Needle litter of
Pinus yunnanensis, the main surface fuel of wildfire in the central Yunnan Province, China, is highly flammable for its high oil content and low moisture. We monitored the moisture contents of surface fuels of
P. yunnanensis forests in central Yunnan during the fire prevention period in 2020. Correlation analysis, common factor variance, variance inflation factor (VIF), and multiple prediction regression model were used to explore the relationships between topographical, meteorological, and stand factors and the moisture contents of surface fuels. The model coefficients were adjusted by deviation standardization method, and the model accuracies were evaluated. The results showed that the factors affecting surface fuel moisture of
P. yunnanensis forests in descending importance were temperature, humidity, wind speed, slope direction, canopy density, slope, elevation, and wind direction. The VIFs of slope direction and canopy density were more than 10, showing a high degree of collinearity. Therefore, regression model
E1 was constructed using temperature, humidity, wind speed, slope, and elevation. The average goodness of fit for moisture of
Yk1,
Ys1, and
Yg1 was 74.35%, and the average error rate was 32.06%; the symbols k, s, and g were branch litter of pine, needle litter of pine, and shrub twig litter and grass litter, respectively. An enhanced regression model
E2 was reconstructed with temperature, humidity, and wind speed as independent variables, which showed significant correlations with the target (
r>0.70). The average goodness of fit of
Yk2,
Ys2, and
Yg2 was 83.99%, and the average error rate was 17.09%, which outperformed those of
E1. Slope direction, slope, elevation, and canopy density, which showed insignificant contributions to surface fuel moisture, were selected as adjustment elements and converted as correction coefficients by the deviation standardization method, to reconstruct the regression model
E3. The average goodness of fit was 89.72%, and average error rate was 8.48% for
Yk3,
Ys3,
Yg3. The accuracy of
E3 outperformed
E1 and
E2, with the mean goodness of fit of
Yk3,
Ys3,
Yg3 being improved by 9.69% vs 2.11%, 4.84% vs 10.77%, and 8.41% vs 4.33%, and the error rates being reduced by 15.65% vs 6.89%, 11.24% vs 13.69%, and 18.01% vs 5.24%, respectively. The added correction coefficients can improve the prediction accuracy of the model, and the model factors are easy to obtain. These findings are useful for forest fire prevention and management, and provide technical support for rapid, accurate, and real-time prediction of surface fuel moisture.
