Modelling the Spatial Distribution of Lake Surface Water Temperature of the Thaw Lakes in Arctic Coastal Plain Using Geographically Weighted Regression Model

A Sub-region of the Arctic coastal plain of Alaska was divided according to latitude and distance from coastline, and were prepared for the selection of spatial impact factors that influence the average lake surface water temperature (LSWT). After analyzing the relationship between each factor with LSWT by isolating the other factors, the factors would be recalculated via logarithm or exponent transformation in order to satisfy a linear relationship if the relationship is nonlinear. The most related factors including lake area, compactness index, mean depth, the distance to Chukchi Sea, the distance to Beaufort Sea and latitude were used to construct the LSWT spatial distribution model. To decrease the spatial non-stationarity of the models, the principal component analysis was applied to eliminate the effect of multicollinearity among variables. Then the LSWT spatial distribution models were constructed by ordinary least squares regression and geographically weighted regression method, respectively. The validation results show that the accuracy of geographically weighted regression model improved compare with ordinary least squares regression model. The coefficient of determination of the geographically weighted Regression model, R-square, is promoted from 0.615 to 0.752. And compared with the OLS model. The MAE and RMSE of GWR model decreased from 0.48 to 0.38 and from 0.65 to 0.44, respectively. It demonstrates that the improved GWR model can moderately depict the spatial distribution of thawed lake surface water temperature on the arctic coastal plain of Alaska.