Precipitation Patterns, Mountaintop Removal Mining, and the July 2022 North Fork Kentucky River Flood

Intense rains during the last week of July 2022 led to deadly flooding and widespread landslides in eastern Kentucky. As many as 45 deaths were attributed to the flooding (one body was never found), more than 600 helicopter rescues and countless swift-water rescues were performed, nearly 9,000 homes housing 22,000 people were destroyed or damaged, and 13 of Kentucky’s 120 counties were declared Federal Emergency Management Agency disaster areas (Dixon and Shelton, 2023; Klesta, 2023). The Kentucky Geological Survey documented nearly 1,100 new landslides and debris flows visible from public roads during the weeks after the flooding (Crawford et al., 2023), and many more are likely to have occurred in inaccessible areas. Estimates of the cost to rebuild the damaged homes range from $450 million to $950 million, depending on the number of homes relocated to less flood-prone areas (Dixon and Shelton, 2023). As author Silas House (House, 2023) wrote about an Appalachian flood from his youth in a poem composed for the 2023 Kentucky gubernatorial inauguration, “All the creeks conspired to the raging river.”

As Kentucky’s state geologist and director of the Kentucky Geological Survey at the time of the flood, I responded to dozens of requests for interviews from local, national, and international journalists. Common questions emerged from almost all interviews: What makes eastern Kentucky so prone to flooding? Was it a 1,000-year flood? Did the region’s legacy of surface coal mining, especially the geomorphologically destructive practice of mountaintop removal mining, increase the severity of the floods? Did climate change play a role? This paper offers some answers and suggestions for future research.

The purpose of this paper is threefold. First, the paper provides a general summary of the July 2022 rainfall event and flood along the North Fork Kentucky River (referred to as simply the North Fork throughout the rest of the paper). Second, the paper explains the disproportionate amount of damage and loss of life along Troublesome Creek, a tributary of the North Fork, using publicly available rainfall and airborne lidar elevation data as input for geographic information system (GIS)–based volumetric flux modeling. Third, the paper evaluates the potential effects of surface coal mines on flood severity using differences in end-member hydrologic models to simulate the effects of increased runoff from areas disturbed by surface coal mining.

The study area described in this paper is the watershed of the North Fork above the U.S. Geological Survey stream gage at Jackson, Kentucky, including upstream gages and subwatersheds at Hazard and Whitesburg (Figure 1). The U.S. Geological Survey gage numbers for the three are 03280000 (Jackson), 03277500 (Hazard), and 03277300 (Whitesburg), and those watersheds drain areas of 2,852, 1,207, and 172 km², respectively (U.S. Geological Survey, 2024). The only major flood control feature in the study area is Carr Creek Lake, which is impounded by a dam completed in 1976 and lies approximately 26 km upstream from Hazard. The weather station in Jackson receives an annual average of 131 cm of precipitation and a July monthly average of 13 cm of precipitation (National Weather Service, 2024a). Gebert et al. (1987) estimated the average annual runoff in the area to be approximately 51 cm/yr, about 39 percent of the average annual precipitation at Jackson, during a 1951–1980 study period.

Ground surface elevations in the study area range from 213 to 975 m above sea level (Figure 2A) and calculated slope angles range up to 69°, with remarkably large flat areas where the natural topography has been altered by mountaintop removal mining (Figure 2B and C; also see Pericak et al., 2018; SkyTruth, 2023).

Except for a narrow strip of southeasterly dipping Mississippian through Pennsylvanian beds along Pine Mountain and adjacent to the Pine Mountain thrust, which marks the southeastern boundary of the study area, the bedrock geology of the study area consists of flat-lying Middle Pennsylvanian strata of the Appalachian Plateau incised by a dendritic network of deep stream valleys and subsidiary unchanneled hollows (McFarlan, 1943; Rice and Wolcott, 1973; McDowell, 1986; and Noger, 1988). Quaternary alluvium has been mapped along the bottoms of larger stream valleys, but published geologic quadrangle maps covering the area are not detailed enough to show features such as individual stream terraces and alluvial fans that might be significant in flood hazard studies (e.g., Rice and Wolcott, 1973).

The rainfall that gave rise to the July 2022 eastern Kentucky flood resulted from the intersection of (1) moisture-laden air brought north from the Gulf of Mexico by strong low-level winds, (2) a stationary front over southeastern Kentucky, and (3) strong upper-level jet stream winds north of the front; the combination repeatedly generated thunderstorms that remained in place for several hours at a time over several days (Day and Chen, 2023; National Weather Service, 2024a). Such events are common in the midcontinental United States, although they occur more commonly during spring than summer, and are known as warm season elevated thunderstorms (Moore et al., 2003).

In contrast to its annual and July monthly averages of 131 and 13 cm, respectively, the weather station at Jackson received 38 cm of rain in July 2022 (National Weather Service, 2024a). Most of the precipitation fell during a 4-day period in the last week of the month. A map of July 26–29, 2022, precipitation totals across the study area from National Centers for Environmental Prediction Environmental Modeling Center (NCEP/EMC) 4-km-gridded (GRIB) stage IV precipitation data (Du, 2011) shows that the amount of rainfall during those days varied considerably in space and ranged over a factor of 4 across the study area (Figure 2D). The heaviest rainfall was concentrated in an area just outside of the North Fork watershed and is not shown in Figure 2D. July 26–29 precipitation totals within the North Fork watershed—the focus of this paper—ranged from just over 8 cm in the far southern portion to nearly 30 cm over portions of the Troublesome Creek watershed, the headwaters of the Whitesburg watershed, east of Jackson, and along the North Fork between Jackson and Hazard. Historic rainfall records were broken at several weather stations in the area, and the recurrence interval for the 4-day rainfall total exceeded 1,000 years (National Weather Service, 2024a).

Figure 3 is a set of flood discharge hydrographs plotted using data from the U.S. Geological Survey (2024) from the Jackson, Hazard, and Whitesburg gages for the period of July 25–31, 2022. The Whitesburg gage was damaged during the flood but recorded data intermittently up to and past the peak discharge; its hydrograph ends midday on July 29 because a new stage-discharge rating curve is being developed for the site. Although it contains the smallest of the three peaks, the Whitesburg hydrograph is remarkable because its 2022 flood peak stage of 6.64 m and discharge of 278 m³/s represent increases of 48 and 27 percent, respectively, over the previously recorded peak annual stage and discharge records of 4.48 m and 219 m³/s, respectively, in 1957 (the year the gage began operation). The stage and discharge at Whitesburg had been as low as 0.5 m and 0.8 m³/s, respectively, as late as July 26. Three other things are notable about the set of hydrographs: First, the Hazard flood peak appears disproportionately small compared to the area of its watershed and those of the Whitesburg and Jackson gages. Second, the flood peak at Hazard lags about 13 hours after it peaked at Whitesburg; however, the peak at Jackson occurred at virtually the same time as the peak at Hazard. Third, the simple shape of the Whitesburg hydrograph suggests that it represents a single precipitation pulse on July 28, whereas the shoulders on the rising limbs of the Hazard—and to a lesser extent, the Jackson—hydrograph may represent responses to at least two pulses in precipitation or flow from tributaries. The Jackson hydrograph also has a shoulder on its falling limb that may represent response to a third such pulse.

Day and Chen (2023) calculated recurrence interval estimates for three gages within watersheds drained by each of the three forks of the Kentucky River, coincidentally including the Whitesburg gage discussed in this paper. They concluded that the two gages within the South Fork and Middle Fork drainages recorded July 27–28, 2022, floods with approximately 1-year recurrence intervals, whereas the Whitesburg gage along the North Fork recorded a July 27–28 flood with a 1,000-year recurrence interval. Day and Chen (2023) attributed the remarkable difference in recurrence intervals to a combination of greater antecedent rainfall, steeper topography, and larger proportion of barren ground as the result of mountaintop removal mining in the watershed above Whitesburg (58.5 percent for the watershed above Whitesburg on the North Fork versus 26.4 and 2.7 percent for the less heavily mined Middle and South Fork watersheds). Day and Chen (2023) did not list a source for their land cover values.

Haneberg (2023) plotted peak annual discharge values over time and calculated recurrence intervals for the July 2022 flood as recorded at the Whitesburg, Hazard, and Jackson gages. The peak annual discharge values showed a general decrease over the periods of record, with the 2022 discharges at Whitesburg and Jackson as significant exceptions to the trend. Haneberg (2023) calculated recurrence intervals of 851, 2, and 94 years, respectively, for the Whitesburg, Hazard, and Jackson gages. In a regional study of peak streamflow trends, Levin and Holtschlag (2022) reported either no monotonic trend or decreases in peak streamflow for selected locations in eastern Kentucky, attributing the decreases to large surface water impoundments.

Christian et al. (2023) published a map showing both the approximate locations of deaths attributed to the 2022 eastern Kentucky floods and the areas disturbed by mountaintop removal mining. They showed that 38.9 percent of the fatalities known at the time of their analysis occurred within 100 m of Troublesome Creek and 47.2 percent occurred within 800 m of Troublesome Creek. Overall, 69.4 percent of the known fatalities occurred within 0.8 km of features in the U.S. Geological Survey National Hydrography Dataset. Although Christian et al. (2023) wrote that their map could not prove mountaintop removal mining was responsible for deaths associated with the July 2022 floods, they did conclude that living near mountaintop removal mining sites “could increase risk of flood-related injury or death.”

Mountaintop removal mining, sometimes referred to more completely as mountaintop removal and valley fill mining or simply MTRVF mining, is a method that has been widely adopted in Central Appalachia in recent decades. In mountaintop removal mining, coal seams are exposed by blasting and removing mountaintops above the coal and placing the resulting debris (known as spoil) in adjacent valleys, resulting in permanent and minimally reclaimable landscape alteration (e.g., Shobe et al., 2024). Although the mountaintop removal offers an economical way to mine coal if externalities such as environmental and human health consequences are not considered, the method has been associated with hydrologic consequences such as increased downstream flooding; erosion, sliding, and chemical weathering of mine spoil; water quality degradation (Wiley and Brogan, 2003; Phillips, 2004; Ferrari et al., 2009; Palmer et al., 2010; McCormick and Eshleman, 2011; Zégre et al., 2014; Miller and Zégre, 2014, 2016; and Reed and Kite, 2020); and various serious human health problems (Christian et al., 2023, and references therein).

The results of published empirical studies of the effects of mountaintop removal mining on downstream flooding have been equivocal. Borchers et al. (1991) compared hydrologic characteristics of mined and unmined watersheds in West Virginia, including areas of mountaintop removal, and concluded that peak stream discharges from the mined watersheds were less than those from the unmined watersheds; however, pH, dissolved manganese, and suspended sediment loads were higher in streams draining mined watersheds. Wiley and Brogan (2003) inferred stream peak discharge during a 2-day rainfall event over six West Virginia watersheds, three of which had been affected by mountaintop removal mining and three of which had not. They calculated flood recurrence intervals ranging from less than 2 years to more than 100 years and attributed their results to the spatial variability of rainfall, which ranged over a factor of 5 across their study area during the rainfall event. Both the two largest and the smallest recurrence intervals were from the three watersheds affected by mountaintop removal mining; the three watersheds not affected by mountaintop removal mining produced intermediate results. In a review of primarily empirical work, Miller and Zégre (2014) concluded that the limited number of relevant mountaintop removal mining studies published at the time showed “significant variability in hydrologic response” to rainfall in watersheds affected by mountaintop removal mining. Zégre et al. (2014) wrote that their analysis of long-term trends from a West Virginia watershed showed “significant decreases in maximum streamflow and variability, and increases in base-flow ratio attributed to valley fills and deep mine drainage.” Miller and Zégre (2016) suggested that valley fills and underground mines could mitigate stormflow peaks by storing and gradually releasing water in watersheds affected by mountaintop removal mining. Nippgen et al. (2017) showed that mined watersheds delivered more water (but also more highly alkaline and saline water) through sustained baseflow during dry periods and significantly less water in response to storms than unmined watersheds. Reed and Kite (2020) documented significant gully and landslide development on supposedly stable valley fill slopes in eastern Kentucky and central West Virginia.

Other authors have used hydrologic models to evaluate the contribution of mountaintop removal mining to flooding. Phillips (2004) used relative runoff and flow detention calculations based on values from a previously studied mountaintop removal fill location (the Starfire mine) within the study area and typical forested Kentucky landscapes to infer that mountaintop removal mining was likely, but not guaranteed, to increase flood potential; he further emphasized that local site conditions could produce a range of results (potentially including reduced flood potential). Phillips (2004) also performed a geostatistical analysis of radar-quantified precipitation patterns from two storms and concluded it was unlikely that nearby watersheds would receive substantially different amounts of flood-producing rainfall (cf. Wiley and Brogan, 2003). Ferrari et al. (2009) and McCormick and Eshleman (2011) concluded that even when reclaimed, the response of surface-mined areas is more like that of urbanized areas than deforested but otherwise unmodified areas because soil compaction during reclamation reduces infiltration capacity and increases stormflow. In particular, Ferrari et al. (2009) concluded that flood magnitudes increase at a rate linearly proportional to the amount of area disturbed by mining and that this rate increases for larger recurrence intervals. Williamson and Barton (2020) used a combination of field hydraulic conductivity measurements and modeling to show that mined areas reclaimed using a modern forestry-driven approach responded to rainfall in a manner more like unmined areas than conventionally reclaimed areas but that all reclaimed soils they studied were thinner and less capable of storing water than natural soils in unmined areas.

The maps described in this paper were produced primarily using the open-source software QGIS, including GRASS (QGIS.org, 2024). The open-source Whitebox Tools geomorphologic analysis plug-in was additionally used for volumetric flux and other specialized calculations (Lindsay, 2016, 2023).

The digital elevation model (DEM) used as the basis for the hydrologic modeling described in this paper was developed by aggregating the publicly available KyFromAbove statewide lidar DEM (Commonwealth of Kentucky, 2024a) from 5 ft (1.5 m) to 25 ft (7.6 m) rasters and clipping it to the study area boundaries. It is based on phase 1 KyFromAbove coverage representing pre-2022 flood conditions. The aggregated DEM reduces file size and computing time while retaining acceptable landform resolution across large areas for many kinds of projects (e.g., terraced valley fill slopes and other mountaintop removal mining landscape artifacts remain visible in the aggregated DEM). Even at its reduced resolution, the aggregated DEM comprises nearly 49 million nonnull rasters. Because subsequent GIS calculations were based on the aggregated version of the statewide DEM, the project datum and coordinate system were the same as the original statewide DEM: NAD83 and Kentucky State Plane Coordinate System (single zone) in U.S. survey feet (EPSG 3089) with distances converted to metric for publication. GIS files using other datums or projections were reprojected to EPSG 3089 within QGIS as necessary. EPSG stands for European Petroleum Survey Group, which publishes a widely used database of map coordinate system information widely used in the GIS community.

The paths of the North Fork and Troublesome Creek were obtained from a statewide 1:24,000 National Hydrologic Dataset blueline stream coverage (Commonwealth of Kentucky, 2024b). Watershed catchment areas above each of the three stream gages were downloaded via the national U.S. Geological Survey Hydro Networked-Linked Index web application programming interface described by Blodgett (2023). A fourth watershed for Troublesome Creek above its confluence with the North Fork, which is not gaged, was extracted for this project using the r.watershed and r.water.outlet functions in GRASS. The watersheds are nested, which is to say that the Jackson watershed incorporates the Hazard, Whitesburg, and Troublesome Creek watersheds; likewise, the Hazard watershed includes the Whitesburg watershed.

Precipitation input was obtained as NCEP/EMC GRIB stage IV precipitation data showing 24-hour totals for the 4 days of most significant rainfall over the North Fork watershed above Jackson (July 26–29, 2022). Daily total precipitation maps were added together to calculate a 4-day total precipitation map (Figure 2D) and then reprojected and reduced to a 7.6-m raster size using bicubic interpolation via the GDAL warp function in QGIS. Resizing the raster cells is necessary because the flux routing function used in this analysis requires all input maps to have identical dimensions and raster cell sizes. The alternative, reducing the precipitation map cell size, would have been to aggregate the DEM upward to 4-km cells, which would have greatly diminished its topographic fidelity and usefulness for this analysis. Borga et al. (2022) raised concerns about the use of unprocessed radar precipitation data for hydrologic modeling; however, Carpenter et al. (2001) concluded that appropriately processed radar data can produce results comparable to those obtained from rain gage observations. NCEP/EMC stage IV data are produced using manual quality checks and integrate available rain gage data (Du, 2011).

The 2022 North Fork flood was simulated and summarized using a volumetric flux approach that allows for more explicit incorporation of spatial variability than the relative runoff and flow detention calculations performed by Phillips (2004) but requires less parameterization than the distributed dynamic flow routing models used by authors such as Ferrari et al. (2009) and Williamson and Barton (2020). This simplified approach also has the advantage of summarizing the aggregate effects of a spatially and temporally complicated event in a single map of cumulative values.

The project DEM was preprocessed using the BreachDepressionsLeastCost function with optional single-cell pit filling in the Whitebox Tools QGIS plug-in (Lindsay, 2023), which creates a DEM without artificial closed depressions. Closed depressions in a DEM will disrupt calculated flow across topographic surfaces, and it is standard practice to remove them from DEMs before hydrologic modeling.

In Eq. 1, V_out is the volume of water flowing out of a raster cell, L is a loading term representing water added to the cell, A is an absorption term with the same unit as L, V_in is the volume flowing downslope into the cell, and 0 ≤ E ≤ 1 is an efficiency factor representing the proportional ability of the raster to transmit volume. In this paper, L is the precipitation falling onto each raster cell and A represents a combination of processes, such as canopy interception, evapotranspiration, and shallow infiltration, that remove water from the cell. The A term is not restricted to physical absorption of the water alone. E and A are somewhat redundant because they both reduce the amount of material leaving the cell, but they do so in different ways. For simplicity, as discussed later, this paper assumes that E = 1 and uses A to account for any reduction in flow out of each cell. The V_in and V_out terms represent water flowing into and out of each raster cell down the topographic slope defined by the DEM, and the water flowing out of each raster is apportioned among multiple downslope cells using a D-infinity algorithm (e.g., Gruber and Peckham, 2009; Lindsay, 2023).

In Eq. 2, the P values represent the proportion of each land use class, the A values represent the absorption values for each land use class, and the V values represent the flow into and out of the entire study area (in this case, taken to be the watershed of the North Fork above the Jackson gage). The inflow, V_in, is the precipitation integrated over the study area, and V_out is the integral of a stream discharge hydrograph at the outlet from the study area.

Reprojection and interpolation of the July 26–29 precipitation map produced a smoothed 7.6-m raster version of the original 4-km raster map with minima and maxima that agree well in both location and magnitude (Figure 4). The 4-day precipitation totals vary considerably across the study area, ranging from 9.2 to 29.6 cm (Table 1). About half of the Hazard watershed received lower-than-average 4-day total precipitation, whereas areas along the North Fork north of the Hazard gage received the largest amounts of 4-day total precipitation. Much of the course of Troublesome Creek and the northern half of the Whitesburg watershed fall within areas of above-average 4-day total precipitation, whereas the southern half of the Whitesburg watershed falls within an area of lower-than-average 4-day total precipitation.

The interpolated 4-day precipitation map (Figure 4B) was converted from centimeters to meters and multiplied by the raster area, 7.6² m², to calculate the volume of water falling onto each raster cell in units of cubic meters. The result was the loading map for the volume flux calculations.

Eq. 3 allows an infinite number of possible A_mined and A_unmined combinations, comprising the solid blue line in Figure 5. Even if mined areas absorb all precipitation falling onto them (A_mined = 1), their proportional area is small enough that the unmined areas would be required to have an average absorption coefficient of A_unmined = 0.4 to account for a calculated flux ratio of V_out/V_in = 0.51. At the other end of the range, a value of A_unmined = 0.61 would be sufficient to account for the calculated V_in/V_out ratio if the mined lands had no absorption capacity (i.e., A_mined = 0). Assuming that reclaimed mine lands, regardless of the methods employed or time since reclamation, have absorption coefficients less than areas that have not been mined, the valid range of A combinations can be restricted to the solid portions of the blue line beneath the gold A_mined = A_unmined line in Figure 5. This places a heuristic limit on the plausible range of absorption values. There could be some unmined areas with small absorption coefficients, for example, areas covered by buildings, roads, or parking lots; however, their effects are already incorporated into the watershed-wide volume calculation in Eq. 3.

This paper uses two end-member scenarios developed to evaluate the maximum possible influence of surface mines on the July 2022 flood: (1) a mined end member in which mined areas have no absorption capacity (A_mined = 0) and, to account for the known watershed-wide V_in/V_out ratio, A_unmined = 0.61, and (2) a unmined end member in which the entire landscape has an absorption coefficient of 0.61 based on the unmined value calculated in the previous paragraph. This could apply to either an entire study area that had not been mined or one in which reclamation practices resulted in restoration of mined areas to the same state as unmined areas. The mined scenario does not represent actual values during the July 2022 storm; instead, it is used to estimate the maximum possible effects of surface mining during the storm by incorporating into the model the most different plausible pair of absorption values (A_unmined = 0.61 and A_mined = 0).

For the unmined scenario, the absorption map was set to 0.61 times the loading map for the entire study area. For the mined scenario, the absorption map was set to 0.61 times the loading map for areas with no previous mining and 0 for all mined areas, as shown in Figure 2C, regardless of age. The efficiency coefficient in Eq. 1 was set to E = 1 for this paper because the absorption coefficients already place a limit on the amount of water flowing out of each cell.

Results of the volume flux calculations are shown in Figure 6, including three magnified areas to illustrate the concentration of flow into the North Fork and Troublesome Creek, and are summarized for key locations in Table 1. The Table 1 discharge values were obtained by manually selecting raster cells close to the stream gage locations or, for Troublesome Creek, just upstream from its confluence with the North Fork. The gage and the mined scenario model cumulative flux results for the Jackson gage agree because the absorption coefficients for the mined scenario were calculated using the gage value.

The gage and modeled cumulative flux increase downstream as a function of watershed area and in qualitative agreement with the flood hydrographs in Figure 3. Figure 7 shows point values and best-fit curves of the form Q = kA^c for both the three gages and the model results (mined and unmined scenarios). In the best-fit curves, Q is the cumulative flux, k is the specific flux (i.e., the cumulative flux predicted per unit area of the watershed), and c is a power law coefficient representing the behavior of the watershed. The gage values follow a well-developed power law relationship of the form c = 1.89, whereas the model values follow nearly linear relationships, with c = 1.07 for the mined scenario and c = 1.06 for the unmined scenario. Different storms will likely produce different results. The variable Q is commonly taken to be an instantaneous value such as the peak annual discharge (e.g., Galster, 2007) but in this paper is the cumulative flux (i.e., discharge integrated over time). The mined model and gage values agree for the Jackson gage because those are the values used to calculate the absorption coefficients based on a weighted combination of mined and unmined areas. The models significantly overpredict the gage flux at Hazard and slightly overpredict the gage flux at Whitesburg. Troublesome Creek is not gaged, so it is impossible to compare the model results to an observed value, but the model significantly overpredicts the best-fit gage curve. Troublesome Creek contributes 10^7.90/10^8.35 or 35 percent of the mined model flux in the North Fork just downstream from the confluence while draining only 28 percent of the area above the confluence because the Troublesome Creek watershed received more rain than the rest of the watershed above the confluence.

Figure 8 illustrates the maximum possible difference in flux attributable to the modeled surface mine impacts, calculated by dividing the difference between the mined and the unmined cumulative flux maps by the unmined cumulative flux map. The maximum possible flow increases range from 21 percent for the North Fork at the Hazard gage, which drains a watershed with both the lowest proportion of its area affected by surface mining and the smallest 4-day precipitation value, to 41 percent for the Troublesome Creek watershed, which drains a watershed with both the highest proportion of its area affected by surface mining and the highest 4-day precipitation. As shown in Figure 7, especially in the magnified panels, the differences between the mined and the unmined scenarios are large within and immediately adjacent to mined areas, in many instances greater than 100 percent, but fall off with distance from the mined areas. Many tributaries have maximum possible increases approaching or even exceeding 100 percent at their confluences with the North Fork and Troublesome Creek.

Much of the variability in gaged flood discharges and the concentration of deaths along Troublesome Creek during the July 2022 flood is correlated with the spatial variability of precipitation during the storm. Even in their GRIB format (Figure 4), the available precipitation data show significant location-to-location variations in cumulative precipitation, which ranged over a factor of 3 across the study area during the last week of July 2022. In an analysis of the effects of mountaintop removal mining on Appalachian flooding, Phillips (2004) suggested that nearby watersheds would not be expected to receive substantially different amounts of flood-producing rainfall. The implication was that significantly different flood discharges from nearby watersheds during a single event could be attributed to the effects of mining, not spatially variable rainfall. However, that conclusion applied to watersheds separated at most by a few kilometers. The watershed of the North Fork above Jackson extends about 76 km along a NW-SE axis and about 50 km along a SW-NE axis, comprising 2,852 km². The results described in this study are more along the lines of those reported by Wiley and Brogan (2003), who observed that precipitation over a watershed of a similar size to that of the North Fork above Jackson ranged over a factor of 5 and recurrence intervals for six watersheds ranged from less than 2 years to more than 100 years during a single storm. There are no known published attribution studies assessing the possible influence of climate change on the severity of the July 2022 flood.

The most notable of the three stream gage records from the North Fork during the July 2022 flood is the one from Whitesburg, which recorded a flood with a recurrence interval at or near 1,000 years. About one-third of the watershed above the Whitesburg gage received more than 20 cm of rain over the July 26–29 period, and the area of heaviest rain coincided with the area of most extensive surface mining north of the North Fork. Compared to the rest of the study area, the Whitesburg watershed is not unique in terms of the proportion mined, slope steepness, or rainfall received during the July 2022 event (Table 1), yet its calculated flood recurrence interval of 851 years (Haneberg, 2023) to 1,000 years (Day and Chen, 2023) is exceptional compared to those for Hazard (2 years) and Jackson (94 years). The most likely explanation for the large Whitesburg flood recurrence interval is, all other things being equal, that a small watershed is less likely than a large watershed to receive an extreme large amount of rainfall over a relatively short period of record (47 years for the Whitesburg gage).

Given the exceptionally large recurrence interval for the flood at Whitesburg, it is also notable that the flood at the next gage downstream, at Hazard, had an estimated recurrence interval of only 2 years (Haneberg, 2023). This can also be understood largely in terms of precipitation patterns. The Hazard watershed (including the Whitesburg watershed upstream) had the lowest average precipitation of any of the four watersheds included in this paper. The precipitation map (Figure 4) further shows that a substantial portion of the watershed received less precipitation than any other part of the study area during the storm.

The largest amount of rainfall in the study area during the July 2022 storm, slightly more than 29 cm, fell along the North Fork between Hazard and Jackson (Figure 4). Secondarily, much of the course of Troublesome Creek aligns with a swath of above-average precipitation. The Troublesome Creek watershed received the largest average rainfall (22.6 cm) and contains the largest proportion of area affected by surface mining (25 percent) of the four watersheds analyzed in this paper. The large amount of rainfall across the Jackson watershed below Hazard, including the Troublesome Creek subwatershed that contributed a disproportionately large amount of the modeled flow relative to its area, likely contributed to the significant 94-year recurrence interval estimated for the flood at Jackson (including any mitigating effects of the Carr Creek reservoir along a tributary that joins the North Fork between Hazard and Jackson).

The absorption coefficient range of 0.51 ≤ A_unmined ≤ 0.61 estimated from available precipitation and stream gage data is consistent with the nationwide runoff map by Gebert et al. (1987), which shows average annual runoff in the project area of approximately 51 cm/yr during a 1951–1980 study period. Average annual precipitation at the weather station in Jackson is 131 cm/yr (National Weather Service, 2024b). Those average runoff and precipitation values suggest a landscape-wide absorption coefficient on the order of 1 − 51/131 = 0.61, which is identical to the unmined value used in this paper for the late July 2022 storm. The exact agreement may be coincidental; however, obtaining a similar value from an independent study conducted when a significantly smaller amount of land had been disturbed by mountaintop removal mining suggests that A_unmined = 0.61 is a reasonable first approximation for the unmined areas in this study and for the late July 2022 storm. Different storms may yield different results.

The power law exponents in the gage and model best-fit curves in Figure 7 differ significantly. Galster (2007) offered the interpretation that all other things being equal, the power law coefficient should be ∼1 (i.e., a linear relationship between watershed area and discharge because of mass conservation). Galster (2007) wrote that a value of ∼1 “seems to be the maximum possible in natural settings” and that values > 1 suggest that either the upper reaches of a watershed are inefficient in delivering precipitation as runoff or that downstream reaches generate disproportionately more runoff than upstream reaches. In either case, the difference could be the result of natural factors or human factors. Galster et al. (2006) discussed a heavily urbanized watershed in which the power law coefficient approached 2, ostensibly because an increase in impermeable urban surfaces contributed disproportionately more runoff than the less developed headwaters.

The best-fit power law coefficient of c = 1.89 for the gage data in this study is close to values of ∼2 that Galster et al. (2006) attributed to the disruptive effects of urbanization. Moreover, both Ferrari et al. (2009) and McCormick and Eshleman (2011) concluded that extensively mined watersheds behave like urban watersheds with large amounts of impervious surfaces. However, it is not clear whether the observed c = 1.89 in this study is attributable to surface mining or is the result of other factors not included in the models. The reason is that both the mined and the unmined scenario models produced nearly linear curves yielding c values of ∼1. Given the detailed maps of areas affected by surface mining, from which the mined and unmined absorption coefficient maps were developed, and precipitation variability used in the models, the model results should have accurately incorporated mining impacts.

Of the three gages, the Hazard gage departs the most from the straight-line model predictions. A large proportion of the Hazard watershed south of the North Fork appears to have been less heavily mined and received less precipitation than other portions of the study area (Figure 2C and D). It is possible that the combination of less mining disturbance and less intense rainfall during the July 2022 storm increased the actual absorption capacity of the land to a degree not incorporated in the models. Using values from Table 1 for the Hazard watershed exclusive of the Whitesburg watershed, Eq. 2 requires values of A_mined = A_unmined = 0.71 or A_mined = 0 and A_unmined = 0.80 for there to be no overprediction of the Hazard gage cumulative flux. Those values are high but might occur under the right conditions. Understanding the effects of precipitation intensity and duration on absorption coefficient values may lead to better model results.

In interpreting the physical significance of the absorption coefficients and their relation to model results, it is important to remember that the coefficients mathematically represent all processes that hinder or otherwise prevent the flow of water out of a raster cell, not just physical absorption of rainfall by soil or colluvium. Those other processes might include canopy interception, evapotranspiration, temporary storage of water by runoff retention features in operating or reclaimed mines, buffering of peak stormflow by transient storage of water within valley fills, or deep infiltration into underground mines. Borchers et al. (1991), Wiley and Brogan (2003), Miller and Zégre (2014), and Nippgen et al. (2017) published or summarized results showing that some valley fills associated with mountaintop removal mining may act as buffers that reduce stormflow during wet periods and enhance baseflow to streams during dry periods. Miller and Zégre (2016) concluded that existing underground mines, which were not considered in this paper, may play a significant role in buffering stormflow. Mine reclamation may have restored some absorption capacity, although field-based studies such as Williamson and Barton (2020) suggest that even areas reclaimed using modern best practices have diminished properties. Of those studies, it appears that only Wiley and Brogan (2003) include results from events as extreme as 100-year flood discharges, and none consider conditions as extreme as an 851- or 1,000-year flood.

Accurate maps distinguishing zones of mountaintop removal from zones of valley fill—not just areas impacted by surface mining that do not discriminate between zones of removal and filling—to which different absorption coefficients might be assigned, would help to improve flood models such as the one used in this paper. One possibility might be to reconstruct premining topography to infer areas of removal versus filling using legacy DEMs; however, that is a task more easily conceptualized than completed. Haneberg (2018) concluded that the 95 percent confidence limit for detection of real change between a modern 1.5-m lidar DEM and a legacy 10-m DEM digitized from photogrammetrically derived contours is about ±18 m in two eastern Kentucky 7.5-minute quadrangles. Much of that difference was attributed to localized zones or bands of significant horizontal errors as large as 30 to 45 m in unmined portions of the legacy DEMs. Therefore, legacy DEMs would have to undergo significant processing (e.g., Zhu et al., 2022) to transform them into conformity with the more accurate lidar DEMs if they are used to reconstruct premining topography. It would not be sufficient to simply layer one DEM over the other. Reconstruction of premining topography to improve Appalachian flood models should be a topic for future research.

Finally, the cumulative flux difference map in Figure 8 shows maximum differences that may have occurred as the result of mining-related elimination of absorption capacity, not necessarily actual values realized during the 2022 flood. It is based on an area-wide estimation of A_unmined = 0.61 and the assumption that any areas with evidence of mining as shown on the SkyTruth (2023) maps have no capacity to absorb floodwater. As discussed earlier, that may or may not be the case, especially given the overprediction of the model for cumulative flow at the Whitesburg and Hazard gages and some published evidence that valley fills may buffer the effects of heavy precipitation by serving as temporary reservoirs for infiltrating water. Overall, however, the mined-case model was calibrated to discharge measurements from the Jackson gage along the North Fork. Changing the ratio of mined to unmined absorption coefficients to other values along the line of allowable combinations in Figure 5 would have no effect on the modeled total cumulative discharge at Jackson, because the combinations simply reflect different ways of partitioning the average value calculated for the entire watershed above Jackson.

The significance of increases in cumulative flux can be understood in terms of recurrence intervals using the best-fit log-Pearson type III flood frequency curves developed by Haneberg (2023). Approximating a flood hydrograph as a triangle, it is easy to show that the proportional increase in cumulative flux will be equal to the proportional increase in peak discharge if the flood duration remains constant (i.e., a 20 percent increase in peak discharge will produce a 20 percent increase in cumulative flux). For the Whitesburg gage, for example, increasing peak discharge by 20 percent from 100 to 120 m³/s increases the recurrence interval of a hypothetical flood from 7 to 12 years, and increasing peak discharge by 40 percent from 100 to 140 m³/s increases the recurrence interval from 7 to 19 years. Likewise, increasing peak discharge at Hazard by 20 percent from 1,000 to 1,200 m³/s increases the recurrence interval of a hypothetical flood from 10 to 19 years, and increasing peak discharge by 40 percent from 1,000 to 1,400 m³/s increases the recurrence interval from 10 to 35 years. Modeled cumulative flux increases—potentially ranging from 22 percent along the North Fork at the Hazard gage to 50 percent along parts of Troublesome Creek, and greater than 100 percent along some tributary streams emanating from mined areas—are cause for concern and call for additional research on the topic.

Publicly available data show that differences in the amount of rain that fell across the North Fork watershed, which ranged from just over 9 cm (3.2 in) to nearly 30 cm (10.6 in) between July 26 and 29, 2022, are consistent with the significant differences in flood recurrence intervals calculated for the North Fork stream gages at Jackson (94 years), Hazard (2 years), and Whitesburg (850–1,000 years). Troublesome Creek, along which a disproportionately large number of people died during the flood, is not gaged; therefore, a recurrence interval cannot be calculated. However, the Troublesome Creek watershed received higher average rainfall than the entire North Fork watershed north of Jackson and the Hazard and Whitesburg subwatersheds. According to the computer model described in this paper, which was calibrated to known rainfall, streamflow, and the amount of area affected by surface mining, Troublesome Creek contributed a disproportionately large proportion of the cumulative flow in the North Fork at the confluence during the July 2022 flood (35 percent of the flow from 28 percent of the watershed area). Results from computer models with and without mining effects show that if known mined areas had no capacity to absorb or retain runoff, they could have increased cumulative flood discharge by 22 percent along the North Fork at Whitesburg to more than 100 percent in small tributaries emanating from mined areas.

Preliminary aspects of this and related research on the July 2022 flood effects were supported by the Kentucky Geological Survey and the National Science Foundation (NSF 21-547, GLD: 2242120). Comments by Jay Christian, Jason Dortch, Meredith Swallom, and three anonymous reviewers helped to improve the clarity the manuscript; however, the results and conclusions are the responsibility of the author.