Empirical Evaluation of QAOA with Zero Noise Extrapolation on NISQ Hardware for Carbon Credit Portfolio Optimization in the Brazilian Cerrado

Optimizing carbon credit portfolios is a critical challenge for climate mitigation, particularly in high-biodiversity biomes such as the Brazilian Cerrado. This study explores the practical application of the Quantum Approximate Optimization Algorithm (QAOA) combined with Zero Noise Extrapolation (ZNE) to address a multi-objective territorial planning problem. We model an 88-variable portfolio optimization involving carbon sequestration, biodiversity connectivity, and social impact metrics, executed on intermediate-scale IBM Quantum hardware (ibm_torino and ibm_fez). The results of seven independent hardware runs demonstrate that the QAOA+ZNE workflow consistently outperforms a classical greedy baseline. The quantum method achieves a mean portfolio score of 58.47 +/- 6.98, corresponding to a 31.6% improvement over the classical heuristic (44.42), with high statistical significance (p = 0.0009) and a large effect size (Cohen’s d = 2.01), where ZNE yields extrapolated expectation values of the portfolio objective rather than a discrete portfolio solution without noise. A validation run conducted after a 13-day interval confirms the temporal stability of the methodology against hardware calibration drifts. These findings establish empirical quantum utility in an environmental science context, showing that current NISQ-era devices, when coupled with rigorous error mitigation, can identify complex territorial synergies that myopic classical approaches overlook. The proposed workflow provides a scalable methodological template for high-precision environmental conservation policies.

💡 Research Summary

The paper presents a concrete demonstration of using the Quantum Approximate Optimization Algorithm (QAOA) together with Zero‑Noise Extrapolation (ZNE) to solve a real‑world, multi‑objective carbon‑credit portfolio optimization problem in Brazil’s Cerrado biome. The authors first construct a dataset of 88 municipalities selected from an initial pool of 246 based on carbon‑sequestration potential, biodiversity relevance, and social impact indicators. Each municipality is represented by a binary decision variable, and the objective function combines three weighted components—carbon sequestration, biodiversity conservation, and social benefit—each containing linear terms (individual scores) and quadratic terms (pairwise synergies captured by adjacency, biodiversity‑synergy, and social‑synergy matrices). The weights are set to wC = 0.33, wB = 0.33, and wS = 0.34, and a cardinality constraint forces exactly 28 municipalities to be selected, yielding a search space of roughly 1.45 × 10²² possible portfolios.

The problem is encoded as a Quadratic Unconstrained Binary Optimization (QUBO) model and mapped onto 88 physical qubits of IBM Quantum devices (ibm_torino and ibm_fez). The QAOA circuit depth is limited to p = 1 and p = 2 to stay within the coherence limits of the hardware. Classical parameters (γ, β) are optimized with the COBYLA optimizer using the expectation value of the cost Hamiltonian as the loss function.

Because current NISQ devices suffer from gate errors and decoherence, the authors apply ZNE as an error‑mitigation layer. They execute each QAOA circuit at three artificially amplified noise levels (scale factors 1, 3, 5), collect 8 192 measurement shots per scale, and perform Richardson extrapolation to estimate the zero‑noise expectation value of the portfolio objective. This yields a “noise‑free” score for each run without requiring full error correction.

Seven independent hardware runs were performed over a 17‑day period, each using fresh calibrations and random seed initializations. The ZNE‑corrected mean portfolio score was 58.47 ± 6.98, compared with a classical greedy heuristic that achieved 44.42 on the same data. This corresponds to a 31.6 % improvement. Statistical analysis using a two‑sample t‑test gave p = 0.0009, and Cohen’s d = 2.01, indicating both high significance and a large effect size. A validation run conducted 13 days later reproduced the same average score and confidence interval, demonstrating robustness against calibration drift.

The authors discuss several limitations. The shallow QAOA depth provides only an expectation‑value estimate rather than a concrete binary solution; the ZNE process dramatically increases the number of required shots, impacting runtime; and the choice of objective weights is somewhat subjective, suggesting a need for sensitivity analyses in policy contexts. Future work is outlined to explore deeper QAOA layers, alternative penalty‑based constraint handling, other mitigation techniques such as digital zero‑noise extrapolation or Clifford data regression, and scaling to >100 qubits as hardware improves.

In conclusion, this study provides the first empirically validated quantum advantage for a high‑dimensional, multi‑criteria environmental planning problem on real quantum hardware. By coupling a variational quantum algorithm with rigorous error mitigation, the authors show that NISQ‑era devices can outperform a strong classical baseline, offering a promising pathway for quantum computing to contribute to climate mitigation and biodiversity conservation strategies.