Random forests for homogeneous and non-homogeneous Poisson processes with excess zeros

Predicting downed woody material carbon stocks in forests of the conterminous United States

Downed woody material (DWM) is a unique part of the forest carbon cycle serving as a pool between living biomass and subsequent atmospheric emission or transference to other forest pools. Thus, DWM is an individually defined pool in national greenhouse gas inventories. The diversity of DWM carbon drivers (e.g., decay, tree mortality, or wildfire) and associated high spatial variability make this a difficult-to-predict component of forest ecosystems. Using the now fully established nationwide inventory of DWM across the United States (US), we developed models, which substantially improved predictions of stand-level DWM carbon density relative to the current national-reporting model ('previous' model, here). The previous model was developed from published DWM carbon densities prior to the NFI DWM inventory. Those predictions were tested using NFI DWM carbon densities resulting in a poor fit to the data (coefficient of determination, or R² = 0.03). We present new random forest (RF) and stochastic gradient boosted (SGB) regression models to prediction DWM carbon density on all NFI plots and spatially on all forest land pixels. We evaluated various biotic and abiotic regression predictors, and the most important were standing dead trees, long-term annual precipitation, and long-term maximum summer temperature. A RF model scored best for expanding predictions to NFI plots (R² = 0.31), while an SGB model was identified for DWM carbon predictions based on purely spatial data (i.e., NFI-plot-independent, with R² = 0.23). The new RF model predicts conterminous US DWM carbon stocks to be 15% lower than the previous model and 2% higher than NFI data expanded according to inventory design-based inference. The new NFI data-driven models not only improve the predictions of DWM carbon density on all plots, they also provide flexibility in extending these predictions beyond the NFI to make spatially explicit and spatially continuous estimates of DWM carbon on all forest land in the US.

A zero altered Poisson random forest model for genomic-enabled prediction

In genomic selection choosing the statistical machine learning model is of paramount importance. In this paper, we present an application of a zero altered random forest model with two versions (ZAP_RF and ZAPC_RF) to deal with excess zeros in count response variables. The proposed model was compared with the conventional random forest (RF) model and with the conventional Generalized Poisson Ridge regression (GPR) using two real datasets, and we found that, in terms of prediction performance, the proposed zero inflated random forest model outperformed the conventional RF and GPR models.

Data-driven and interpretable machine-learning modeling to explore the fine-scale environmental determinants of malaria vectors biting rates in rural Burkina Faso

Background

Improving the knowledge and understanding of the environmental determinants of malaria vector abundance at fine spatiotemporal scales is essential to design locally tailored vector control intervention. This work is aimed at exploring the environmental tenets of human-biting activity in the main malaria vectors (Anopheles gambiae s.s., Anopheles coluzzii and Anopheles funestus) in the health district of Diébougou, rural Burkina Faso.

Methods

Anopheles human-biting activity was monitored in 27 villages during 15 months (in 2017–2018), and environmental variables (meteorological and landscape) were extracted from high-resolution satellite imagery. A two-step data-driven modeling study was then carried out. Correlation coefficients between the biting rates of each vector species and the environmental variables taken at various temporal lags and spatial distances from the biting events were first calculated. Then, multivariate machine-learning models were generated and interpreted to (i) pinpoint primary and secondary environmental drivers of variation in the biting rates of each species and (ii) identify complex associations between the environmental conditions and the biting rates.

Results

Meteorological and landscape variables were often significantly correlated with the vectors’ biting rates. Many nonlinear associations and thresholds were unveiled by the multivariate models, for both meteorological and landscape variables. From these results, several aspects of the bio-ecology of the main malaria vectors were identified or hypothesized for the Diébougou area, including breeding site typologies, development and survival rates in relation to weather, flight ranges from breeding sites and dispersal related to landscape openness.

Conclusions

Using high-resolution data in an interpretable machine-learning modeling framework proved to be an efficient way to enhance the knowledge of the complex links between the environment and the malaria vectors at a local scale. More broadly, the emerging field of interpretable machine learning has significant potential to help improve our understanding of the complex processes leading to malaria transmission, and to aid in developing operational tools to support the fight against the disease (e.g. vector control intervention plans, seasonal maps of predicted biting rates, early warning systems).

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s13071-021-04851-x.