Operation mercury: Impacts of national‐level armed forces intervention and anticorruption strategy on artisanal gold mining and water quality in the Peruvian Amazon

Artisanal and small‐scale gold mining (ASGM), a wealth‐generating industry in many regions, is nonetheless a global challenge for governance and a threat to biodiversity, public health, and ecosystem integrity. In 2019, the Peruvian government mobilized a targeted, large‐scale armed intervention against illegal ASGM, which has caused deforestation and water resource degradation in this Tropical Biodiversity Hotspot. Before the intervention, the extent of waterbodies created by mining (mining ponds) was increasing by 33%–90%/year; after, they decreased by 4%–5%/year in targeted zones. Mining activity indicators showed 70%–90% abandonment. New mining activity accelerated in nearby areas outside the targeted area (pond area increases: 42%–83%; deforestation increases +3–5 km2/year). Far from intervention zones, mining remained stable during the study period. Our analysis demonstrates that targeted, large‐scale government intervention can have positive effects on conservation by stopping illegal mining activity and shifting it to permitted areas, thereby setting the stage for governance. Continued conservation efforts must further address the impacts of informal mining while (1) limiting environmental degradation by legal mining; (2) remediating former mining areas to reduce erosion and enable reforestation or alternative uses of the landscape; and (3) sustaining such efforts, as some miners began to return to intervention areas when enforcement relaxed in 2022.

Recent attempts to mitigate the accelerating impacts of mining via formalization and small military operations have been largely unsuccessful (Damonte, 2016(Damonte, , 2021)).Such dispersed, smallholder mining is thought to be ungovernable by the state due to the complex interactions among actors (Peluso, 2018).In Madre de Dios (MDD) and elsewhere, military-and police-based interventions have been limited in both scope and duration by corruption, lack of resources, poor interagency cooperation, and inconsistency in laws and enforcement (Espin & Perz, 2021;Hilson & Maconachie, 2020).Difficulties in evaluating the success of mitigation efforts may also stem from inadequate monitoring and lack of objective criteria for judging outcomes.Assessing miner re-dispersal across a wide territory with limited road access is challenging.As a result, governments seeking to limit the negative environmental and human impacts of rapid ASGM expansion are hindered by uncertainty about the likely effects of intervention.Developing new intervention strategies and improving the success of these efforts require understanding the relative success of different intervention strategies and the likely trajectories of postmining recovery in affected ecosystems.
The focus of this study is widespread ASGM activity in the MDD region of southeastern Peru, where mining has led to the conversion of over 120,000 hectares of forest into a vast landscape of sprawling mining ponds and related drainage networks (Espejo et al., 2018).In the past decade, mining activity has rapidly expanded outside legal concessions to buffer zones surrounding several protected areas in MDD, including a Communal Reserve (Amarekeri), a National Park (Bahuaja-Sonene), and a natural reserve (Tambopata) (Espejo et al., 2018).In response to pressure to reduce illegal ASGM and the associated widespread rise of organized crime, human trafficking, and loss of territorial control in these areas (Bernet Kempers, 2020;Cortés-McPherson, 2020), in February 2019 the Government of Peru (GoP) launched "Operation Mercury" (OM, Operación Mercurio in Spanish), a broad series of interventions to eradicate illegal mining.These efforts began with a declaration of a state of emergency, the deployment of 1500 armed military and national police, and the eviction of miners and destruction of equipment within a 1000-km 2 mining region known as "La Pampa."La Pampa is located inside the established buffer zone for the Tambopata National Reserve, an area in which mining is prohibited (Ministry of the Interior, 2019).After most mining operations were shut down, a series of continuously occupied military and national police bases were established at mining zone access points and across the region to prevent the return of illegal mining and assess the intervention's effectiveness.Associated anticorruption measures included monthly rotation and replacement of garrisons and commanders.
The Peruvian government's unprecedented effort to enforce national laws governing both small-scale mining activity and protected areas initially appeared to succeed in its primary goals, as miners were largely excluded and mining operations were shut down.To provide objective, repeatable measures of shifts in mining activity following OM, we build on existing, deforestation-based estimates of ASGM activity in MDD.We quantify the hydrologic and water quality impacts of mining activity, major yet rarely measured outcomes of ASGM (Dethier et al., 2019).We contextualize changes during the intervention by measuring spatial and temporal variations in mining activity during rapid mining expansion (2016−2019) and post-OM intervention (2019−2021).This work provides an environmental analysis of policy enforcement that includes long-term monitoring of mining activity and recovery, with findings relevant to policymakers in Amazonian countries and around the world experiencing increases in alluvial gold mining.

Study area
We evaluated ASGM activity between 2016 and 2020 in nine mining areas in the Peruvian department of MDD (∼13.0 S, 70.0 W; Figure 1).Six of these mining areas are in the La Pampa mining zone, a vast mining area straddling the Interoceanic Highway (IOH).La Pampa mining areas located south of the IOH were the target of OM enforcement, which sought to eliminate all illegal mining and remove all unofficial settlements.In two mining areas in the La Pampa mining zone north of the IOH, land is titled for mining and has long-standing active mining operations; these areas were not targeted by OM despite similar levels of environmental degradation.We also examined three more distant sites that were not directly targeted by OM enforcement in Huepetuhe and Delta mining areas.Operations in Huepetuhe and Delta rely on varying degrees of mechanization, including considerable overlap with typical La Pampa mining methodology.Contemporaneous operations at these sites in the absence of direct intervention provide valuable context for understanding regional mining trends (Espejo et al., 2018).

Approach
In this analysis, we used multisensor satellite analysis to quantify mining activity.We integrated data from the European Space Agency's Sentinel-1 (10-m resolution) and Sentinel-2 satellites (a combination of 10-m and 20-m resolution) that can reliably detect deforestation (Tavares et al., 2019), water (Binh et al., 2017), and water quality (Toming et al., 2016), making them indispensable tools for analyzing mining activity.
We measured changes to mining pond extent (using both S2-MSI and S1-SAR) and water quality (using S2-MSI) during all study periods.Both satellites take images on a weekly basis, but to allow for cloud-free image analysis and consistent period of cross-comparison, we created wet and dry season composite images for each year from 2014 to 2021 for S1-SAR and from 2017 to 2021 for S2-MSI.We used these images to analyze spatiotemporal changes in land cover and water properties, in each case with respect to the February 2019 beginning of OM intervention, comparing the pre-and postintervention time series.We measured pond color, specifically pond yellowness (summed surface reflectance for red and green light wavelengths), which was identified by principal component analysis (PCA) as an indicator of mining activity using unmined oxbow lakes in the region for reference data.We use pond yellowness as a proxy for mining activity using thresholds of 0.36 and 0.28 to define "active" and "transitional" mining ponds.These thresholds are the 25th percentile of pond yellowness in La Pampa south of the IOH during the pre-OM and post-OM period, respectively.
We measured deforestation and revegetation with a Normalized Difference Vegetation Index (NDVI) by detecting temporal changes from vegetated to unvegetated pixels and vice versa, respectively.We measured precipitation using the Climate Hazards Group InfraRed Precipitation with Station (CHIRPS) daily data.In addition to evaluating aggregate change at each mining area, we conducted individual analysis of ponds >200 m 2 .We further describe the image-and object-based analyses of these satellite data in the Supporting Information; all the code used in analysis and figure generation is available at: www.github.com/evandethier/mining.

RESULTS
We find that OM significantly decreased mining-related deforestation and decreased the creation of excavated mining pits (referred to here as "mining ponds") in areas targeted by OM law enforcement efforts.During the same period, mining activity increased immediately adjacent to areas targeted by law enforcement, in concessions where mining is permitted.

Changes to mining pond extent
Mining areas in all areas of La Pampa were highly active and dynamic from 2016 to February 2019, indicating the rapid expansion of mining during that period at all sites except for one area, La Pampa Northwest, where mining activity was minimal (Figure 2; Table 1).Mining expansion rates were relatively uniform at sites south of the IOH (33%−92% new mining pond area/year), whereas expansion rates were considerably slower north of the IOH (0%−18% new mining pond area/year), and somewhat slower but also approximately linear at Huepetuhe, Delta, and Delta East (13%−27% new mining pond area/year).
TA B L E 1 Summary of changes in mining pond extent, a measure of mining intensity for each mining area, as determined by linear model fits for pre-and post-Operation Mercury (OM) periods.Following the initiation of OM in February 2019, mining pond expansion in La Pampa south of the IOH immediately halted, shifting from large increases to decreases of 4%−5%/year (2.2 km 2 /year active mining pond area lost) (Table 1; Figures 2 and 3).Adjacent areas north of the IOH had limited pre-OM activity, but experienced large increases immediately post-OM.Thus, in late 2019 and 2020, the locus of mining activity shifted from south to north of the IOH.Rates of mining pond expansion slowed in the presence of limited intervention during the same period at Delta and Delta East (respectively, 19% and 27%/year preintervention vs. 9% and 15% postintervention), but mining pond area continued to increase unabated at Huepetuhe in 2019 and 2020 despite COVID-19 lockdowns that forbade mining activity beginning in March 2020 (Smith-Roberts et al., 2021).

Deforestation and revegetation
As alluvial mining activities require land clearing to access soil sediments, trends in increasing mining pond extent track deforestation closely (Figure S3).In intervention areas, deforestation halted in 2019 after OM began.In general, former pond areas have seldom been recolonized by vegetation.Rather, incipient revegetation in abandoned mining areas is largely located at the edges of deforested swaths adjacent to former mining operations.In the intervention zone, revegetation rates range from 1 to 3 km 2 /year (3%−7% of the deforested area), with La Pampa (South) revegetating most rapidly.Given the recency of abandonment, this revegetation warrants continued monitoring in the future.Regionally, rapid deforestation continued at Huepetuhe and Delta (3−10 km 2 /year) and accelerated F I G U R E 3 Mining pond reflectance in red and green wavelengths (perceived pond "yellowness" due to mining-related suspended sediment in the water column) decreases rapidly where operations were abandoned after Operation Mercury intervention in February 2019.Symbols display average true color (RGB) for mining ponds in each region during the specified period, and symbol size indicates total mining pond area.
at La Pampa north of the IOH (3−5 km 2 /year).Overall, revegetation in the intervention area is slow due to intense soil degradation (Román-Dañobeytia et al., 2021) and was offset more than 4:1 by new deforestation in other areas.

Changes to mining pond color
Ponds affected by mining were distinct from unaffected ponds in the control dataset across all visible wavelengths (α = 0.05; p < 0.0001) and thus provide a measure of mining activity independent from deforestation or pond expansion.The color distinction was most pronounced in green and red wavelengths (Figure S1).PCA showed that the "yellowness" of ponds accounts for 75% of the total spectral variance in the dataset, connecting the metric of "yellowness" with increased suspended sediment in the water column (Dethier et al., 2020).Ponds with ongoing mining activity universally appear yellow due to sediment muddying the water.As mining became established in La Pampa areas south of the IOH, average pond reflectance increased, then stabilized.During the peak mining period from 2017 to 2019, almost every pond south of the IOH indicated constant use for mining activities; thus, average mining pond yellowness did not continue to increase after 2017 while miners continued to expand pond area (Figure S5).As sediment rapidly settled in abandoned ponds throughout the region, mining pond color decreased after OM intervention, with trends similar to, but more pronounced than, those observed in mining pond extent (Figures 3 and S4).During period, high-resolution imagery and ground truthing corroborate our findings that ponds were abandoned, water supply to them diminished, and sediment settled.Rates of decrease in mining pond yellowness outpaced rates of mining pond extent decrease.Pond yellowness decreased from an average of 0.49−0.57in 2018 to 0.21−0.38 in 2020 (decrease of 33%−55%) at mining areas targeted by intervention.
Rapid sediment settling and dilation in these ponds may have been accentuated during this period by low precipitation years during the dry seasons of 2019 and 2020.However, during this same low-precipitation period, we see rapid pond growth and yellowness increase in La Pampa Northeast and Northwest, which indicates that mining activity is the main cause of suspended sediment in the system.Indeed, we do not observe any region-wide response to changing precipitation except a slight seasonal effect in yellowness (0%−4%) and pond area (0%−10%) and thus attribute nearly all the observed hydrologic changes to direct human activity.
While mining pond activity decreased south of the IOH, mining activity rapidly increased in permitted areas.At La Pampa Northwest, mining pond yellowness increased 43% relative to pre-OM levels, synchronous with mining pond expansion; at La Pampa Northeast, where mining was more established before OM, mining pond yellowness was stable as miners continued to work or rework all existing mining ponds while progressively adding new mining pond area (Figures 3 and 4).
In Huepetuhe and Delta, where heavily mechanized ASGM is dominant relative to the minimally mechanized mining found in La Pampa, ponds associated with mining have broader or bimodal distributions of yellowness due to the presence of active and nonactive ponds throughout the year and across years (Figure 4).Though mining pond extent continued to increase steadily at each of these sites as new ponds were created (Figure 3), average pond yellowness did not indicate any notable trends during the study period and distributions from pre-and post-OM intervention in La Pampa are stable.

DISCUSSION
In areas targeted by OM in La Pampa, mining ponds rapidly (<6 months) lost the signature of active mining activity.Though most ponds remained on the landscape, 2 years after the intervention, active mining ponds occupied less than 50% of their peak area, with only 10% of ponds still active by late 2020 in La Pampa South, the largest illegal mining area (Figures 5 and S4).The small number of active ponds within the intervention area may indicate continued and clandestine illegal mining activity, but also may be due to sediment mixing in ponds that are hydraulically connected to streams.Region-wide, nearly 40% of pond area was classified as inactive by 2021, with approximately 20% of pond area in a "transitional" phase but trending toward the "inactive" threshold.
In the La Pampa mining areas that were located north of the IOH (i.e., outside the Tambopata National Reserve buffer zone and immediately adjacent to the target zone of OM), rapid increases in mining pond extent and utilization occurred after OM intervention south of the IOH.After 2 years of increased activity, these new mining areas were not yet as extensive as the abandoned mining zones south of the IOH.However, rates of mining pond expansion north of the IOH were fast enough that ∼3−5 years of new mining (including reworking of previously mined areas and newly deforested areas) would exceed the extent of all the abandoned mining zones targeted by OM.
Rates of deforestation at active mining areas that were not targeted by OM are currently much faster than revegetation at abandoned mining areas.All the incipient revegetation has occurred at the edges of mining areas, and not around mining ponds, and it remains unclear how the system will respond to continued abandonment.Mining-related deforestation has both eliminated aboveground biomass and caused near-total soil loss.Vegetation cover may return on decadal timescales in the absence of new mining activities, but full ecological recovery may take centuries to millennia, given the soil nutrient depletion (Fu et al., 2017) and slow rates of soil regeneration (Richter et al., 1999).The continued lack of vegetation and slow rates of natural forest regeneration and succession (Peterson & Heemskerk, 2001), presence of ponded water, and abundance of loose sandy sediment maintain the profound alteration of the landscape relative to its premining state.
One point of interest following our findings regards the possible recovery trajectory of the mining ponds themselves.Though some abandoned mining ponds have partially drained, most retain their approximate active shape.In addition, such sediment is likely contaminated with heavy metals, most notably elemental mercury (typically used in ASGM to concentrated particulate gold in mined sediments) (Diringer et al., 2015), which methylates to methylmercury (MeHg) and is taken up and magnified in aquatic food webs, creating mercury exposure risks to piscivorous wildlife and fish-consuming human populations (Wyatt et al., 2017;Yard et al., 2012).Since mercury methylation rates are higher in water bodies with high levels of dissolved organic matter, anoxic bottom conditions, and high rates of sulfate-reducing bacteria (Guimaraes et al., 2000;Lino et al., 2019) (conditions typical of Amazonian post-ASGM mining ponds), these abandoned ponds may function as centers of high MeHg production.During flooding events common during the Amazonian rainy season (Jan-May), MeHg may be transported to the greater watershed as many ponds become temporarily connected with streams and rivers (Gerson et al., 2020).Elevated levels of mercury present in post-ASGM mining ponds may limit safe future pond use (particularly for aquaculture) despite a local interest in using these ponds for post-ASGM economic development.
Though OM could not be fairly characterized as a conservation or purely environmental intervention, our results have implications for intervention strategies that rely on armed forces for conservation goals.Militarized interventions used to enforce environmental conservation are often referred to by the umbrella term "green militarization" and have led to complex outcomes (Duffy et al., 2015;Hilson & Maconachie, 2020), including negative impacts on local social or economic conditions (Hilson & Maconachie, 2020;Neumann, 2004;Van Bockstael, 2019).Interventions against ASGM are often unpopular among miners and some outside observers, criticized for shortsighted targeting of artisanal miners without a clear development strategy (Spiegel, 2014), lack of prolonged success (Prescott et al., 2020), and heavy-handed tactics (Hilson, 2020;Hilson & Maconachie, 2020).
In Peru, past military and police operations against mining had short duration, limited scope, and often poor coordination with other government stakeholders on issues surrounding mining, such as criminal syndicates, public F I G U R E 5 Time series of regional (A) mining pond activity relative to its 2019 peak and (B) mining pond utilization (the percent of mining ponds used for active mining) indicates the cumulative effects of mining pond abandonment forced by Operation Mercury intervention at some mining areas in Madre de Dios.Mining pond area was rapidly identifiable as transitional or inactive following the intervention.Active mining pond area comprises <50% of its peak, and almost 40% of ponds are now classified as inactive, down from a high point of 80% utilization in 2019.
health, corruption, land titling, and so forth, resulting in minimal effect on mining activity, and have thus received similar criticism (Damonte, 2016;Espin & Perz, 2021).However, alternative-livelihood interventions also have a sporadic record of success (Hilson & Banchirigah, 2009;Prescott et al., 2020;Roe et al., 2015), and formalization of ASGM has been challenging in the absence of robust enforcement.Miners often return to banned locations or practices as enforcement wanes and are deterred from formalization by the lack of clear mining policy, mining policy that changes frequently, multilayered bureaucracy, and fees (Alvarez-Berrios et al., 2021).
OM differed from prior interventions in that it was not limited to short-term military and law enforcement action, but instead garnered the participation of a full range of government ministries and featured an extended occupancy of intervening forces with a clear anticorruption strategy that included frequent force rotation (Riehl, 2020).Seeking to shift miners to permitted areas with ongoing mining, the operation preserved legal mining concessions immediately adjacent to the conservation zone.We find that miners did not return to OM-targeted areas in the 2 years following the intervention, even when enforcement weakened due to COVID-19.Rather, mining generally shifted from places where it was not titled to places where it is titled.However, as this study was in review, there have been returns of mining to areas of the intervention zone close to the IOH beginning in the summer of 2021 in the areas most highly susceptible to corruption (Finer et al., 2021).Though areas targeted by OM remain mostly abandoned, incipient but rapid return of mining also shows that enforcement may not be successful without associated changes in governance-in this case, the formalization of mining in titled areas and/or improvement of existing formalization processes (OAS DOTOC, 2021) coupled with sustained enforcement in areas where mining is prohibited.Because widespread environmental degradation persists in both titled and untitled mining areas in MDD, further work in mitigation and rehabilitation are likely necessary to reduce the negative impacts of mining and restore postmining landscapes after mining ceases.Follow-on examination of the social and economic ramifications of displaced and resurgent mining is critical to generate a more holistic analysis of the mid-and long-term consequences of this intervention approach for addressing illegal ASGM in tropical frontier regions in Amazonia.

CONCLUSION
The growth of illegal artisanal gold mining has been highlighted as a challenge for effective governance, an avenue for widespread corruption, and a major threat to biodiversity, public health, and ecosystem integrity in the Amazon Region.This study is the first to examine the interrelated land cover and hydrological consequences of the outcomes of OM, a large and sustained armed forces-led intervention focused on illegal gold mining within protected areas in the Amazon Region.Our results indicate that OM led to nearuniversal abandonment of mining operations in targeted areas for 2 years.Continued observation of mining in MDD is essential to characterize how mining activity is displaced to other protected areas in the region.As illegal gold mining continues to proliferate in the Amazon and other tropical regions worldwide, we believe that the techniques presented here can be useful for such assessment of conservation and mitigation efforts.Already in MDD, mining activity has resumed in some forbidden areas as enforcement has been relaxed, underscoring the importance of comprehensive governance required to limit mining where it is not legal and manage mining in a socially responsible way where it is.

F
I G U R E 1 Map showing major mining areas in Madre de Dios, Peru.Heavily mechanized mining that utilizes earth-moving vehicles predominates in Delta and Huepetuhe, while minimally mechanized mining relying on suction pumps and human labor is used almost exclusively in La Pampa.The Interoceanic Highway, shown by a dashed line, separates La Pampa into Northern and Southern zones; illegal mining operations in the Southern zone (labeled South, East, Central, and West) were the target of Operation Mercury in February 2019.Light gray areas did not have wide-scale gold mining activity, though mining operations are expanding rapidly into new territory.
At locations targeted by OM, trends in mining pond extent show statistically significant reversals, changing from increasing to decreasing.Sites north of the Interoceanic Highway show dramatic increase in new mining pond area following OM.*Significance is evaluated at 0.01.F I G U R E 2 Change to mining pond planform area as recorded by S1 synthetic aperture radar.The beginning of Operation Mercury is shown as a vertical line in February 2019.Gray bands in each panel indicate uncertainty (±1 SD).
Distribution of pond yellowness (reflectance in the red wavelengths + reflectance in the green wavelengths of the S2-MSI) for pre-Operation Mercury (OM) in 2018 (unfilled bars) and post-OM in 2020 (colored bars) shows decreases at sites targeted by OM (green), increase or no change at mining areas proximal to but not targeted by OM, where miners diverted after OM, and no discernible change in pond color distribution, despite an increase in the total number of ponds, at sites not targeted and distal to the OM intervention.