aquaculture (Huang et al., 2010;Martino et al., 2020), including production disruptions and higher operating costs (e.g., oxygenation, intensified monitoring/mitigation, and adaptive harvesting). In this sense, the estimated upstream TP response provides an empirical starting point for future work that quantifies downstream economic benefits. Specifically, future research could link station-level TP changes to estuarine exposure measures and merge these with fisheries and aquaculture production and revenue data to explore whether upstream load controls are associated with economically meaningful reductions in downstream risk in coastal systems.

    This study has several limitations. The estimated average treatment effects become sensitive when including year fixed effects (and high-dimensional variants such as monitor×year), often losing significance and occasionally changing sign. Two factors likely contribute: concurrent changes around the policy window (other regulations, facility investments, and hydrologic shocks) that are difficult to fully separate, and a short post-period that leaves limited residual variation once fine time effects absorb common shocks. This study therefore interprets the 3.6% reduction as suggestive rather than a definitive causal magnitude. This study reports alternative fixed-effect specifications as sensitivity checks and recommends cautious interpretation. In addition, this study does not estimate downstream economic outcomes such as fisheries or aquaculture production and revenues; instead, this study focuses on the policy's first-order effect on inland TP concentrations.

    Despite these caveats, this paper moves beyond monitoring station-specific pre-post audits and provides quasi-experimental evidence consistent with TP improvements under the TMDL. By combining baseline DID with flexible and honest DIDs, the analysis provides suggestive evidence of the TMDL policy's impact at the national-scale. Future research could quantify how upstream TP changes translate into fisheries and aquaculture outcomes by linking riverine nutrient dynamics to estuarine exposure and incorporating additional outcome data.

    Figures

    FBA-57-1-23_F1.jpg
    Flexible DID (left panel) and honest DID (right panel)

    Tables

    Descriptive statistics
    The number of water quality monitoring stations
    Average treatment effect of TMDL policy
    Note: The dependent variable is inverse hyperbolic sine transformed total phosphorus level at the monitoring station level. Models 1-3 exclude covariates, while models 4-6 include flow and water temperature as covariates. Models 1 and 4 include monitor, year, and month fixed-effects, models 2 and 5 include monitor and year-by-month fixed-effects, and models 3 and 6 include monitor-month fixed-effects. Standard errors are clustered at the monitoring station level.
    *p-value < 0.1, **p-value < 0.05, ***p-value < 0.01.

    References

    1. Baek, S. H., Lee, C. H., Kim, M., Hong, S. and Lim, Y. K. ( 2025), “Seasonal effects of Nakdong River freshwater inflow and coastal environmental changes on phytoplankton community structure, including harmful species, in eastern Jinhae Bay, Korea”, Journal of Marine Science and Engineering, 13(4), 669.
    2. Bueno, M. and Valento, M. ( 2019), “The effects of pricing waste generation: A synthetic control approach”, Journal of Environmental Economics and Management, 96, 274-285.
    3. Callaway, B. and Sant’Anna, P. H. C. ( 2021), “Difference-in-differences with multiple time periods”, Journal of Econometrics, 225(2), 200-230.
    4. Cho, Y. C., Im, J. K., Han, J., Kim, S. H., Kang, T. and Lee, S. ( 2023), “Comprehensive water quality assessment using Korean water quality indices and multivariate statistical techniques for sustainable water management of the Paldang Reservoir, South Korea”, Water, 15(3), 509.
    5. Cho, Y. C., Kang, H. Y., Son, J. Y., Kang, T. and Im, J. K. ( 2024), “The spatiotemporal eutrophication status and trends in the Paldang Reservoir, Republic of Korea”, Sustainability, 16(1), 373.
    6. Christensen, P., Keiser, D. A. and Lade, G. E. ( 2023), “Economic effects of environmental crises: Evidence from Flint, Michigan”, American Economic Journal: Economic Policy, 15(1), 196-232.
    7. Chungbuk Daily ( 2012), “Water pollution load shock: What were local governments doing?” February 27 [available at http://www.inews365.com/news/article.html?no=220601].
    8. Hankyoreh Media Group ( 2010), “Han river, ‘mandatory TMDL program’ from 2013,” June 2 [available at https://www.hani.co.kr/arti/society/environment/423767.html].
    9. Huang, L., Smith, M. D. and Craig, J. K. ( 2010), “Quantifying the Economic Effects of Hypoxia on a Fishery for Brown Shrimp Farfantepenaeus aztecus”, Marine and Coastal Fisheries, 2(1), 232-248.
    10. Jung, K., Lee, I., Lee, K., Cheon, S., Hong, J. and Ahn, J. ( 2016), “Long-term trend analysis and exploratory data analysis of Geumho River based on seasonal Mann-Kendall test”, Journal of Environmental Science International, 25(2), 217-229.
    11. Keiser, D. A. and Shapiro, J. S. ( 2019), “Consequences of the Clean Water Act and the demand for water quality”, Quarterly Journal of Economics, 134(1), 349-396.
    12. Kim, E., Kim, Y., Rhew, D., Ryu, J. and Park, B. K. ( 2014), “A study on the water quality changes of TMDL unit watershed in Geum River Basin using a nonparametric trend analysis”, Journal of Korean Society on Water Environment, 30(2), 148-158.
    13. Kim, T. H. ( 2024), “The impact of a coastal water quality policy in South Korea: Evidence from the total pollution load management program”, Water Economics and Policy, 10(3), 2450009.
    14. Korea Law Information Center ( 2025), Ministry of Government Legislation, accessed 1 October 2025 [available at https://www.law.go.kr/main.html].
    15. Korea’s Water Environment Information System ( 2025), Ministry of Climate, Energy and Environment, accessed 8 September 2025 [available at https://water.nier.go.kr/web].
    16. Lee, D. I. and Kim, J. K. ( 2007), “Estimation of total allowable pollutant loads using eco-hydrodynamic modeling for water quality management on the southern coast of Korea”, Journal of the Korean Society for Marine Environmental Engineering, 10(1), 29-43.
    17. Lefcheck, J. S., Orth, R. J., Dennison, W. C., Wilcox, D. J., Murphy, R. R., Keisman, J., Gurbisz, C., Hannam, M., Landry, J. B., Moore, K. A., Patrick, C. J., Testa, J., Weller, D. E. and Batiuk, R. A. ( 2018), “Long-term nutrient reductions lead to the unprecedented recovery of a temperate coastal region”, Proceedings of the National Academy of Sciences of the United States of America, 115(14), 3658-3662.
    18. Ly, Q. V., Nguyen, X. C., Lê, N. C., Truong, T. D., Hoang, T. H. T., Park, T. J., Maqbool, T., Pyo, J., Cho, K. H., Lee, K. S. and Hur, J. ( 2021), “Application of machine learning for eutrophication analysis and algal bloom prediction in an urban river: A 10-year study of the Han River, South Korea”, Science of the Total Environment, 797, 149040.
    19. Ma, J. ( 2018), The effect of the total maximum daily loads (TMDL) management implementation plan on total phosphorus (TP), Master’s thesis, University of Kentucky, Lexington, KY, USA.
    20. Martino S., Gianella F. and Davidson K. ( 2020), “An approach for evaluating the economic impacts of harmful algal blooms: The effects of blooms of toxic Dinophysis spp. on the productivity of Scottish shellfish farms,” Harmful Algae, 99.
    21. Ministry of Climate, Energy and Environment, Republic of Korea ( 2012), “Assessment of water pollution load management underway amid local government feedback,” accessed 3 November 2025 [available at https://www.korea.kr/docViewer/iframe_skin/doc.html?fn=42a2b8d671b9ad04001876dab00ea52d].
    22. Muhammetoglu, A., Akdegirmen, O., Dugan, S. T. and Orhan, P. ( 2025), “A modeling framework for control of nonpoint source pollution and evaluation of best management practices for identification of critical source areas”, Environmental Earth Sciences, 84(10), 257.
    23. Park, J. H., Ryu, D. H. and Jung, D. I. ( 2011), “Water quality status of the unit watersheds in the Yeongsang/Seomjin River Basin since the management of total maximum daily loads”, Journal of Korean Society on Water Environment, 27(6), 719-728.
    24. Rambachan, A. and Roth, J. ( 2023), “A more credible approach to parallel trends”, Review of Economic Studies, 90(5), 2555-2591.
    25. Roth, J. ( 2022), “Pretest with caution: Event-study estimates after testing for parallel trends”, American Economic Review: Insights, 4(3), 305-322.
    26. United States Environmental Protection Agency ( 2007), Developing effective nonpoint source TMDLs: An evaluation of the TMDL development process, accessed 1 November 2025 [available at https://www.epa.gov/sites/default/files/2015-10/documents/developing-effective-nonpoint-source-tmdls.pdf].
    Copyright © The Korean Society of Fisheries Business Administration. All rights reserved.
    Pukyoung Univ.,599-1, Daeyeon 3-dong, Nam-gu, Busan, Korea.