Simulation of Spatial Distribution of Monthly Average Precipitation Driven by Temporal Variation Function

  Objectives  High-precision precipitation field is important data support for hydrological, meteorological, and environmental analysis, which directly affects the accuracy of relevant services.The traditional simulation of precipitation distribution mostly relies on the spatial driving factors but neglects the impact of the temporal characteristics of precipitation on its spatial distribution.

  Methods  We employ the average cumulative precipitation data from 83 national meteorological stations and 28 provincial meteorological stations in Hubei Province from 2015 to 2017. According to the correlation analysis, the arch height (C) and nugget value (C₀) of the theoretical semivariogram model of precipitation time series are introduced as the influencing factors. We apply the geographically weighted regression (GWR) model to simulate the monthly average precipitation distribution model of Hubei Province.

  Results  The results show that: (1)The time-series semivariogram curve of monthly precipitation at each station is consistent with the seasonality of precipitation. 25.3% of the theoretical variation function model of site time series become stable in 4 months, and 36.14% in 6 months.(2)C and C₀ of the theoretical variation function model of precipitation time series at each station are correlated with the annual average cumulative precipitation in December of 0.01 level (both sides). The average correlation coefficients are 0.745 and 0.526, which are greater than the influence of geographical location and elevation on precipitation.(3)The introduction of C and C₀ is helpful to improve the global goodness of fit and interpolation accuracy of GWR model. By comparing the GWR model with latitude and longitude and the GWR model with variation function features of time series theory, the three-year average overall goodness of fit increased from 0.852 to 0.912. The accuracy evaluation of interpolation in the verification set site showed that the mean absolute error, the root mean square error and the mean absolute percentage error all declined by more than 60% in three years.

  Conclusions  Therefore, the spatial-temporal weighted regression model with time series theory variational function features can obtain high precision precipitation simulation results and is more suitable for the estimation of precipitation spatial distribution in the regions with abundant historical precipitation data.