| Accurate and real-time local canopy structure data sets, including canopy height and fractional cover, are required to monitor grassland health and aboveground biomass (forage availability) as well as to provide the validation data for satellite remote sensing products; however, most emerging remote sensing techniques, such as vegetation indices-based optical remote sensing, are only a reflex of the fractional cover information. They are incapable of directly detecting the canopy height information, are thus subjected to the severe saturation problem associated with estimating biomass with a high canopy cover. In recent years, the emergence of the low-cost light unmanned aerial vehicle (UAV) and portable optical camera and laser radar (Lidar) sensors have paved the way for UAV-based remote sensing applications in environment monitoring and disaster relief. This also leads a tendency for the UAV-based remote sensing to replace the traditional aerial remote sensing. The coverage area of individual orthoimage is typically very small, and it is necessary to mosaic thousands of individuals to create a larger image to cover a big meadow area (> 1 km2). Thus, the application of the UAV-based remote sensing technology in meadow areas is seriously dependent on the efficiency and the automatization level of the mosaicking algorithms. In this research, a vector building-based seam detection method was proposed for quickly generating the seams for aerial imaginary mosaics. The proposed method is shown to find the lowest-weight candidate seam from vector building maps. The proposed method provides a positive reference for mosaicking thousands of individual images captured in meadow areas, and establishes a good foundation for estimating the grassland canopy height, fractional cover, and biomass from the UAV-based photographic data.In addition, non-growing seasons of grasslands typically account for nearly three quarters of a year. Grasses in arid and semi-arid environments are generally only 10-50 cm tall, and it is difficult to find feature points for optical imaginary matching. All the factors mentioned have restricted the application of the UAV-based remote sensing in grasslands. To this end, we also proposed a method for estimating the canopy height and fractional cover from the UAV discrete Lidar data, which were then combined to estimate the aboveground biomass. The experiments was conducted in the cattle-grazing plot of Hulunber Grassland Ecosystem Observation and Research Station (HGEORS), located at the center of the Hulunber meadow steppe in the northeastern region of Inner Mongolia. The UAV discrete Lidar technology is independent from the sunlight, vegetation chlorophyll and water content, does not need to find feature points for imaginary matching, and can directly detect both the canopy height and fractional cover. Therefore, the technology is far superior to the traditional optical remote sensing technologies in estimation of the non-growing season aboveground biomass at the plot scale. Furthermore, the influences of flight height and geolocation mismatch on Lidar estimates were analyzed, and an optimized Lidar experimental design was proposed to improve the accuracy of Lidar estimates in grassland surveys.Main findings presented in this report can be summarized as follows:(1) The vector building-based seam detection method can significantly accelerate the computation efficiency and improve the mosaicking quality. Experiments show that the number of crossed buildings using the proposed method is～10%-40% of those crossed by the vector-road based method, and ～1%-15% of those using Dijkstra’s algorithm. The total computation time of the proposed method is shown to be approximately 1-4 times that of the vector-road based method, but crucially～10%-20% that of Dijkstra’s algorithm. The proposed method has an important revelation to mosaicking the imaginary captured in meadow areas. The mosaicking efficiency can be improved by utilizing various vector data to guild the generation of the seams. For the areas where several flight experiments have been conducted, such as the HGEORS, the available vector data includes the historical seam data and various vector objects, including vector roads, building, fence boundaries, and other facility lands.(2) Although canopy height and fractional cover were underestimated in the Lidar data, particularly for areas with high canopy cover, strong correlations (R2>0.56 and RMSE< 107.7 g/m2) were observed in this study between the aboveground biomass and the Lidar-derived canopy structure indices. Combination of canopy height and fractional cover slightly improved the correlation to biomass, with R2 and RMSE values being 0.784 and 108.9 g/m2 (approximately 18.6% of the maximum aboveground biomass).(3) Flight height has a small influence on the derived canopy height, but has an obvious influence on the derived fractional cover and details of the data. The non-coincidence between the field quadrat centers and Lidar-derived metric centers also has a high influence on the accuracy of Lidar estimates. For example, the correlation between the lidar-derived and field measured mean canopy heights, when using RTK GPS and search distance d≤4 m, was 0.033 higher than that of the portable GPS and the RMSE was 0.629 lower than that of the portable GPS on average. Use of real time kinetic (RTK) global positioning system (GPS) or a proper Lidar data sample window can reduce the influence of geolocation mismatch. The recommended window side length is equal to the bigger horizontal coordinate RMSE between RTK GPS and Lidar data where most Lidar estimates are stable.The follow-on studies will include:(1) Applying the revised vector building-based seam detection method to mosaic the UAV-based low attitude imaginary for meadow areas and developing algorithms for estimating the canopy height, fractional cover, and biomass from the UAV photographic data.(2) Developing a Lidar sensor with a small laser beam divergence and short transmit pulse width to estimate the canopy height, fractional cover, and biomass accurately, and determining the optimal flight height for grasslands with different fractional cover.(3) Developing algorithms for estimating grassland biomass using spaceborne SAR, hyperspectral, and multispectral remote sensing data, and validating the results using the UAV remote sensing estimates. This will significantly improve our ability to monitor the grasslands at any time (day and night), in any weather (sunny and cloudy), and at any scale (plot scale to national scale). |
PDF Full Text Request