This study combined artificial intelligence and image recognition technologies to construct a machine learning predictive model based on image features was constructed to achieve rapid, objective, and accurate estimation of surface fuel loading in coniferous forests, providing data support for refined forest fire prevention, forecasting, and related endeavors.

Taking the needles of typical Pinus massoniana forests in Guizhou Province were selected as the research object. The actual variation range of fuel loading in the needle litter layer was determined through setting up sample plots and conducting random quadrat surveys. In the laboratory, needle litter layers with different fuel loading (30 cm×30 cm) were constructed, and each fuel loading was photographed three times repeatedly, resulting in a total of 150 vertically oriented images. Image features were extracted using OpenCV-Python. Following Z-score standardization and principal component analysis (PCA) for feature selection, three machine learning methods were employed to construct prediction models for needle litter layer loading, and the model performance was evaluated.

1) Field measurements revealed that the fuel loading of the needle litter layer ranged from 3.9-16.5 t·hm^-2. Among the shape features extracted from images of litter layers with different fuel loading, the maximum perimeter value reached 600 223.95, while the edge density in edge features was the smallest, with a value of 4.11; 2) Severe multicollinearity (VIF>10) was observed among all eight image feature values; 3) The first three principal component scores (PC1, PC2, and PC3), extracted via PCA, were used as new independent variables. The fuel loading showed statistically highly significant linear correlations with PC1, PC2, and PC3, indicating their suitability for model construction; 4) Machine learning prediction models were constructed using the training set and validated with the testing set. The K-nearest neighbor (KNN) model achieved the best prediction performance, with a mean relative error (MRE) of 18.64%, mean absolute error (MAE) of 2.29 t·hm^-2, root mean square error (RMSE) of 3.31 t·hm^-2, R-squared (R²) of 0.79. Its low Jensen-Shannon divergence (JSD) value of 0.005 indicated highly similar distributions between predicted and true values. The random forest regression (RFR) model performed suboptimally (MRE=39.03%, R²=0.57), while the multiple linear regression (MLR) model yielded the worst prediction results.

The machine learning model constructed based on image features proved feasible for estimating the fuel loading of the needle litter layer. It overcame the drawbacks of traditional measurement methods, such as being time-consuming, subjective, and low in accuracy, achieving rapid and objective estimation of fuel loading. This approach provided a novel methodology for research on forest surface fuel loading and held significant implications for forest fire forecasting and scientific management.

Surface fine fuels are critical ignition materials for forest fires, and the accurate prediction of their moisture content is essential for fire prevention. The fine fuel moisture code (FFMC), as a typical semi-physical model, is widely used for moisture content prediction. This study aims to evaluate the applicability of directly using FFMC in predicting the moisture content of surface fine fuels in Pinus massoniana forests and to improve the model to enhance prediction accuracy.

Focusing on typical Pinus massoniana forests in Guizhou, the dynamic changes in the moisture content of surface fine fuels were monitored hourly during the fire prevention period, and their driving factors were analyzed. The hourly FFMC values were calculated. Through comparative analysis, the applicability of directly using hourly FFMC in predicting the moisture content of Pinus massoniana forests was evaluated, and the model was improved to test its prediction accuracy.

1) The dynamic changes in moisture content are mainly influenced by meteorological factors, with air temperature and wind speed showing a significant negative correlation with moisture content, while relative humidity shows a significant positive correlation; 2) When the moisture content of surface fine fuels in Pinus massoniana forests is below 35%, directly using FFMC for prediction shows good applicability, with mean absolute error and relative error of 0.88% and 4.81%, respectively. However, when the moisture content exceeds 35%, directly using FFMC is not applicable, with a mean relative error as high as 66.37%; 3) A stepwise regression method was selected to establish a moisture content (>35%) prediction model based on FFMC and meteorological factors, significantly improving prediction accuracy. The improved model's mean absolute error and relative error were 15.81% and 21.41%, respectively.

This study validates the applicability of using hourly FFMC to predict the moisture content of surface fine fuels in Pinus massoniana forests and enhances prediction effectiveness through model improvement. The research findings are significant for a deeper understanding of the moisture dynamics of surface fine fuels in Pinus massoniana forests, supporting fire risk forecasting, and scientifically advancing forest fire prevention and suppression efforts.

As the igniter of forest fire, the loading of fine dead fuel on the forest surface determines a series of fire behaviors and the risk degree of fire occurrence. It is of great significance for forest fire risk prediction and scientific forest fire management to quickly and accurately obtain the loading value of fine dead fuels on the surface.

In this study, the fine dead fuel on the surface of 6 typical forests (Pinus massoniana Lamb. forest, Cupressus funebris Endl. forest, Soft broad forest, Cunninghamia lanceolata forest, Broadleaf mixed forest and shrub forest) in Dalou Mountain in the southwest forest region were taken as the research object. Through field investigation and indoor drying, the forest characteristics and loading values were measured to obtain the influencing factors of the fine dead fuel on the surface, and a prediction model was established.

1) The fine dead fuel loading of different forest types was significantly different, and that of Pinus massoniana Lamb. and Broadleaf mixed forest was significantly higher than that of other forests. The difference in carrying capacities between different standard plots of Pinus massoniana Lamb. forest and broadleaf mixed forest was significant, and there was no significant difference in carrying capacities among different standard plots of other stands; 2) Except for Cunninghamia lanceolata forest and shrub forest, the surface fine dead fuel loading of other forests had a significant correlation with some forest characteristic factors (forest age, canopy density, mean diameter at breast height, mean tree height, mean diameter at ground level, slope and elevation). The forest density and average crown width had no significant impact on the surface fine dead fuel loading of the study area; 3) The prediction model of the surface fine dead fuel loading of Pinus massoniana Lamb. forest, Cupressus funebris Endl. forest, soft broad forest and Broadleaf mixed forest was established. The prediction effect of Cupressus funebris Endl. forest was the best, with an error of only 3.6%. The prediction error of Pinus massoniana Lamb. forest was as high as 27.4%, which could not be applied in practice.

Through this study, the basic data of the surface fine dead fuel loading of typical forest stands in Dalou Mountain and the influencing factors and prediction models have been obtained, which is of great significance for the study of the regional loading and scientific forest fire management.

As an important part of forest fuel, shrub have a significant impact on the vertical burning of forests and the possibility of extreme fires. As the second largest forest region in China, the southwest forest region has the highest annual fire frequency in China. It is of great significance for forest fire prevention and forest fire management to investigate the shrub loading of typical forests in this region, analyze its influencing factors, and establish prediction model.

In this study, shrubs in six typical forest stands in Dalou mountain were taken as the research object, and standard plots were set up in the field to investigate the basic information. Through field experiments and indoor experiments, the aboveground stems, branches, leaves and total loading of shrubs were obtained, and their influencing factors were obtained and prediction models were established respectively.

1) Through the study on different standard plots of the same forest type, it was concluded that the shrub branch loading of Pinus massoniana forest and the shrub leaf loading of Cupressus forest had significant differences, while the shrub branch loading and leaf loading of broad-leaved mixed forest had significant differences, and the dry loading and total loading had extremely significant differences; There was no difference in stem, branch, leaf and total loading among different forest types. 2) The stem, branch, leaf and total loading of shrubs in the same forest type were correlated with the average tree height, average ground diameter, average height, average diameter at breast height, stand density, slope, elevation, etc. (the correlation gradually weakened); The stem, branch, leaf and total loading of shrubs in different forest types were correlated with the average ground diameter, average tree height, average diameter at breast height, average height, slope and average crown width (the correlation gradually weakened). 3) Establishment of models: the fitting effect of the model of stem and branch loading of Cunninghamia lanceolata (Lamb.) forest and the model of stem, branch and total loading of shrub forest (R²=0.998) was the best, while the fitting effect of the model of branch loading of P. massoniana forest is the worst (R²=0.318).

This study provides basic data and methods for the investigation of shrub loading in typical forests in Dalou mountain area, and is of great significance for the investigation of fuel loading in forest fire risk prediction and forest fire risk survey.

The loading of forest surface litter affects the occurrence of forest fires and a series of fire behavior characteristics exhibited by forest fires. Accurately obtaining the loading of surface litter is crucial. The Euler number of image feature can characterize the number of objects in the image, analyze the relationship between Euler number and loading, and establish a load prediction model based on image Euler number, which is of great significance for load research.

The litter in typical forest stands of Cryptomeria fortunei and Phyllostachys heterocycla in Guizhou province was taken as the research object. Through forest stand and loading investigation, taking litter images and image feature processing, the relationship between Euler number and surface litter loading was analyzed. A load prediction model based on image Euler number was established, and the accuracy of the model was tested.

1) After selecting different thresholds for image binarization, not all extracted Euler numbers were correlated with the litter loading. A threshold of 0.1 showed a highly significant correlation between the Euler numbers of binarized images and the two types of litter loading; 2) As the Euler number of the image increased, the surface litter loading of forests of C. fortunei and P. heterocycla showed an overall downward trend; 3) Linear regression was chosen to establish a litter loading prediction model based on image feature Euler number. The absolute errors of the prediction models for the litter load in C. fortunei and P. heterocycla forests were 1.60 t·hm^-2 and 1.72 t·hm^-2, respectively, with mean relative errors of 20.03% and 20.71%. The predicted effect of surface litter loading in C. fortunei forest was better than that in P. heterocycla forest.

Through this study, the feasibility of predicting forest surface litter loading based on image features has been preliminarily verified, providing new ideas for accurately obtaining load research and of great significance for scientific forest fire management.