–> Launch PostCRPT <–
Overview
The Postfire Conifer Reforestation Planning Tool (PostCRPT) seeks to answer the questions of where conifers will regenerate on their own after wildfire and, conversely, where conifer reforestation efforts might be targeted. The app is designed to simplify the process of creating predictive maps of postfire conifer regeneration.
PostCRPT predicts the probability of post-fire conifer regeneration under varying postfire precipitation and seed production scenarios. The predictive model was fit using data from 1,234 plots, spanning 19 wildfires, each measured five years after wildfire. PostCRPT predicts the probability that regeneration will occur within 4.4-m radius (60 m2) plots across the burn footprint. Refer to Stewart et al. (2020) for details.
The app requires the user to input a zip file containing a fire perimeter and a raster of burn severity (RdNBR). The RAVG website has both RdNBR and fire perimeter data sets available for all fires with at least 1,000 acres of National Forest land from 2007 to the present. A sample dataset is also available in the about tab of the tool.
A screenshot from the Postfire Conifer Reforestation Planning Tool; prediction for the 2021 Caldor Fire.
Presentations
Accuracy
Confidence intervals for the PostCRPT statistical models are visualized using reliability diagrams (below). The reliability diagrams depict the performance of PostCRPT predictions when applied to new data, and were created using leave-one-fire-out cross-validation. Predictive performance is high in the all-Conifer and fir (Abies) models. The pine (Pinus) model has much room for improvement. See Stewart et al. 2020 for a full discussion.
Reliability diagrams show how well the PostCRPT models performed when used to predict regeneration for new fires. Central horizontal lines are medians. Boxes cover from the first to the third quartiles, whiskers extend to the most extreme value no further than 1.5 times the interquartile range from the first and third quartiles. Additional data are plotted as outliers.
Statistical Models
Seeking to follow the principle of Ockham’s razor, PostCRPT uses just six input variables to predict the probability of regeneration. The variables are seed availability, burn severity, historical precipitation, postfire precipitation anomaly, slope, and aspect. Other candidate variables either did not improve out-of-sample model performance or resulted in response functions that appeared to be the result of overfitting.
Marginal effects of explanatory variables for the Abies regeneration model used by PostCRPT. Panels depict the effect of one variable when all other explanatory variables are set to their mean value, with 95% confidence intervals.
Citation
PostCRPT is a user interface for models developed by:
PostCRPT was adapted for the web from the poscrptR package, developed in a collaboration between the US Geological Survey Western Ecological Research Center and the US Forest Service:
The models developed by Stewart et al. (2020) build on an earlier spatial model developed by Shive et al. (2018) and analyses of sensitivity to postfire climate presented by Young et al. (2019).
Version History
PostCRPT version 0.123 – Released January, 2022. Allows users to incorporate the effects of multiple sequential fires. Uses the 2020 version of LEMMA GNN forest structure maps. Updates the Stewart et al 2020 regeneration model to use 2020 version of GNN structure maps. Adds useful warnings. Expands geographic coverage. Corrects the spatial domain.
PostCRPT version 0.010 – Released January, 2021. This initial release of the app cannot account for multiple sequential burns occurring after 2012. It uses the original Stewart et al 2020 regeneration models and the 2012 version of GNN basal area maps. It was adapted for the web from poscrptR and incorporates only small modifications of the poscrptR code.