LAND USE CLASSIFICATION USING DEEP MULTITASK NETWORKS

Weather stations are often sparse and systematic under/overestimation of a global weather forecast system leads to bias. Most of the available bias correction methods do not consider higher order moments of a probability distribution and they use same distributions families to estimate both marginal and multivariate distributions. We propose three new copula-based bias correction methods, which describe the dependence structure between air temperature and covariates.

We present a Bayesian estimation of uncertainty in half-hourly GPP partitioned from flux tower measurements of NEE. The results show that it is possible to do this at any desirable time step. This, in turn, can be used to quantify the propagated uncertainty when validating process-based simulators. We further show the importance of using non-informative priors compared to informative priors of the parameters of flux partitioning model as they speed up calculation without loss of precision.