Introduction

The Center for Conservation Innovation at Defenders of Wildlife has completed analyses quantifying whether a causal relationship can be identified between mountaintop removal mining activities in Appalachia and a degradation in water quality for aquatic species. We are interested in this relationship as it pertains to two intersecting environmental laws, the Endangered Species Act (ESA) and the Clean Water Act.

Under the ESA, federal agencies must ensure that their actions do not jeopardize the existence of species listed as threatened or endagered on the endagered species list (hereafter ‘listed species’). Often they accomplish this by consulting with the US Fish and Wildlife Service, which implements the ESA, on the impacts proposed projects and regulations pose to listed species. The permitting and regulation of surface mines in the United States is governed by the Surface Mining Control and Reclamation Act of 1977 (SMCRA). In 1996, the US Fish and Wildlife Service consulted on a set of regulations and procedures by which state regulatory authorities could issue mining permits under SMCRA. In the Service’s judgement, these procedures were sufficient to prevent the jeopardization of listed species potentially affected by mining activity.

The Clean Water Act is administered by the Environmental Protection Agency (EPA), and is the primary federal law governing water pollution. Among other regulations, the Clean Water Act requires states to develop water quality standards for waterways to meet different usage criteria, one being suitability for aquatic life. If these standards are consistently not met a waterway may be designated as impaired under section 303(d). The water quality standards thus represent an agreed upon set of thresholds necessary to maintain suitability of waterways for aquatic life.

This report aims to answer the question as to whether the creation and operation of mountaintop removal mines impair the conservation and recovery of aquatic listed species. We address this goal by analyzing three relationships:

  1. Relationship between mined area and observed values of different water quality measures.
  2. Relationship between mined area and the frequency that chronic thresholds for water quality were exceeded.
  3. Relationship between mined area and frequency of waterway impairment under the Clean Water Act.

Methods

Data

We obtained spatial data delineating the footprints of all large surface mines across central Appalachia in each year from 1985 to 2015. These data come from a previous analysis measuring trends in the extent of mining activities over time, conducted by a partner organization SkyTruth. In order to avoid commission errors in mine identification affecting subsequent analyses, we cross-referenced these footprints with each year of available data from the National Land Cover Dataset (NLCD), and eliminated all footprints that overlapped with areas flagged as agriculture or development

We used these mine footprints to define our study area as a contiguous selection of all the US counties overlapping these mines.

Fig. 1 Central Appalachia was covered in thousands of mountaintop removal mines as of 2015. This map displays the footprints of mined areas detected using satellite imagery in 2015. Equivalent data from each year from 1985 through 2015 were used in analyses.

For a clearer look at these data, the map below shows the mining footprint data for Boone county, WV in 1985, 1995, and 2005. These three years are shown for map rendering purposes, but data are present for each year 1985 - 2015.

Fig. 2 The extent of mountaintop removal mines change over time. An example of mining footprints measured in 1985, 1995, 2005, and 2015 in Boone county, WV are shown here illustrating changes in the extent of mined area.

We obtained measurements of water quality from the national water quality data portal using the dataRetrieval package for R. The national water quality data portal aggregates data from monitoring stations nationally, primarily from the US Geological Survey and the EPA, but also the National Park Service and other federal agencies. We selected a set of water quality measures related to mining activity that could also be important to aquatic species. Potentially important measures were determined by examining federal documents pertaining to listed species including listing decisions, recovery plans, and five-year reviews. Data from 1985 through 2015 were collected - corresponding to the same period over which mining footprint data was available - from all monitoring stations within the counties comprising our study area.

toxins <- c('Aluminum',
            'Arsenic',
            'Cadmium',
            'Calcium',
            'Conductivity',
            'Copper',
            'Flow rate, instantaneous',
            'Iron',
            'Lead',
            'Manganese',
            'Mercury',
            'pH',
            'Selenium',
            'Sulfate',
            'Temperature, water',
            'Turbidity',
            'Total dissolved solids',
            'Total suspended solids',
            'Zinc'
            )

toxinData <- readWQPdata(countycode = appCounties$portalCode,
                        startDate = '1985-01-01',
                        endDate = '2016-01-01',
                        siteType = 'Stream',
                        sampleMedia = 'Water',
                        characteristicName = toxins,
                        minresults = 1)

All measures of concentration were standardized to ug/L. Temperature was standardized to Celsius. Conductivity was standardized to uS/cm, and Turbidity to NTU. To account for potentially misrecorded data, we flagged as outliers any observations above the 99.9th percentile or below the 0.1th percentile for a given measure. These thresholds were determined empirically by plotting the number of observations falling outside successively larger quantiles and selecting the lowest value at which observations plateaued.

To identify acute and chronic toxicity thresholds for different measures, we used the state water quality standards administered by Virginia under the Clean Water Act. These standards are approved by the EPA and are used to determine waterway impairment. The Virginia standards were identical to those from Kentucky and West Virginia. For each measure we flagged any observation where levels exceeded either the acute or chronic exposure thresholds. Finally, we also obtained the locations of all waterways declared as ‘impaired’ under section 303(d) of the Clean Water Act occurring within our study area. Impaired waterway designations were made on biannual cycles beginning in 1991.

Finally, we obtained a list of all listed species occurring within the counties comprising our study area. We read all available listing decisions, five-year reviews, and recovery plans for these species to identify specific streams that are designated as important to the species survival or recovery. We determined streams were important to listed species if they are mentioned in these documents as containing extant populations or as necessary for recovery.

Spatial analyses

To assess the impacts of mining on water quality measures at downstream monitoring stations, we used the watershed and flow modeling tools from the pyshed package for Python to delineate the drainages that feed into each station. These analyses require a model of surface elevation as input, and we used a 30m digital elevation model provided by NASA. We then calculated the proportion of each drainage area covered by surface mines in each year from 1985 to 2015. We also created a many-to-one spatial join identifying which mines fell within the drainage basin of each monitoring station.

We used the annual Cropland Data Layers and NLCD data to estimate the proportion of these drainage basins that were covered by agriculture or impervious surface in each year. Cropland data were available annually beginning in 2009, and NLCD impervious surface data were available from 1992, 2001, 2004, 2006, 2008, 2011 and 2013. We interpolated between and extrapolated from these observed data points to estimate agricultural and impervious surface data for missing years between 1985 and 2015. We repeated these procedures to obtain the area mined, and covered by agriculture and impervious surface within a contiguous selection of watersheds overlapping the mine footprint data set. We used USGS HUC12 hydrologic units to represent watersheds.

An example drainage basin and corresponding monitoring site

An example drainage basin and corresponding monitoring site

Unless otherwise indicated, all spatial analyses were performed using the Google Earth Engine python API.

Statistical analyses

In all analyses estimating the relationship between mining and water quality measures, we attempted to account for attenuation in pollutant concentrations with increasing distance from mines. We adjusted the area of individual mine footprints within each drainage basin in proportion to the square root of the distance from the mine to the monitoring station. We refer to this measure as adjusted mined area.

Our first objective was to quantify the relationship between between values of each water quality measure and the proportion of drainage basins that were mined in a given year, while controlling for the amount of agriculture and impervious surface in the drainage basin. We specified linear mixed effects models with normal error distributions and random intercepts per year nested within monitoring sites. These models were used to estimate the increase or decrease in mean water quality measures as a function of adjusted mine area, percent agriculture, and percent impervious surface.

pollutants <- unique(cleanData$CharacteristicName)
pollutants <- pollutants[!pollutants %in% c('Temperature, water', 'Flow rate, instantaneous')]

make_df <- function(measure, type, cnt){
  df <- filter(cleanData, !outlier)%>%
    select(MonitoringLocationIdentifier, CharacteristicName, ResultSampleFractionText, convertedValue, outlier, year, chronic, acute)%>%
    left_join(cons_site_areas, by = c("MonitoringLocationIdentifier" = "Monitoring", "year" = "year"))%>%
    left_join(site_counts, by= c('MonitoringLocationIdentifier'))%>%
    filter(CharacteristicName == measure,
           ResultSampleFractionText == type,
           Area > 400,
           count > cnt,
           convertedValue > 0)%>%
    mutate(Year = as.character(year))
  return(df)
}

run_stan_model <- function(toxin, measure, cnt){
  model <- stan_lmer(
    data = make_df(toxin, measure, cnt),
    convertedValue ~ (1|MonitoringLocationIdentifier/Year) + Adj_pMine + pAg + pImperv
  )
  stats <- as.data.frame(summary(model,
                         pars = c("Adj_pMine", "pAg", "pImperv"),
                         probs = c(0.025, 0.50, 0.975),
                         digits = 3
  )
  )
  df <- bind_cols(data.frame(
                    Toxin = rep(toxin, 3),
                    Measure = rep(measure, 3),
                    Param = row.names(stats)
                  ),
                  stats
  )
  return(df)
}

stanAnalysis <- lapply(pollutants, function(x){
  lapply(c("Total", "Dissolved"), function(y){
    tryCatch(
      run_stan_model(x,y, 10),
      error = function(cond){
        return(c(x, y, rep(NA, 8)))
      }
    )
  })%>%bind_rows()
})%>%bind_rows()

Our second objective was to determine whether the proportion of a drainage basin that was mined in a given year affected the probability that pollutant levels would exceed thresholds deemed safe for aquatic life. We specified a generalized linear mixed model using a binomial error distribution and logit link, with random intercepts per year nested within monitoring site. These models were used to predict the probability that an observed value for a given measure would exceed chronic thresholds as a function of adjusted mined area, percent agriculture, and percent impervious surface.

Our last objective was to estimate the relationship between the proportion of a watershed that was mined and the probability that a waterway therein would be designated as 303(d) impaired. We specified a generalized linear mixed model with a logit link and binomial error distribution with random intercepts per year nested within monitoring stations. These models were used to predict the probability that a watershed would contain an impaired waterway at the end of a given biannual evaluation cycle as a function of adjusted mined area, percent agriculture, and percent impervious. Because impaired waterways are only tallied during biannual cycles, we used the maximum percent agriculture and impervious surface within two-year cycles as predictor variables. To account for lagged and cumulative effects, we used the cumulative sum of percent mined area within drainage basins over time.

impairModel <- stan_glmer(data = impairData,
                  YN ~ (1|HUC12) + maxSum + maxImperv + maxAg,
                  family = binomial(link = 'logit'))

In all regression analyses we included only sites with at least 10 observations. We estimated model parameters in a Bayesian framework using the rstanarm package for R. For each model we generated four MCMC chains and tested for convergence using the Rhat statistic. A significant relationship was determined between mined area and response variables if the 95% credible interval around the relevant parameter estimate did not overlap zero.

Results

We obtained water quality data from 4260 different water quality monitoring sites across our study area. The number of observations at each site ranged from 1 to 275

Fig. 3 Water quality monitoring stations across central Appalachia recorded different amounts of data. Circles on the map indicate the location of water quality monitoring stations from which data was available. Circle size is proportional to the number of observations recorded at a given stations between 1985 and 2015.

55 listed aquatic species potentially occur within the counties comprising our study area. These include 39 mollusk, 12 fish, 3 crustacean, and 1 snail species. Of these listed species, 15 had designated critical habitat. Additionally, for 50 of these species we were able to identify specific streams that were important to the species survival and recovery in either listing decisions, five-year reviews, or recovery plans. Chronic and acute toxicity thresholds were exceeded thousands of times in drainage basins containing streams important to listed species conservation and recovery. The most commonly exceeded threshold was that for pH, followed by Manganese, Aluminum, and Conductivity.

Models indicated significant positive relationships between the proportion of a drainage basin that is mined and levels of Conductivity, Manganese, Sulfate, Sulfur, Total Dissolved Solid, Total Suspended Solids, and Zinc. No measures were significantly negatively associated with adjusted mined area.

Fig.4 Increases in the proportion of drainage basins that were mined lead to increases in multiple measures of water quality that are detrimental to aquatic species. Graphs show the change in water quality measures per change in mined area as estimated by linear mixed models. Dashed lines encompas the 95% credible interval around estimated relationships.

Measures of conductivity all exceeded chronic exposure thresholds, and we were unable to model probability of exceeding thresholds. Models indicated that probabilities for Copper, Lead, Manganese were all positively related to the area mined within drainage basins. No water quality measures exhibited a significant negative relationship.

Fig.5 Increases in the proprotion of drainage basins that were mined increased the probability that chronic exposure toxicity thresholds would be exceeded for three water quality measures. Graphs show the change in probability of exceedence per change in mined area as estimated by linear mixed models. Dashed lines encompas the 95% credible interval around estimated relationships.

We also found significant positive relationships between the cumulative proportion of a watershed that was mined over time and the log odds that a stream in that watershed would be designated impaired.

Discussion

Our results demonstrate that surface mining likely degrades water quality in ways that affect the survival and recovery of listed species. We found consistent evidence linking changes in mined area and increases in concentrations of pollutants, conductivity, and dissolved and suspended solids. Far from being innocuous side effects, the measures most strongly affected by mining were some of those that are directly important to the survival of aquatic species. We found substantial increases in stream conductivity, and the concentrations of Manganese, Sulfate, Sulfur, Zinc, dissolved and suspended solids as a result of increases in the size of areas that are mined. These results establish a direct relationship between mined area and degradation in water quality and habitat suitability. Furthermore, we found evidence that this degradation can progress to potentially lethal levels for aquatic species. We found a significant relationship between the probability that scientifically determined and federally regulated thresholds for acceptable Copper, Lead, and Manganese concentrations would be exceeded and the expansion of mined area. The potential impact of these events to listed species is illustrated by the frequency with which these events occurred in watersheds containing streams designated by the US Fish and Wildlife Service as important to listed species survival and recovery. Finally, these processes likely contribute to the emergent outcome that a waterway was more likely to fail to meet water quality standards and be declared impaired under the Clean Water Act as the proportion of it’s watershed that was mined increased. A large body of previous ecological and hydrologic research has shown that surface mining can negatively impact water quality, and reduce the suitability of streams for aquatic species. The findings presented here illustrate that these effects are not limited only to sites immediately proximate to mines, but at a landscape scale. Our results indicate that, in situ the growth and continued operation of surface mines in Appalachia directly limits the prospects for survival and recovery of over 50 listed species.