On track to achieve no net loss of forest at Madagascar's biggest mine

The UN Sustainable Development Goals highlight the need to balance economic growth with biodiversity conservation. No Net Loss (NNL) policies aim to allow development while preventing environmental damage by implementing a mitigation hierarchy: avoiding, minimizing, restoring, and offsetting biodiversity loss. While many countries mandate or support biodiversity compensation, robust evaluations of offset effectiveness are scarce. This paper focuses on the Ambatovy nickel and cobalt mine in Madagascar, one of the world's largest lateritic nickel mines, situated in a highly biodiverse and deforestation-threatened rainforest. Ambatovy committed to NNL of biodiversity, implementing an averted loss offset strategy by preventing deforestation from shifting agriculture in four designated areas. The study uses statistical approaches to assess whether the offsets have successfully compensated for forest loss at the mine site. The challenge lies in estimating the counterfactual scenario – what would have happened without the offset intervention. This requires robust statistical methods, capable of testing the robustness of estimates across different model specifications, to mitigate the risks of arbitrary modeling choices influencing results.

The existing literature highlights the complexities and controversies surrounding biodiversity offsets, particularly concerning permanence, equivalence, and equity. Questions arise about the use of proxy measures for biodiversity (like forest loss), the accuracy of counterfactual loss estimations, and the potential for leakage (deforestation displacement). While averted loss offsets can be effective where high-quality habitat is threatened, accurate assessment necessitates robust statistical methods that account for various potential biases and model specifications. Existing evaluations of biodiversity offsets, especially those in forested ecosystems, are limited in number and often lack robust methodologies, underscoring the need for this study.

The study uses a combination of statistical matching and regression models to assess the impact of Ambatovy's four biodiversity offsets on deforestation. The former province of Toamasina in eastern Madagascar served as the study area. A grid-based sampling strategy selected a subsample of 30 x 30 m² pixels that were forested in 2000, excluding protected areas and a 10 km buffer around the offsets to minimize the influence of leakage effects. The outcome variable was the annual deforestation rate from the Global Forest Change (GFC) dataset. Statistical matching paired offset pixels with control pixels from the wider landscape exhibiting similar characteristics (slope, elevation, distance to road, forest edge, and previous deforestation). This matching controlled for confounding factors potentially influencing deforestation. Site-based difference-in-differences regressions were conducted for each offset, comparing deforestation rates in offset and control areas before and after offset implementation. A fixed-effects panel regression was used to analyze all four offsets simultaneously. To ensure robustness, 116 alternative model specifications were explored, varying matching parameters (distance measure, caliper size, control-to-treated ratio, replacement), and the inclusion of additional covariates (precipitation, distance to river, cart track, settlement, population density). A posteriori validity checks ensured the models met essential assumptions (sufficient matching, covariate balance, parallel trends). The avoided deforestation was then estimated and compared to the forest loss at the mine site to assess whether NNL was achieved.

The site-based difference-in-differences regressions indicated significant deforestation reduction in two offsets: Ankerana (96% reduction) and the Conservation Zone (66%). Torotorofotsy showed no significant effect, while CFAM couldn't be analyzed due to a lack of parallel trends in the pre-intervention period. The fixed-effects panel regression, incorporating all four offsets, estimated an overall 58% reduction in annual deforestation. Robustness checks across 116 alternative model specifications confirmed these findings, with the vast majority supporting significant avoided deforestation. The mine destroyed 2,064 ha of forest. Based on the difference-in-differences regressions, 1,948 ha of deforestation was avoided (94% of mine-caused loss), primarily in Ankerana. The fixed-effects model estimated 1,644 ha avoided (79%), suggesting NNL could be achieved by the end of 2021, potentially earlier than the company's initial projection. Analysis of leakage effects within a 10 km radius of the offsets showed no significant impact on deforestation outside the offset boundaries. Converting the findings to Cohen's d effect size showed the Ambatovy offsets were more effective at reducing deforestation than 97% of interventions in a comparison dataset, surpassing even most protected area interventions. The Conservation Zone's reduction is likely due to existing access restrictions.

The study provides strong evidence that Ambatovy's biodiversity offsets significantly reduced deforestation, surpassing the effectiveness of many other conservation interventions. Several factors could explain this success: the explicit focus on measurable impact (NNL), substantial funding, and the project's structured approach to achieve a quantifiable outcome. However, the reasons for differing success across the individual offsets (e.g., Ankerana's success versus Torotorofotsy's lack thereof) remain unclear, potentially due to varying levels of enforcement and community engagement. This highlights the importance of considering factors beyond simply habitat protection for successful outcomes. The study addresses the research question regarding the effectiveness of Ambatovy's offsets in achieving NNL by demonstrating its significant contribution to avoided deforestation. The results are relevant to the field of biodiversity offsetting and conservation by offering robust evidence of success while acknowledging caveats. This highlights potential avenues for more effective biodiversity offsetting policies and projects.

This study provides robust evidence that Ambatovy's biodiversity offsetting program is successfully mitigating deforestation related to mining activities. Although NNL of forest may be achieved by the end of 2021, the researchers emphasize the importance of continued monitoring given the inherent complexities of biodiversity offsetting and the need for effective long-term management of the offsets after the mining company's involvement ceases. Future research should focus on examining the factors that influenced the varying success across different offsets, the longer-term effects of the interventions, and exploring more holistic approaches to biodiversity conservation that simultaneously address social and economic factors.

While the study employs robust methodology and addresses several potential biases, limitations remain. The reliance on forest cover as a proxy for biodiversity, the inherent uncertainties in counterfactual estimations from observational data, the possibility of indirect impacts from the mine on regional deforestation and the relatively small sample size (n = 38 for some regressions) affecting the precision of the estimates should be noted. Further limitations exist regarding capturing small-scale forest damage and the lack of species-level data. Assessing the true permanence of achieved conservation gains, given Madagascar's challenging governance context, remains a critical challenge.