Work

Vulnerability evolution and interactive feature attributions of vegetation productivity in the Jinsha river dry-hot valley using physics-guided neural networks

Year: 2026

Type: article

Abstract: Vegetation productivity dynamics in ecologically fragile alpine-canyon regions reflect ecosystem resilience under intense environmental stress. Taking the Jinsha River dry-hot valley as the study area, this research integrated a multi-dimensional geospatial database covering 2000 –2025, with near-infrared reflectance of vegetation as the core proxy for vegetation productivity. A physics-guided neural network and SHAP framework were employed to evaluate productivity evolutionary trajectories, model simulation advantages, and feature contributions. Regional vegetation productivity exhibited an overall upward trajectory increasing from 0.096 to 0.114, characterized by three distinct phases including an initial instability period, a prolonged recovery phase, and a recent stabilization stage. Positive growth areas accounted for 83.24% of the total domain, whereas degrading regions formed linear low-value corridors along main river channels with a mean coefficient of variance of 0.2168. Incorporating empirical ratio vegetation index-based saturation constraints into the ResNet framework significantly improved simulation robustness, increasing R² values by approximately five percentage points across the evaluated periods. Furthermore, SHAP dependency analysis suggested a potential shift in feature contribution patterns from a persistent elevation-related background toward stronger multi-variable meteorological associations involving vapor pressure deficit and temperature during some later periods. This pattern may indicate increased sensitivity of lower-elevation environments, although any inference of a vertical vegetation shift requires independent validation.

Source: Ecological Informatics

DOI: 10.1016/j.ecoinf.2026.104042

Authors: Xiong Duan, Yu Yu, Haiying Wang

Institutions: China West Normal University, Chengdu University of Technology, State Key Laboratory of Oil and Gas Reservoir Geology and Exploitation

Language: English

Topic: Remote Sensing in Agriculture

Subfield: Ecology

Field: Environmental Science

Domain: Physical Sciences

SDGs: Life on land, Climate action

Open Access status: gold

Estimated APC paid: $2,770