Statistical properties of positively-censored estimators of herbaceous primary productivity derived from successive independent surveys

The estimation of herbaceous primary productivity is usually derived from successive independent surveys. In order to avoid negative estimates, ecologists often use positively-censored estimators, i.e. they keep positive values but replace negative values by zero. The statistical properties of these positively-censored estimators are analysed in this paper : their relative bias, variance and mean squared error can be expressed as junctions of the sampling coefficient of variation (i.e. the ratio between the average sampling error and the true unknown productivity). When this coefficient increases, bias increases while variance and mean squared error decrease. A case study from a south Indian savanna illustrates that the bias is not negligible, especially for the months of low herbaceous productivity. In the case of normal distributions, analytical formula are derived, which can further be used to design sampling strategies that optimize their accuracy and precision. (Résumé d'auteur