Background The remote sensing image obtained with unmanned aerial vehicle (UAV) has been widely used in soil and water conservation monitoring. However, compared with other industries, there are shortcomings in the UAV remote image, such as inadequate research depth, few application functions, and low monitoring accuracy. Moreover, images shot with different cameras and application of different classification methods will have impact on the monitoring accuracy of forest and grassland coverage in the monitoring process. The objective of our study is to provide a rapid and accurate method to monitor the coverage rate of forest and grass.

Methods This study conducted a case study in Kenli, Dongying, Shandong province. The forest and grass coverage was calculated based on the multispectral data of UAV, and then the coverage was compared with the vegetation information extracted from the visible light. The visible light and multispectral images with high-resolutions were simultaneously obtained via UAV equipped with visible light and multispectral cameras having 5 multispectral sensors. Each camera was equipped with the same spectral resolution of 2 megapixels. Sixteen vegetation indices, including 8 multispectral vegetation indices and 8 visible light vegetation indices, were established, and the object-oriented threshold method and support vector machine method were used to extract the vegetation information and calculate the forest and grass coverage, respectively. Finally, the optimal vegetation index and classification method were chosen via the confusion matrix.

Results The accuracy of vegetation index identified by multi-spectrum was over 90%, and the Kappa coefficient was over 0.90. Three types of visible light vegetation indices had reached the above level. The 11 vegetation indices mentioned above met the requirements of practical applications in soil and water conservation monitoring for the production and construction projects. In the classification method of multispectral vegetation indices, the available support vector machine methods were used in green ratio vegetation index (GRVI), soil-adjusted vegetation index (SAVI), green normalized difference vegetation index (GNDVI) and normalized differential index with red edge (NDRE), while the threshold methods were used in normalized difference vegetation index with red edge (RENDVI), enhanced vegetation index2 (EVI2), normalized difference vegetation index (NDVI) and optimized soil-adjusted vegetation index (OSAVI). The threshold methods were all used in the visible light vegetation indices, including red (R), green (G), and excess green vegetation index (EXG).Confirmatory experiments in three study areas showed that under normal conditions, better effects of vegetation information identification in calculating the forest and grass coverage might be obtained under both of the multispectral and visible light vegetation indices. In the presence of shadows, the single-band visible lights R and G were not well classified. On the contrary, the multispectral vegetation index presented better applicability and stability compared with the visible light covered index.

Conclusions This study provides a high-precision extraction method to monitor the forestry and grass coverage for soil and water conservation based on UAV visible light data and multispectral data.