Date/heure
30 mai 2024
09:15 - 10:15
Lieu
Salle de conférences Nancy
Oratrice ou orateur
Trinh Duong (LIF, LabEx ARBRE)
Catégorie d'évènement Groupe de travail Probabilités et Statistique
Résumé
Surveying natural populations is challenging due to their scattered distribution across a territory. To create spatially balanced samples, surveys typically divide the territory into a spatial grid and either use the grid nodes to form the sample or select points within the grid cells. Sampling the cells adds an additional stage, as currently employed by the French National Forest Inventory (NFI) for annual estimates. However, little attention has been given to accounting for this stage. Double sampling for stratification is a general method that helps reduce the size of a field sample, which is particularly costly. To improve sampling efficiency, we propose a new framework called two-stage two-phase sampling, incorporating a two-stage sampling design in the first phase.
The Horvitz-Thompson estimator is used to estimate the total value. In the first stage, cells are sampled using spatially systematic sampling, and in the second stage, points within these cells are sampled uniformly. The classification of first-phase points into strata is performed through photo-interpretation. In the second phase, points are sampled using spatially systematic sampling over the first-phase sample, based on varying sampling intensities across the strata. To calculate the variance estimator, the global first-phase sample is modeled as uniform sampling, and the global second-phase sample is modeled as stratified simple random sampling. Our results indicate that the expansion estimator remains unbiased and the variance estimators are moderately conservative for the sampling design used by the French NFI.
Additionally, the forest is undergoing rapid changes due to various disturbances, which can be large-scale, such as windthrow or fire, or small-scale, like bark beetle infestations. Our project focuses on large-scale disturbances. Estimating the area affected by such disturbances, known as the area of interest, is interesting for foresters. To address this, we are considering the intensification method, which increases sampling intensity in the area of interest. This method requires higher sampling intensity in specific zones compared to others, resulting in different sampling intensities across regions. A two-stage two-phase sampling framework is particularly useful for managing these varying sampling rates during the second phase, as disturbance information only becomes available at this phase.