Climate physical risks and technological innovation in the grain industry chain: an empirical analysis based on machine learning of patent texts in China

Abstract

Context

Climate change-induced physical risks are permeating the entire grain industry chain (GIC), triggering systemic restructuring pressures. While global adaptive responses increasingly prioritize technological innovation as the most promising solution, empirical validation remains scarce due to measurement challenges in quantifying GIC technology innovation (GICTI). Critically, prior studies focus narrowly on single production segments, lacking a deconstruction of innovation responses across the full chain. This also leaves the linkage mechanism between climate physical risks and different innovation types still in a black box, hindering effective policy-making.

Objective

This study investigates the structural impacts of climate physical risks on GICTI: 1) Quantify risk effects across five segments: pre-production R&D segment, cultivation-harvest segment, storage-transport-processing segment, distribution-consumption segment, and whole-chain technologies; 2) Measure asymmetric effects on disruptive versus incremental innovation; 3) Verify the pivotal role of R&D capital allocation in risk transmission; 4) Identify regional heterogeneity patterns.

Methods

1) Innovation measurement: Using Chinese GIC patent data, we constructed a multi-level semantic analysis framework with machine learning: BERT model decoded patent abstracts to classify innovations into specific chain segments; BGE model transformed text into semantic vectors, identifying disruptive/incremental technological innovations via average vector distance thresholds. 2) Climate physical risk quantification: Provincial climate physical risk indices were developed from multi-dimensional perspectives. 3) Empirical analysis: Bidirectional fixed-effects models examined the impacts and mechanisms of climate risks on innovation levels/structure across Chinese provinces (2004–2023).

Results and conclusions

1) Climate physical risks exhibit significant long-term driving effects on GICTI. 2) Structural heterogeneity prevails: Cultivation-harvest and storage-transport-processing segments show robust positive responses, while incremental innovation responds more systematically than disruptive innovation. 3) Mechanistically, R&D capital deepening constitutes the core transmission channel, though excessive R&D intensity and private-public R&D investment ratios substantially weaken innovation incentives. 4) Regionally, innovation responsiveness is significantly stronger in grain-producing provinces and western China.

Significance

1) Methodologically, the BERT-BGE dual-model framework transcends IPC coding limitations, establishing a replicable paradigm for industry-specific innovation measurement. 2) Theoretically, we empirically integrate climate physical risks, chain-wide innovation, and spatial heterogeneity, unraveling risk-innovation dynamics and advancing climate-adaptive agriculture and innovation economics. 3) Practically, findings enable diagnosis of regional innovation gaps, optimization of climate-adaptive R&D resource allocation, and targeted policy formulation for differentiated innovation incentives.