Methods for Retrieval of Forest Parameters from Satellite Remote Sensing

FOREMMS Methods for retrieval of forest parameters from satellite remote sensing 2 Norsk Regnesentral NOTAT SAMBA/26/01 Sammendrag/Abstract: One of the key questions in the FOREMMS project, is what kind of forest parameters satellite sensors can provide. In this note, we review previous approaches to estimation of forest parameters from existing satellite optical data, and discuss basic methodological technology and further research needs. The most fundamental parameter to FOREMMS is the forest cover, which can be derived from a classification of forest type or land cover. This review also includes methods for estimation of forest coverage fraction, and classification of forest type from time series. Biomass and wood volume can be estimated by regression models or by kNN interpolation models. Leaf Area Index (LAI) is a key biophysical variable. Many methods are based on empirical relationship between vegetation indices and LAI. However, the most promising methods seem to be the physically founded ones that have been developed for the production of the standard MODIS LAI product. Detection of forest changes and damages involves time-series of images and separation of effects like seasonal responses of the vegetation, inter-annual variability, and directional change. Deforestation and afforestation can be detected by comparing a classified forest image to the CORINE land cover database. Biodiversity can be mapped from three parameters named 'dominance', 'contagion' and 'fractal dimension'. We then discuss classification algorithms and algorithms for estimating a continuous parameter. For classification we suggest the Gaussian ML and Haslett's contextual MAP classifier. We also briefly discuss multi-resolution analysis, multi-temporal image classification models and data fusion models. For estimation of a continuous parameter we suggest general regression models that can use local models for spatial context, e.g. by using Markov random field models. This approach would use the same methodological framework as proposed for spatial context and ancillary data in image classification. FOREMMS Methods for retrieval of forest parameters from satellite remote sensing 3 Norsk Regnesentral NOTAT SAMBA/26/01 FOREMMS Methods for retrieval of forest parameters from satellite remote sensing 4 Norsk Regnesentral NOTAT SAMBA/26/01 Preface This report is written as a part of the FOREMMS project. Its purpose was initially to give a general documentation and description of the state-of-the-art for forest parameter retrieval from satellite remote sensing. During the process of studying literature and writing this text, the specification and goal of the FOREMMS system has been developing and changing into a system that are more focused, …