Synergistic use of Sentinel-1, Sentinel-2, and Landsat 8 in predicting forest variables
Gengsheng Fang,Hao Xu,Sheng-I Yang,Xiongwei Lou,Luming Fang
DOI: https://doi.org/10.1016/j.ecolind.2023.110296
IF: 6.9
2023-04-27
Ecological Indicators
Abstract:Forest variables measurement is indispensable to supervising the dynamic of forest carbon storage. As commonly quantified forest variables, canopy cover (CC), diameter at breast height (DBH), and tree height (TH) were the predicted objects in our study. Recently, remote sensing technology, with high efficiency and wide temporal and spatial coverage, has been extensively used in practice. However, integrating open access satellite resources Sentinel-1 (S1), Sentinel-2 (S2), and Landsat 8 (L8) to quantify forest CC, DBH, and TH has not been extensively examined in the literature. In this study, we explored the potential of individual satellite bands/indices and the integration of different sensors in predicting mean CC, mean DBH, and mean TH in the plots dominated by four tree species (total area 102,887.6 ha). We acquired the satellite patterns of S1, S2, and L8 covered our research plots in 2017, and extracted the mean pixel values of the spectral images for every plot we detected. Specifically, seven sets of features derived from the satellite images of S1, S2, and L8 were used to predict CC, DBH, and TH for the survey plots. The results mainly indicated that group S2 + L8 (RMSEr (the relative root mean square error) = 17.9%, 21.0%, 23.0%) or S2 + L8 + S1 (RMSEr = 17. 8%, 21.0%, 23.1%) achieved the best accuracy for CC, DBH, and TH of all plots. Additionally, group S2 + L8 outperformed the other groups within two sensors and was superior to the individual sensors. Adding S1 synthetic aperture radar (SAR, the sensor carried by S1) had no evident improvement or deviation for most models, nevertheless, numerous bands/indices of SAR occupied important positions in variable importance and were been selected by stepwise regression, which denoted that it is available and valuable for forest parameters of some dominant tree species. Moreover, shortwave infrared (SWIR) bands played a crucial role in the three forest parameters prediction during the rainy season. SWIR bands (RMSEr = 20.5%), NIR bands (RMSEr = 24.1%), and red-edge bands (RMSEr = 27.1%) exerted the most significant implication for predicting CC, DBH, and TH apiece as univariate models. Green bands and NIR also occupied more positions of higher variable importance in multivariate models for estimating DBH and TH, while SWIR took up more for CC in all dominant species. Furthermore, non-parametric models outperformed the parametric model on the whole. In the meanwhile, the best models also were influenced by the data analysis methods. In a nutshell, a combination of multiple satellite sensors was proposed for forest variable selection, especially different optical sensors. SAR data were also inspired to be pitched in related forest measurement work, but it requires some data processing before applying to the research in many cases.
environmental sciences