Search for: All records

ABSTRACT Nature‐based climate solutions in Earth's forests could strengthen the land carbon sink and contribute to climate mitigation, but must adequately account for climate risks to the durability of carbon storage. Forest carbon offset protocols use a “buffer pool” to insure against disturbance risks that may compromise durability. However, the extent to which current buffer pool tools and allocations align with current scientific data or models is not well understood. Here, we use a tropical forest stand biomass model and an extensive set of long‐term tropical forest plots to test whether current buffer pool contributions are adequate to insure against observed disturbance regimes. We find that forest age and disturbance regime both influence necessary buffer pool sizes. In the majority of disturbance scenarios in a major carbon registry buffer pool tool, current buffer pools are substantially smaller than required by carbon cycle science. Buffer pool tools and estimates urgently need to be updated to accurately assess disturbance regimes and climate change impact on disturbances based on rigorous, open scientific datasets for nature‐based climate solutions to succeed. 

Over the past three decades, assessments of the contemporary global carbon budget consistently report a strong net land carbon sink. Here, we review evidence supporting this paradigm and quantify the differences in global and Northern Hemisphere estimates of the net land sink derived from atmospheric inversion and satellite-derived vegetation biomass time series. Our analysis, combined with additional synthesis, supports a hypothesis that the net land sink is substantially weaker than commonly reported. At a global scale, our estimate of the net land carbon sink is 0.8 ± 0.7 petagrams of carbon per year from 2000 through 2019, nearly a factor of two lower than the Global Carbon Project estimate. With concurrent adjustments to ocean (+8%) and fossil fuel (−6%) fluxes, we develop a budget that partially reconciles key constraints provided by vegetation carbon, the north-south CO₂gradient, and O₂trends. We further outline potential modifications to models to improve agreement with a weaker land sink and describe several approaches for testing the hypothesis. 

Privacy-preserving machine learning (PPML) enables multiple distrusting parties to jointly train ML models on their private data without revealing any information beyond the final trained models. In this work, we study the client-aided two-server setting where two non-colluding servers jointly train an ML model on the data held by a large number of clients. By involving the clients in the training process, we develop efficient protocols for training algorithms including linear regression, logistic regression, and neural networks. In particular, we introduce novel approaches to securely computing inner product, sign check, activation functions (e.g., ReLU, logistic function), and division on secret shared values, leveraging lightweight computation on the client side. We present constructions that are secure against semi-honest clients and further enhance them to achieve security against malicious clients. We believe these new client-aided techniques may be of independent interest. We implement our protocols and compare them with the two-server PPML protocols presented in SecureML (Mohassel and Zhang, S&P’17) across various settings and ABY2.0 (Patra et al., Usenix Security’21) theoretically. We demonstrate that with the assistance of untrusted clients in the training process, we can significantly improve both the communication and computational efficiency by orders of magnitude. Our protocols compare favorably in all the training algorithms on both LAN and WAN networks.