Statistical modeling experiment of land precipitation variations since the start of the 20th century with external forcing factors

Abstract

Fitting of the historic climate change with multiple linear regression (MLR) models often encounters the problem of residual autocorrelation and multiple correlation among the dependent factors, which leads the model not to be the Best Linear Unbiased Estimator (BLUE). To solve these problems, a set of sophisticated statistical modeling procedures are developed in this paper to detect the contributions of the external forcing to the global and regional land precipitation changes during the recent century. The radiative forcing (effective radiative forcing) data of multiple natural and anthropogenic factors are taken as independent variables and a combined model of MLR and Autoregressive Integrated Moving Average (ARIMA) is first used to separate the impacts of natural forcing and human activities (signals) and internal variability (noise) due to land precipitation changes since the start of the 20th century. Based on the results, a partial least squares regression (PLSR) model is further adopted to quantify the respective contributions of the each components of anthropogenic forcing. Some preliminary conclusions were drawn: (1) The combined model of MLR and ARIMA has clearly separated the influence of anthropogenic forcing signals in the global and regional (middle and high latitude in Northern Hemisphere) land precipitation anomaly at 5% significance level, and the explained variance is relatively large. (2) The PLSR model overcomes the morbidity of ordinary linear regression equations to a certain extent, and better separated the respective impact (contribution) of various natural and anthropogenic forcing factors on global and regional (middle and high latitude in Northern Hemisphere) land precipitation anomaly; and the relative strength of each factor’s contribution are determined. It has good application value for modeling and impact studies with multiple correlations among different independent variables. (3) The above combined model explains about 40% of the total variance of land precipitation anomalies change over global and middle latitude regions in Northern Hemisphere, and nearly 60% of the total variance in high latitude regions. Based on the fitting results, both the human activities and natural forcing are the deterministic impact factors of global land precipitation variations. But the natural forcing’s contribution is not significant, while human activities still explain a large part of the variance and thus have high significance for the regional land precipitation variations. In addition, the autocorrelation should also be considered as one of the important impact factors for global and regional land precipitation anomaly changes. (4) The anthropogenic aerosol and contrails have clear and significant positive contributions to the global and regional land precipitation anomalies, while the contributions of the remaining factors show obvious positive and negative differentiation, indicating that there is certain degree of uncertainty. Therefore, it is necessary to further optimize and improve the factors and models based on more accurate factor datasets for each region. (5) The modeling approach described above is primarily based on the idea to decompose and model the historical climate series. On one hand, it may put forward certain requirements for the “integrity” of the observational data; while on the other hand, it is not restricted by the study domains, climate variables and the independent variables. Moreover, it may even model/separate the anthropogenic/natural effects of climate changes in local scales. In addition, it can independently detect and analyze the relative contribution and significance of human activities or natural forcing to climate change in the historical observational series without using the climate system model. Therefore, it is proved to be a convenient approach for detecting the causes of climate change and a powerful supplement to model attribution.