Assessment of karst rocky desertification from the local to regional scale based on unmanned aerial vehicle images: A case‐study of Shilin County, Yunnan Province, China

Abstract

Karst rocky desertification is a major ecologic and geologic problem in Southwest China that has restricted the sustainable development of society and the economy. Many methods have been used to evaluate rocky desertification based on satellite remote sensing, but the results of these methods are affected by the heterogeneous surroundings of the karst region and the low resolution of sensors. In this study, a new method that combines satellite images and unmanned aerial vehicle (UAV) images was used to quantitatively extract information and evaluate rocky desertification. First, we extracted the bare rock ratio from the local high‐resolution UAV images, and then the regression models were established between the bare rock ratio of UAV images and the band reflectivity and eight rock indices of satellite images to invert the bare rock ratio at the county scale. The results showed that the overall accuracy and F1 of the classification of UAV images was 97% and 90%, respectively. The linear regression model between the reflectivity of the pixels in band 2 of the LANDSAT image and the bare rock ratio extracted from the UAV images was the best (R2 = 0.86). In addition, our method in which rocky desertification was assessed by the inversion model based on UAV images and LANDSAT images was superior to the traditional approach based on vegetation coverage. These results suggested that we can extract information on rocky desertification based on high‐resolution UAV images and assess rocky desertification by the inversion model from the local to regional scale.

Extraction of the bare rock rate in counties of typical karst areas based on unmanned aerial vehicle images and satellite data, China

“…There were significant differences in the inversion of the BRR by the two methods. The multiband regression model‐extracted BRR was significantly lower than the NDRI‐derived BRR (Figure 12); notably, the NDRI‐derived BRR was overestimated, and this trend matches the conclusion of Dai's study (Dai et al, 2021). The mean BRR calculated by the NDRI was 40.09%, while the mean BRR calculated by our method was 28.54% (Table 2).…”

“…The development of UAV remote sensing technology and inversion models provides a new way to study rocky desertification, but this approach is laborious and costly for large‐scale areas. Due to the high accuracy of UAV images and the wide coverage of satellite data, UAV images can be combined with satellite data to extract better remote sensing information, which has been widely used (Dai et al, 2021; Di Mauro et al, 2015; Fraser et al, 2017; Gevaert et al, 2015). Researchers have also combined UAV images and satellite data for BRR extraction.…”

“…Dai verified that the evaluation method based on UAV images and LANDSAT data for BRR inversion was better than the traditional method based on the vegetation coverage. The results demonstrated that rocky desertification information could be extracted based on high-resolution UAV images and assessed from local to regional scales via inversion models (Dai et al, 2021). However, many researchers have only performed single upscaling experiments (such as the upscaling method from UAV to LANDSAT images), and no one has performed experiments using the stepwise upscaling method.…”

Monitoring and classification of karst rocky desertification with Landsat 8 OLI images using spectral indices, multi-endmember spectral mixture analysis and support vector machine

“…In this study, a comparison was made between the spectral indices and the DPM model in terms of the determination of karst rock desertification. This comparison revealed different results from the spectral indexes with a high accuracy rate, as in other studies in the DPM literature [7,45].…”

Machine learning algorithm for estimating karst rocky desertification in a peak-cluster depression basin in southwest Guangxi, China

“…sensing data used to monitor the degree of karst rocky desertification have transitioned from multispectral 13 to hyperspectral images 14,15 , and low resolution data have been replaced by high resolution data 16,17 .…”