Comparison of design-based and model-based estimates for tropical forestry resource with post-stratification

Forest resource management planning provides today the guide line for harvesting: the forest is cut in high yield units from which forest managers successively extract quantities of timber. Management mapping needs local assessment of the resource over the whole forest from a forest inventory. Prediction of the resource is made by two methods of kriging: one is ordinary kriging and the other one, named stratified kriging, takes into account an auxiliary qualitative variable. Results obtained from these technics are compared with those yielded by classical sampling. Thanks to an exhaustively sampled area, one can judge more objectively the suitability of each technique. If the total is similar for the different estimators, variances obtained by kriging are smaller. The difference between ordinary and stratified kriging concerns the distribution of the estimation. Taking into account the stratification was not very efficient and it might be better to use survey estimators. (C) 1999 Inra/Editions scientifiques et medicales Elsevier SAS.