Objective Large-scale and accurate monitoring of pine wilt disease (PWD) plays an important role in forest epidemic prevention and management. Existing studies often use single-phase data, resulting in greater uncertainty in the accuracy of monitoring PWD, which is easily affected by forest background and non-host trees. In addition, single data have the limitation of insufficient characterization of disease characteristics. For example, passive optical data focus on describing the horizontal structure of forest canopy, and are easily affected by cloud and rain, resulting in missing data; Active microwave data are sensitive to forest vertical structure and moisture content, but there are problems such as high noise, low pigment sensitivity and large terrain impact. Therefore, the combination of active microwave and passive optical time-series images is expected to reduce the impact of environmental factors while tracking the time-series change characteristics of the same forest stand, which helps to improve the accuracy and robustness of PWD monitoring.

Method First, using centimeter-level resolution drone images to obtain samples. Based on extreme gradient boosting algorithm, PWD monitoring models were constructed by combining Sentinel-1 C-band microwave and Sentinel-2 optical time-series images. The performance of microwave model, optical model, combined microwave and optical model in the monitoring of PWD was evaluated respectively. At the same time, compare the performance of the optimal model under different environment conditions.

Result (1) The accuracy of combined microwave and optical model (overall accuracy = 80.62%, Kappa coefficient = 0.61) was slightly higher than that of the single optical model (overall accuracy = 79.58%, Kappa coefficient = 0.59), but significantly higher than that of the single microwave model (overall accuracy = 68.87%, Kappa coefficient = 0.36), and its showed the value of combined microwave and optical time-series data in the monitoring of PWD. (2) The analysis of different environment conditions showed that the model generally exhibited higher accuracy under gentle slope, sunny slope, low altitude, and high coverage conditions.

Conclusion This study makes full use of multi-source remote sensing satellite data and provides new technical support for large-scale accurate monitoring of PWD.