No evidence of increased forest loss from a mining rush in Madagascar's eastern rainforests

The study investigates whether a major artisanal and small-scale mining (ASM) rush—the 2016 sapphire rush at Bemainty within Madagascar’s Corridor Ankeniheny–Zahamena (CAZ) protected rainforest—led to increased deforestation or forest degradation. ASM often occurs in biodiversity hotspots and can cause habitat loss, chemical pollution (notably from mercury in gold mining), erosion, and increased hunting pressures. However, robust quantitative assessments, especially using counterfactual methods, are scarce, and much evidence comes from descriptive case studies focused on high-impact gold mining regions (e.g., Amazon, Ghana). At Bemainty, media and some reports alleged extensive forest loss and threats to lemurs, but previous analyses conflated time periods and misattributed deforestation predating mining. The authors aim to rigorously evaluate the forest impacts of the Bemainty rush relative to a counterfactual of no mining and to contextualize ecological and social outcomes with lemur surveys and informal interviews.

Prior work documents multiple pathways by which ASM affects biodiversity: direct habitat loss from clearing, fuelwood and construction timber harvesting; chemical contamination (e.g., mercury, cyanide) in gold mining; erosion and siltation from mining along waterways; and indirect pressures such as increased logging, agriculture, and bushmeat hunting. Quantitative evidence of ASM-driven deforestation is limited and concentrated on gold mining in the Amazon and Ghana, where impacts are large due to mercury use and mechanization. Studies using satellite data report wide variation in forest loss near ASM sites (0.1%–46% within 5 km buffers), but typically lack counterfactual designs. Madagascar’s ASM has rapidly expanded with gemstone rushes often overlapping protected areas, heightening concern. Conflicting narratives around Bemainty include claims of high deforestation versus critiques that much valley clearing predated mining and was agricultural. This underscores the need for counterfactual, context-specific evaluation.

The authors use a synthetic control method to estimate counterfactual forest loss in the absence of mining, comparing observed outcomes in the Bemainty drainage basin to a weighted combination of similar control basins. Unit of analysis: Level 9 HydroBASINS drainage basins, chosen to match the expected spatial extent of impacts for alluvial (secondary) sapphire deposits and potential resource-use travel distances. Study period: 1991–2021, with pre-intervention 1991–2011 and post-intervention from 2012 (onset of mining; rush begins 2016). Donor pools: Primary donor pool comprises eight CAZ drainage basins with ≥70% forest cover in 2011, similar protected status transitions, and with known gem-mining sites (e.g., Didy) and basins overlapping >10% with other protected areas excluded; overlapping sections <10% were trimmed. A secondary, wider donor pool includes 13 forested basins from the former Toamasina province (the eight CAZ basins plus five additional unprotected forested basins) to test robustness. Covariates for matching include pre-2012 proxies for potential confounders: 2011 population density, 2001–2011 population growth, mean distance to settlements, elevation, slope, annual precipitation, distances to cart tracks, roads, rivers, and forest edge (2011), plus forest area and similar protected status. Outcome variables: Forest loss derived from the Tropical Moist Forests (TMF) dataset at 30 m resolution (1990–2021), masked to Madagascar 1990 forest extent. Deforestation is long-term canopy loss (>2.5 years); degradation is temporary canopy loss (<2.5 years) with recovery. For deforestation, annual deforested area per basin is taken from TMF Deforestation Year. For degradation, the Annual Disruptions product is reprocessed in Google Earth Engine to identify the start year of each degradation episode (including repeated events), masked to pixels classified as degraded in the TMF Transition Map, yielding annual degraded area per basin. Outcomes are measured three ways: (1) annual rate (% of forest at start of year), (2) raw hectares, and (3) cumulative hectares. Synthetic control construction: Implemented with Synth in R, weighting control basins to match pre-intervention outcomes and covariates; fit assessed via pre-treatment Mean Square Prediction Error (MSPE) and visual checks. Significance testing: In-space placebo tests assign false treatment (2012) to each donor basin; only placebo pairs with pre-treatment MSPE ≤5× that of Bemainty are retained. If Bemainty’s post-treatment differences exceed placebo ranges, effects are considered significant. Validation: In-time placebos assign false treatment to 2009 to test method credibility and robustness. Field data: Complementary lemur surveys and informal interviews conducted Oct–Nov 2019. Lemur surveys: five ~7 km transects from villages into adjacent forests, each repeated 5–6 times (27 transects; ~189 km total), during diurnal activity windows; one limited nocturnal survey. Interviews: 73 semi-structured interviews in five settlements (29 miners, 44 farmers), opportunistically sampled, conducted in local dialect, covering resource use, environmental change perceptions, tree cutting, and hunting.

- No significant increase in deforestation or forest degradation attributable to the Bemainty mining rush. Across both outcomes (deforestation, degradation), three metrics (annual rate, annual hectares, cumulative hectares), and two donor pools (CAZ-only and wider Toamasina), observed forest loss at Bemainty did not exceed the synthetic control beyond the range indicated by placebo tests. - Although deforestation increased at Bemainty in 2016–2017 and was higher than the synthetic control in 2017 (notably for cumulative deforestation), the difference fell within placebo-derived statistical noise and was thus not significant. Seven of eight similarly forested CAZ basins also saw increased deforestation from 2016 to 2017, suggesting broader regional drivers (e.g., shifting agriculture) rather than mining-specific effects. - Post-2012, Bemainty often exhibited equal or lower forest loss than its synthetic control (differences generally within uncertainty bounds). Isolated years of significantly lower deforestation (e.g., 2013) were not consistent across all outcomes or scales. - Lemur surveys (Oct–Nov 2019) recorded 10 of the 13 lemur species known from CAZ, including frequent encounters with critically endangered Indri (Indri indri) and black-and-white ruffed lemur (Varecia varecia), indicating an apparently healthy, diverse diurnal lemur community two years after the rush. Five small-bodied lemur traps were found; nocturnal species were under-sampled. - Interviews (n=73) suggest miners and farmers both used wood for fuel and construction but reported avoiding mature trees; few respondents cited shifting agriculture as ongoing deforestation. No respondents self-reported lemur hunting; several noted local taboos (fady) protecting Indri, also linked by miners to luck in finding sapphires. - The per-capita deforestation footprint of ASM at Bemainty appears substantially lower than that of shifting agriculture, which remained the dominant driver of forest loss and likely explains why overall forest loss did not exceed the counterfactual. - The rush directly supported livelihoods of up to 30,000 people at its peak, but brought social costs (insecurity, disease, disrupted rice production, price inflation). Potential unmeasured environmental impacts include increased erosion, turbidity, and altered water flows in streams.

Findings contradict prominent media claims that the Bemainty sapphire rush caused extensive deforestation and severe lemur declines. The absence of a significant mining-induced increase in forest loss likely stems from five context-specific factors: (1) deposit geology—secondary alluvial sapphires confined mining to narrow valley bottoms; (2) legacy clearing—valley floors were substantially cleared for agriculture before mining, reducing the need for forest clearance; (3) short duration—the peak rush lasted less than a year; (4) limited timber demand—miners reported using dry wood and small poles, not engaging in charcoal production; and (5) the dominance and larger footprint of shifting agriculture, which continued to drive most forest loss. Together, these conditions can yield minimal incremental forest impacts from unregulated ASM, at least relative to counterfactual land uses. Interview and survey evidence suggests fady (taboos) protecting Indri persisted despite the influx of migrants, with miners’ beliefs potentially reinforcing protection; the observed presence of Indri and ruffed lemurs supports low hunting pressure. Nonetheless, the rush imposed noteworthy social trade-offs (crime, insecurity, disease, food insecurity), and likely caused freshwater ecosystem impacts (sedimentation, turbidity) not captured by the forest-cover analysis. The study underscores the heterogeneity of ASM’s environmental effects across contexts and minerals, cautioning against generalizing from high-impact gold-mining cases. It advocates for nuanced, context-specific policy responses that weigh ASM’s socio-economic benefits against environmental trade-offs, and for broader, more robust interdisciplinary evaluations, including higher-resolution forest data and direct ecological and social measurements.

An unregulated sapphire mining rush involving tens of thousands of miners in a protected Malagasy rainforest did not increase forest loss beyond a counterfactual of no mining, challenging narratives of widespread mining-driven deforestation. Shifting agriculture remained the predominant driver of forest loss, and ASM’s per-capita forest footprint was comparatively small. Field evidence two years post-rush indicates an apparently healthy, diverse lemur community. While this is a single case, results highlight the heterogeneity of ASM impacts and the importance of assessing them relative to alternative land uses. More case studies using robust counterfactual methods and integrated ecological and social data are needed to inform proportionate, evidence-based policies that enhance ASM’s socio-economic benefits while minimizing environmental trade-offs.

- Remote sensing resolution and scope: TMF data at 30 m may miss small-scale forest structure changes (e.g., selective logging) and cannot assess non-forest impacts such as water quality, erosion, and aquatic biodiversity. - Donor pool size and matching: The primary donor pool within CAZ is small (n=8), limiting placebo power; although a wider pool (n=13) was used for robustness, suitable well-matched basins are scarce. - Timing and representativeness of field data: Interviews and lemur surveys were conducted two years after the rush, with opportunistic sampling and limited nocturnal surveying, introducing recall bias, social desirability bias, and potential non-representativeness of peak-rush conditions. Lack of pre-rush ecological baselines prevents assessing immediate or transient impacts on lemur populations. - Generalizability: Context-specific factors (alluvial deposit confinement, prior valley clearing, short duration, and local land-use patterns) limit extrapolation to other ASM settings, especially mechanized or mercury-using gold mining. - Potential confounding: Despite careful covariate selection and protected status matching, unobserved factors and management changes could still influence outcomes, though the synthetic control and placebo tests aim to mitigate this.