Investigating causal relationships between grassland deterioration and climate and socioeconomic changes through time-series computational learning
Geography and Geology
Journal of Cleaner Production
The paper investigates how climate changes and socioeconomic transformations affected grassland productivity from 2000 to 2017 in counties in Inner Mongolia of China. The grassland productivity and climatic change variables included 162 readings (9 periods in the growing season × 18 years), while ten socioeconomic variables contained 18 yearly readings. The mismatch between the environmental observations and socioeconomic data compounds quantitative analysis of causal relations between them. We applied several time-series processing methods to generate five datasets. We fit them with conventional, dependent-lagged, and dynamic-lagged panel regression models to investigate which combinations augment causal relationships between grassland productivity, climate, and socioeconomic changes.
The results confirmed that the conventional model with the original data failed to reveal causal relationships between these factors. The panel regression models with the original data tended to intensify the causal effects of climate factors on grassland productivity. The models with the transformed datasets disclosed more socioeconomic variables showing causal relationships with grassland productivity. The dependent-lagged model with the empirical mode decomposition transformed data produced the highest R-squared value and lowest prediction error. GDP, state fixed asset investment, highway construction, grain plantation, and livestock density were primary factors affecting grassland productivity. In brief, the integration of time-series data mining and time-lagged panel analysis effectively unveiled the causes of grassland deterioration under climate and socioeconomic changes.
Link to Published Version
Zhou, C., Liang, J., & Xie, Y. (2022). Investigating causal relationships between grassland deterioration and climate and socioeconomic changes through time-series computational learning. Journal of Cleaner Production, 366, 132963. https://doi.org/10.1016/j.jclepro.2022.132963