Stand density estimation based on fractional vegetation coverage from Sentinel-2 satellite imagery

Abstract

• A linear regression model based on phenological factors well explains the correlation between stand density and FVC. • A quick method to calculate stand density is proposed, namely calculating it with sentinel-2 data on hectare scale. • Ten of thousands of individual evergreen coniferous trees within 100 ha were identified and marked for modeling or validation. • Images with lower-than-canopy spatial resolution can be used in stand density calculation, according to this study. Given that forest stand density is an important parameter for studies of carbon, water, and energy cycles and a core indicator for forest management, it requires accurate mapping to better assess how it impacts the eco-environment. Unfortunately, the calculation of stand density has long relied on the identification of individual trees for small-scale fine mapping or on empirical methods for macro estimations of large areas, making it difficult to balance cost and accuracy. Thus, this work proposes a more efficient method to estimate absolute stand density (n/ha) based on the fractional vegetation coverage retrieved from remote sensing image by establishing the correlation between the large- and small-scale approaches from the perspective of the hectare scale. The study area covered by planted evergreen coniferous forests featuring Pinus tabulaeformis and Pinus sylvestris in Shaanxi province, China. Taking into account that FVC is made up of contributions from evergreen and deciduous vegetation, phenological factors were considered to minimize the influence of background deciduous vegetation in the forest. A Sentinel-2 satellite multispectral remote sensing image sensed at late November 2020 was selected to estimate stand density when the deciduous vegetation is negligible. The accuracy was verified by using WorldView-3 satellite image with high spatial resolution images. The regression relationship was established by using 22 sample plots, and the stand density was estimated with fractional vegetation coverage in other 108 sample plots. The result shows that mean absolute error and root mean square error was 41.69 n/ha and 117 n/ha for sample plots, respectively, and the relative estimation accuracy of the total number of evergreen coniferous trees in the entire study area reached 81.57%. We thus conclude that this proposed approach to estimate the absolute stand density using Sentinel-2 MSI data with spatial resolution of 10 m is a feasible way on the hectare scale.