Forecasted reference sample plot data in estimations of stem volume using satellite spectral data and the kNN method

When applying remote sensing in forestry, there is often a need for reference sample plot data as a link between what is seen in the image and the actual ground conditions. In many cases such data exist, although it may be old and need to be updated. In this study, the effects on the estimation accuracy of using forecasted reference sample plot data when estimating stem volume in forests are investigated. The k nearest neighbour (kNN) estimation method is applied, using spectral information in a satellite image and sample plot reference data forecasted up to 25 years. The study is based on simulation. The basic components of the simulation model are, however, derived from empirical data. The mean square error (MSE) of volume estimates was found to increase with increasing length of the forecasts, although after excluding disturbed plots (e.g. due to thinnings) from the reference material, use of old reference data led to a very modest decrease of the accuracy. Thus, if disturbed plots can be identified, the results point towards the possibility of using updated old reference data as a means to increase the cost-efficiency in this kind of inventory. However, the results are shown to depend on the correlation between ground state and satellite spectral data. When this correlation is higher, the relative increase in MSE with forecasting length was higher, as when disturbed plots had been removed from the reference material.