The invention discloses a construction method of a spatial variable coefficient PM2.5 concentration estimation model based on a Re-ESF algorithm. The construction method comprises the following steps:step 1, multi-source AOD product fusion; step 2, data processing of related influence factors and PM2.5 concentration; step 3, obtaining independent variable values and PM2.5 concentration values ofall parts of the monitoring site; step 4, constructing an adjacent matrix; step 5, centralizing the spatial adjacency matrix and calculating a feature value and a feature vector; step 6, solving a regression coefficient of an SVC model based on the Re-ESF; step 7, judging whether variables in the model are remarkable or not; Step 8, model precision evaluation; step 9, 10-fold cross validation; step 10, judging the relation between the number k of neighborhoods and the number n of monitoring station points; and step 11, selecting an optimal model according to the precision evaluation. On the basis of effectively eliminating the influence of spatial heterogeneity and spatial autocorrelation on PM2.5 concentration modeling, a random effect and spatial variable coefficient method is introduced, so that the precision of the concentration estimation model is further improved, and the model resolving time is shortened.