Data assimilation in stand-level forest inventories

Abstract

The development of remote sensing methods through research and large-scale application nowadays makes it possible to obtain stand-level estimates of forest variables at short intervals and at low cost. This offers substantial possibilities to forestry practitioners, but it also poses challenges regarding how cost-efficient data acquisition strategies should be developed. For example, should cheap but low-quality data be acquired and discarded whenever new data become available or should investments be made in high-quality data that are continuously updated to last over a longer period of time? We suggest that the solution could be to establish data assimilation (DA) procedures linked to forest inventories to make appropriate use of data from several sources. With DA, old information is updated through growth forecasts and when new information becomes available it is assimilated with the old information; the different sources of information are made use of to the extent motivated by their accuracy. In this study we made a general assessment of the usefulness of DA in connection with stand-level forest inventories and we compared two different methodological approaches, the extended Kalman filter and the Bayesian method. Not surprisingly, the relative advantage of DA was found to be largest for cases where low-precision estimates of growing stock volume were obtained at short intervals and forecasts were made with accurate growth prediction models. The methodological comparison revealed a tendency of the extended Kalman filter to underestimate the variance of the estimates.