Forests, Год журнала: 2024, Номер 15(12), С. 2100 - 2100
Опубликована: Ноя. 27, 2024
With climate change and the intensification of human activity, drought event frequency has increased, affecting Gross Primary Production (GPP) terrestrial ecosystems. Accurate estimation GPP in-depth exploration its response mechanisms to are essential for understanding ecosystem stability developing strategies adaptation. Combining remote sensing technology machine learning is currently mainstream method estimating in ecosystems, which can eliminate uncertainty model parameters errors input data. This study employed extreme gradient boosting, random forest (RF), light use efficiency models. Additionally, we integrated solar-induced chlorophyll fluorescence (SIF), near-infrared reflectance vegetation, leaf area index (LAI) construct various The standardised precipitation evapotranspiration (SPEI) was utilised at timescales analyse relationship between SPEI during dry years. Moreover, potential pathways coefficients environmental factors that influence were explored using structural equation modelling. Our key findings include following: (1) combining SIF RF algorithms exhibits higher accuracy applicability vegetation arid zone Xinjiang, with an overall (MODIS R2) 0.775; (2) Xinjiang had different characteristics drought, optimal timescale respond 9 months, a mean correlation coefficient 0.244 grass land SPEI09, indicating high sensitivity; (3) modelling, found temperature affect both directly indirectly through LAI. provides reliable tool methodology conclusions important references similar environments. In addition, this bridges research gap timescales, mechanism natural on scientific basis early warning management. Further validation longer time series required confirm robustness model.
Язык: Английский
