Mapping forest age and fine classification of forestland using surface spectral endmember space

Forestland has been one of the most important ecological land and natural resources in modern agriculture. The quantity, type, spatial layout, and age structure of forestland can have a significant impact on its functional performance. There is a high demand to accurately and rapidly acquire forest information at a regional scale. Fortunately, a linear spectral mixture model can be expected to avoid the influence of mixed pixels on the spectral information. This study aims to carry out the fine classification of forestland and forest age mapping using a linear spectral mixture model combined with a random forest and logistic model. The study area was taken as the Zhangjiakou City, Hebei Province, China. Landsat series remote sensing images were also captured in 2020 using surface spectral endmember space. The results showed that: 1) The forestland was divided into five types: evergreen coniferous, deciduous shrub, deciduous small-leaved, deciduous broadleaved, and deciduous needle-leaf forest. The overall high accuracy of classification was 90.58%, and the Kappa coefficient was 0.89. Furthermore, there were significant differences in the area occupied by various forestland types and their distribution in the horizontal space. Among them, the deciduous shrub forest shared the widest distribution area in the study area, accounting for 31.25%. The second was deciduous broadleaved forest, accounting for 3792.80 km². The smallest distribution area was found in the deciduous needle-leaf forest, accounting for only 480.32 km². 2) The classification of forest age was validated to compare the field sample point data and high-resolution images from the Google Earth platform. The average absolute error and deviation of the forest age mapping were within two and three years, respectively, indicating the more effective mapping. There was a significant variation in the spatial distribution pattern of each age group. Forestland with an age of 20-30 years shared the largest distribution area among the five age groups, with an area of 3704.75 km². While the forestland with an age of more than 30 years had the smallest area, accounting for only 119.98 km². 3) The distribution areas of different forestland types showed a tendency to increase and then decrease with the increase in elevation and slope. Among them, the evergreen coniferous forest was primarily distributed at an elevation range of 1500-2000 m. Deciduous shrubs and deciduous needle-leaf forests were both concentrated within the elevation range of 1000-1500 m. Deciduous small-leaved and deciduous broadleaved forests were aggregated within the elevation ranges of 1000-1500 m and 1500-2000 m, respectively. In terms of slope distribution, the deciduous small-leaved forest predominantly grew with a gradient of 2-6°, while the rest were primarily concentrated on the gradients of 6°-15° and 15°-25°. As for different forest ages, the area of each age group reached the highest level in the elevation band of 1000-1500 m. The age groups of 5-10 years and 10-20 years shared the largest distribution area in the slope zone of 15°-25°. And the forest age of 20-30 years occupied the largest area in the slope zone of 6°-15°. The area of each age group was smaller than that in the flat land with a slope of <2°. This finding can provide a visual basis and technical support to the spatial distribution and age structure of regional forestlands. It was of great significance to optimize the forest resources and ecologically sustainable construction.