Estimation of forest variables using satellite image data and airborne Lidar

Remote sensing data can be used to estimate forest variables or to classify areas into groups of classes. Presently, remote sensing sensors can be used for a variety of different purposes, and the number of sensors is constantly increasing. The development of new sensor systems that are either satellite-borne or airborne, and the development of the Global Positioning System, are both key issues for the future use of remotely sensed data in forest inventories. The main objectives are to investigate the possible use of different sensors and methods for estimation of forest variables in the Swedish National Forest Inventory (NFI). Five studies concerning airborne lidar, NOAA AVHRR data and Landsat TM data are reported. Three different study areas were used: one located north of Umeaa (Lat. 64 deg 15 min N, Long. 20 deg 50 min E), one located south of Stockholm (Lat. 58 deg 56 min N, Long. 18 deg 15 min E) and one loacted in the south of Sweden (centered at Lat. 57 deg 50 min N, Long. 14 deg 20 min E). In all areas, the forest is dominated by Scots pine (Pinus sylvestris) and Norway spruce (Picea abies). Laser measured tree heights were found to underestimate field measured tree heights by 2.1-3.7 m on a plot level, depending on the laser footprint size. A linear regression function relating the laser measurements to stand volume was derived. A method whereby Swedish NFI plots, a Landsat TM image, and a NOAA AVHRR image were used to assess forest estimates across large areas was evaluated. For three out of the four counties situated partially within the TM scence, the mean estimates for most variables tested were found to be within 10% of the estimates obtained using only NFI plots. Two of the studies concerning Landsat TM data were designed to evaluate the "k nearest neighbour estimation method". In this method, multiple variables are simultaneously calculated as weighted mean values of spectrally nearby samples. It was found that at least 5-10 nearby samples should be used when estimating wood volume, pine volume, mean diameter, and stand age, in order to get a sufficiently high estimation accuracy at a pixel level. It was also found that only one nearby sample should be used if the correlation between the forest variables was to be maintained. In the final study, the effects of combining Landsat TM data and tree height measurements were investigated. The results show that the estimation error (RMSE) for wood volume decreased from 56 m3 to 37 m3/ha when Landsat TM data was combined with field measured tree height data, rather than using TM data alone. The studies show that remote sensing data can be integrated in the Swedish NFI to estimate forest variables such as wood volume on both local and regional levels.