Search

The main sources of uncertainty in CO2 and H2O fluxes up-scaling using empirical methods can be summarized in 1) degree of correlation between inputs and target variables, 2) uncertainty in inputs values and 3) method used in the extrapolation. In this work a fluxes up-scaling dedicated algorithm will be presented. It uses only remotely sensed input in order to estimate GPP fluxes in spatially explicit way. The study has been conducted at canopy level, ecosystem level and global scale. One of the most important factor that affect carbon dioxide fluxes is water stress. Primary, at canopy level correlation between multispectral spectrometer data and vegetation water status variables have been evaluated. Preliminary hypothesis has been carried out by simulations using a radiative transfer model (PROSAIL). A water stress experiment has been also carried out, inducing water status in plants and acquiring spectrometer data to check the degree of correlation between multispectral vegetation indices and water status variables. A library of satellite available vegetation indices has been evaluated. In addition, all available multispectral bands in the range 350-2500 nm at bandpass of 10 nm, have been used to calculate all available vegetation indices and correlation with vegetation water status variables in order to evaluate other band combination not acquired currently by satellite. At ecosystem scale, relationships between eddy-covariance tower latent heat (LE) and gross primary production (GPP) to MODIS satellite multispectral indices has been evaluated. Degrees of correlation have been analysed in order to evaluate the if different fitting can be put in relation to plant functional type (PFT). The effects of meteorological gradient and sites specifics characteristics on degree of correlation between fluxes and vegetation indices haves been also evaluated. In order to define the climatic conditions where relationships between fluxes and remotely sensed indices in water stress situations works better, period with high water stress level (no precipitation for at least 24 days) have been selected and the shape of the relation studied. These studies are useful in order to elaborate empirical model for GPP up-scaling. We choose a model tree algorithm in order to up-scale GPP fluxes. The reason was related to capability of model tree to generalize relationships for datasets characterized by high variability and interaction. Model tree is able to stratify dataset in homogeneous subset in order to maximize degree of fitting of multiple regression function to a target variable. Model was been evaluated by leave one out method. Only remotely sensed input are used reducing the level of uncertainty due to errors in the input dataset. The degree of accuracy was similar to other remotely sensed methods that, in addition use ancillary interpolated and modeled input (es. MOD 17). This results must be encourage the use of empirica methods based only on measured data (like radiations) that have the advantage of a reduction of others sources of uncertainty. Dottorato di ricerca in Ecologia forestale