Answers are grounded in this paper and cite their sources.
Accurate estimation of forest aboveground carbon stock (AGC) is essential for carbon governance, but potential geospatial correlations may affect estimation accuracy. Although hyperspectral imaging (HSI) remote sensing can characterize vegetation changes caused by geospatial differences, its redundancy hinders models from capturing geospatial information. To this end, a key information-guided geospatial–spectral image fusion framework (KIG-GeoSIF) is proposed to improve the model's ability to capture geospatial information while reducing HSI redundancy. First, a multiagent (MA) key information selection module, MA-S ² -MMS-PPO, is designed based on proximal policy optimization (PPO). It fuses spectral–spatial (S ² ) correlation features extracted from Gramian angular field (GAF)-encoded images with historical actions to construct a multimodal state representation (MMS), and incorporates an efficiency-oriented composite reward mechanism to guide multiple agents in differentially selecting AGC-sensitive bands. Then, they are input into the GC-bidirectional cross-attention fusion (BCAF) module, which integrates geospatial features from the graph attention network and spectral image features from the convolutional neural network (CNN) through BCAF. In the methodological validation experiment, the spectral branch constructed from the 42 bands selected by MA-S ² -MMS-PPO achieved the best performance (mean five-fold R ² = 0.7394), while fixed-band-number comparisons further verified the effectiveness of the MMS-based MA mechanism. Furthermore, using the 46 key bands selected exclusively based on the modeling set, GC-BCAF achieved R ²=0.8348 on the independent spatial test set. KIG-GeoSIF enhances geospatial information utilization through key-band guidance to improve AGC estimation, providing scientific support for carbon governance.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: