Estimating cavity tree and snag abundance using negative binomial regression models and nearest neighbor imputation methods

Abstract

Cavity tree and snag abundance data are highly variable and contain many zero observations. We predict cavity tree and snag abundance from variables that are readily available from forest cover maps or remotely sensed data using negative binomial (NB), zero-inflated NB, and zero-altered NB (ZANB) regression models as well as nearest neighbor (NN) imputation methods. The models were developed and fit to data collected by the Forest Inventory and Analysis program of the US Forest Service in Washington, Oregon, and California. For predicting cavity tree and snag abundance per stand, all three NB regression models performed better in terms of mean square prediction error than the NN imputation methods. The most similar neighbor imputation, however, outperformed the NB regression models in predicting overall cavity tree and snag abundance. Résumé : Les données sur l'abondance des arbres creux et des chicots sont extrêmement variables et contiennent plusieurs observations nulles. Nous prédisons l'abondance des arbres creux et des chicots à partir de variables facilement disponibles dans les cartes du couvert forestier ou parmi les données obtenues par télédétection en utilisant des modèles de régression binomiale négative (BN), BN à excès de zéros (ZINB) et BN tronquée à zéro (ZANB), ainsi que des méthodes d'imputation par le plus proche voisin. Les modèles sont élaborés et ajustés aux données collectées par le programme d'analyse et d'inventaire forestiers du U.S. Forest Service dans les É tats de Washington, de l'Oregon et de la Californie. Les trois modèles de régression BN offraient une meilleure performance en terme d'erreur quadratique moyenne de prédiction que les méthodes d'imputation par le plus proche voisin pour prédire l'abondance des arbres creux et des chicots par peuplement. Cependant, l'imputation par le voisin le plus semblable donnait de meilleurs résultats que les modèles de régression BN pour prédire l'abondance globale des arbres creux et des chicots. [Traduit par la Rédaction]

Comparison of nearest neighbour approaches for small area estimation of tree species-specific forest inventory attributes in central Europe using airborne laser scanner data

“…Since CF resulted in unbiased estimates comparable to MSN the bad poll of RF may be ascribed to the biased variable selection of the base learners (Strobl et al 2007). Biases in RF predictions were also identified in other studies (e.g., Eskelson et al 2009;Breidenbach et al 2010).…”

“…In contrast to Hudak et al (2008a, b), Eskelson et al (2009) conclude that MSN is better suited for their aims compared to RF. In an application for classification purposes (Chen et al 2004), RF proved to be biased towards the most frequent classes.…”

“…While biases for RF were not reported in Hudak et al (2008a, b), it was found in Eskelson et al (2009) that RF produced biased results for count data. In contrast to Hudak et al (2008a, b), Eskelson et al (2009) conclude that MSN is better suited for their aims compared to RF.…”

A Comparison of Selected Parametric and Non-Parametric Imputation Methods for Estimating Forest Biomass and Basal Area

“…Therefore, GWR may be sufficient for the estimation of biomass per acre if that is the only variable of interest; while, the nearest neighbor imputations are preferred when multiple response variables of interest are present in the analysis. When predicting a single variable, Eskelson et al (2009b) reported that parametric methods resulted in better performance than non-parametric methods.…”

Predicting tree recruitment with negative binomial mixture models

“…There are also a few studies so far have addressed this issue in forestry (e.g., Keefe, 2004;Rathbun and Fei, 2006;Eskelson et al, 2009;Flores et al, 2009;Li et al, 2011). Fortin and DeBlois (2007) applied zero-inflated Poisson model (ZIP) and zero-inflated discrete Weibull (ZIdiW) to model tree recruitment of hardwood stands.…”