Travaglini D* (1), Bottalico F (1), Chirici G (2), Puletti N (1), Nocentini S (1), Barbati A (3), Giannetti F (4), Corona P (3)
(1) DISTAF, Università di Firenze; (2) Dipartimento STAT, Università degli Studi del Molise; (3) DISAFRI, Università della Tuscia, Viterbo; (4) IPLA, Torino
Collocazione: c7.1.5 – Tipo Comunicazione: Presentazione orale
7° Congresso SISEF
Sessione 1: “Inventariazione e monitoraggio ambientale”
Contatto: Davide Travaglini (email@example.com)
Abstract: The aim of this study is to develop procedures for the estimation and mapping of quantitative forest attributes coupling forest inventory data, satellite imagery and other ancillary information, like digital elevation model. The values of quantitative attributes measured in the field by sample plots are extended over the whole forest area using k-Nearest Neighbors (k-NN) assisted procedures. Distinctively several procedures for the local optimization of k-NN are tested contrasting the use of raw spectral information of satellite bands and the use of their principal components. The accuracy of k-NN estimates are calculated for the different algorithm configurations in terms of number of k, type of multispectral distance and use of raw data or different numbers of first principal components. Then, large area assessment is performed by estimator unbiasedly aggregating pixel-level k-NN estimates. The proposed procedure is tested in the southern part of the Piedmont Region (Italy) over a forest area of about 4000 km2, with reference to the assessment of basal area of forest stands and with respect to different levels of administrative units. The obtained k-NN assisted estimates are compared with the Horvitz-Thompson estimates.
Citazione: Travaglini D , Bottalico F , Chirici G , Puletti N , Nocentini S , Barbati A , Giannetti F , Corona P (2009). Developing k-NN assisted procedures to assess quantitative attributes of forest stands . 7° Congresso Nazionale SISEF, Isernia – Pesche (IS), 29 Set – 03 Ott 2009, Contributo no. #c7.1.5