Search

As part of the development of the 2011 National Land Cover Database (NLCD) tree canopy cover layer, a pilot project was launched to test the use of high-resolution photography coupled with extensive ancillary data to map the distribution of tree canopy cover over four study regions in the conterminous US. Two stochastic modeling techniques, random forests (RF) and stochastic gradient boosting (SGB), are compared. The objectives of this study were first to explore the sensitivity of RF and SGB to choices in tuning parameters and, second, to compare the performance of the two final models by assessing the importance of, and interaction between, predictor variables, the global accuracy metrics derived from an independent test set, as well as the visual quality of the resultant maps of tree canopy cover. The predictive accuracy of RF and SGB was remarkably similar on all four of our pilot regions. In all four study regions, the independent test set mean squared error (MSE) was identical to three decimal places, with the largest difference in Kansas where RF gave an MSE of 0.0113 and SGB gave an MSE of 0.0117. With correlated predictor variables, SGB had a tendency to concentrate variable importance in fewer variables, whereas RF tended to spread importance among more variables. RF is simpler to implement than SGB, as RF has fewer parameters needing tuning and also was less sensitive to these parameters. As stochastic techniques, both RF and SGB introduce a new component of uncertainty: repeated model runs will potentially result in different final predictions. We demonstrate how RF allows the production of a spatially explicit map of this stochastic uncertainty of the final model. Key words: tree canopy cover, predictive mapping, classification and regression trees, random forest, stochastic gradient boosting. Dans le cadre de l'elaboration de la couche cartographique du couvert forestier de la National Land Cover Database 2011, un projet pilote a ete lance pour tester l'utilisation de la photographie haute resolution couplee a de multiples donnees accessoires pour cartographier la distribution du couvert forestier dans les etats contigus des Etats-Unis. Deux techniques de modelisation stochastique, les forets aleatoires (FA) et le ' gradient boosting ' aleatoire (GBA) sont comparees. Les objectifs de cette etude consistaient : premierement a explorer la sensibilite des deux techniques face aux choix pour le reglage des parametres; et deuxiemement a comparer la performance des deux modeles finaux en evaluant l'importance des variables predictives et leurs interactions, les mesures d'exactitude globale derivee d'un dispositif de test independant, de meme que la qualite visuelle des cartes du couvert forestier qui sont produites. L'exactitude des previsions des deux techniques etait remarquablement similaire dans les quatre regions pilotes. Dans les quatre regions, l'EQM du dispositif de test independant etait identique a trois decimales pres; le plus grand ecart etait au Kansas ou la technique FA produisait un EQM de 0,0113 tandis que la technique GBA produisait un EQM de 0,0117. Avec des variables predictives correlees, la technique GBA avait tendance a concentrer l'importance des variables sur moins de variables alors que la technique FA avait tendance a repartir l'importance parmi davantage de variables. La technique FA est plus simple a appliquer que la technique GBA etant donne qu'elle compte a la fois moins de parametres qui ont besoin de reglage et qu'elle est aussi moins sensible a ces parametres. En tant que techniques stochastiques, tant la technique FA que la technique GBA introduisent une nouvelle composante d'incertitude : des simulations repetees vont potentiellement produire differentes predictions finales. Nous illustrons comment la technique FA permet de produire une carte spatialement explicite de cette incertitude stochastique du modele final. [Traduit par la Redaction] Mots-cles : couvert forestier, cartographie predictive, classification et arbres de regression, forets aleatoires, ' gradient boosting ' aleatoire., 1. Introduction The tree canopy cover in a given area is a primary structural characteristic of both forested and nonforested landscapes. Understanding and quantifying the spatial distribution of tree canopy [...]

Abstract Background Although significant amounts of carbon may be stored in harvested wood products, the extraction of that carbon from the forest generally entails combustion of fossil fuels. The transport of timber from the forest to primary milling facilities may in particular create emissions that reduce the net sequestration value of product carbon storage. However, attempts to quantify the effects of transport on the net effects of forest management typically use relatively sparse survey data to determine transportation emission factors. We developed an approach for systematically determining transport emissions using: 1) -remotely sensed maps to estimate the spatial distribution of harvests, and 2) - industry data to determine landscape-level harvest volumes as well as the location and processing totals of individual mills. These data support spatial network analysis that can produce estimates of fossil carbon released in timber transport. Results Transport-related emissions, evaluated as a fraction of transported wood carbon at 4 points in time on a landscape in western Montana (USA), rose from 0.5% in 1988 to 1.7% in 2004 as local mills closed and spatial patterns of harvest shifted due to decreased logging on federal lands. Conclusion The apparent sensitivity of transport emissions to harvest and infrastructure patterns suggests that timber haul is a dynamic component of forest carbon management that bears further study both across regions and over time. The monitoring approach used here, which draws only from widely available monitoring data, could readily be adapted to provide current and historical estimates of transport emissions in a consistent way across large areas.

ABSTRACT Gridded temperature data sets are typically produced at spatial resolutions that cannot fully resolve fine-scale variation in surface air temperature in regions of complex topography. These data limitations have become increasingly important as scientists and managers attempt to understand and plan for potential climate change impacts. Here, we describe the development of a high-resolution (250 m) daily historical (1979-2012) temperature data set for the US Northern Rocky Mountains using observations from both long-term weather stations and a dense network of low-cost temperature sensors. Empirically based models for daily minimum and maximum temperature incorporate lapse rates from regional reanalysis data, modelled daily solar insolation and soil moisture, along with time invariant canopy cover and topographic factors. Daily model predictions demonstrate excellent agreement with independent observations, with mean absolute errors of <1.4 °C for both minimum and maximum temperature. Topographically resolved temperature data may prove useful in a range of applications related to hydrology, fire regimes and fire behaviour, and habitat suitability modelling. The form of the models may provide a means for downscaling future temperature scenarios that account for potential fine-scale topographically mediated changes in near-surface temperature. [ABSTRACT FROM AUTHOR]

As part of the development of the 2011 National Land Cover Database (NLCD) tree canopy cover layer, a pilot project was launched to test the use of high-resolution photography coupled with extensive ancillary data to map the distribution of tree canopy cover over four study regions in the conterminous US. Two stochastic modeling techniques, random forests (RF) and stochastic gradient boosting (SGB), are compared. The objectives of this study were first to explore the sensitivity of RF and SGB to choices in tuning parameters and, second, to compare the performance of the two final models by assessing the importance of, and interaction between, predictor variables, the global accuracy metrics derived from an independent test set, as well as the visual quality of the resultant maps of tree canopy cover. The predictive accuracy of RF and SGB was remarkably similar on all four of our pilot regions. In all four study regions, the independent test set mean squared error (MSE) was identical to three decimal places, with the largest difference in Kansas where RF gave an MSE of 0.0113 and SGB gave an MSE of 0.0117. With correlated predictor variables, SGB had a tendency to concentrate variable importance in fewer variables, whereas RF tended to spread importance among more variables. RF is simpler to implement than SGB, as RF has fewer parameters needing tuning and also was less sensitive to these parameters. As stochastic techniques, both RF and SGB introduce a new component of uncertainty: repeated model runs will potentially result in different final predictions. We demonstrate how RF allows the production of a spatially explicit map of this stochastic uncertainty of the final model. [ABSTRACT FROM AUTHOR]

Copyright of Canadian Journal of Forest Research is the property of Canadian Science Publishing and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)