Mozgeris, G. 2008. Estimation and Use of Continuous Surfaces of Forest Parameters: Options for Lithuanian Forest Inventory. Baltic Forestry 14(2): 176-184

Deriving of the continuous surfaces of key forest characteristics from satellite images and point-wise field forest estimates is considered in this paper as a general methodological research framework to integrate sampling based and conventional stand-wise inventory systems. Several non-parametric and parametric estimation methods (two-phase sampling with stratification, k-nearest neighbor and regression) and sources of remotely sensed data (SPOT 4 HRVIR, Landsat-5 TM, panchromatic aerial images) together with the information available from stand-wise forest inventories have been studied on a test area in Dubrava forest in the central part of Lithuania for the accuracy of pixel-level estimates of key forest characteristics. Integration of the information available from satellite images and limited number of field sample plots with the characteristics of forest compartments, derived during the conventional stand-wise inventory, reduced the root mean square errors of estimates of pixel-level forest characteristics by 9-41%. The idea to convert forest reflectance in the satellite image into the continuous surfaces of forest parameters and use such surfaces to construct spatial units corresponding to conventional forest compartments is discussed as well. The relative efficiency of segmentation, expressed by the minimization of variance of key characteristics within the forest block via delineation of compartments, improves by 13-33% if the continuous surfaces of forest parameters are used instead of original satellite image as an input.

Key words: forest inventory, satellite images, pixel-level forest characteristics, two-phase sampling, k-nearest neighbor, regression, segmentation