Predicting tree species presence and basal area in Utah: A comparison of stochastic gradient boosting, generalized additive models, and tree-based methods

Many efforts are underway to produce broad-scale forest attribute maps by modelling forest class and structure variables collected in forest inventories as functions of satellite-based and biophysical information. Typically, variants of classification and regression trees implemented in Rulequest’s © See5 and Cubist (for binary and continuous responses, respectively) are the tools of choice in many of these applications. These tools are widely used in large remote sensing applications, but are not easily interpretable, do not have ties with survey estimation methods, and use proprietary unpublished algorithms. Consequently, three alternative modelling techniques were compared for mapping presence and basal area of 13 species located in the mountain ranges of Utah, USA. The modelling techniques compared included the widely used See5/Cubist, generalized additive models (GAMs), and stochastic gradient boosting (SGB). Model performance was evaluated using independent test data sets. Evaluation criteria for mapping species presence included specificity, sensitivity, Kappa, and area under the curve (AUC). Evaluation criteria for the continuous basal area variables included correlation and relative mean squared error. For predicting species presence (setting thresholds to maximize Kappa), SGB had higher values for the majority of the species for specificity and Kappa, while GAMs had higher values for the majority of the species for sensitivity. In evaluating resultant AUC values, GAM and/or SGB models had significantly better results than the See5 models where significant differences could be detected between models. For nine out of 13 species, basal area prediction results for all modelling techniques were poor (correlations less than 0.5 and relative mean squared errors greater than 0.8), but SGB provided the most stable predictions in these instances. SGB and Cubist performed equally well for modelling basal area for three species with moderate prediction success, while all three modelling tools produced comparably good predictions (correlation of 0.68

[1]  Trevor Hastie,et al.  Generalized linear and generalized additive models in studies of species distributions: setting the scene , 2002 .

[2]  Yoav Freund,et al.  Experiments with a New Boosting Algorithm , 1996, ICML.

[3]  R. Kauth,et al.  The tasselled cap - A graphic description of the spectral-temporal development of agricultural crops as seen by Landsat , 1976 .

[4]  S. Wood,et al.  GAMs with integrated model selection using penalized regression splines and applications to environmental modelling , 2002 .

[5]  Anthony Lehmann,et al.  GRASP: generalized regression analysis and spatial prediction , 2002 .

[6]  David B. Clark,et al.  EDAPHIC FACTORS AND THE LANDSCAPE-SCALE DISTRIBUTIONS OF TROPICAL RAIN FOREST TREES , 1999 .

[7]  Jacob Cohen.  A Coefficient of Agreement for Nominal Scales , 1960 .

[8]  T. Yee,et al.  Generalized additive models in plant ecology , 1991 .

[9]  Rick L. Lawrence,et al.  Classification of remotely sensed imagery using stochastic gradient boosting as a refinement of classification tree analysis , 2004 .

[10]  Keith D. Shepherd,et al.  Rapid characterization of Organic Resource Quality for Soil and Livestock Management in Tropical Agroecosystems Using Near Infrared Spectroscopy. , 2003 .

[11]  Leo Breiman,et al.  Bagging Predictors , 1996, Machine Learning.

[12]  Thomas C. Edwards,et al.  Use of generalized linear models and digital data in a Forest Inventory of Northern Utah , 1999 .

[13]  Thomas W. Yee,et al.  Vector generalized additive models in plant ecology , 2002 .

[14]  Niklaus E. Zimmermann,et al.  A new GLM-based method for mapping tree cover continuous fields using regional MODIS reflectance data , 2005 .

[15]  Eric Bauer,et al.  An Empirical Comparison of Voting Classification Algorithms: Bagging, Boosting, and Variants , 1999, Machine Learning.

[16]  G. Ridgeway.  The State of Boosting ∗ , 1999 .

[17]  William A. Bechtold,et al.  The enhanced forest inventory and analysis program - national sampling design and estimation procedures , 2005 .

[18]  T. Hastie,et al.  Comparative performance of generalized additive models and multivariate adaptive regression splines for statistical modelling of species distributions , 2006 .

[19]  L. Breiman.  Arcing Classifiers , 1998 .

[20]  Ronald E. McRoberts,et al.  Stratified estimation of forest area using satellite imagery, inventory data, and the k-Nearest Neighbors technique , 2002 .

[21]  J. Friedman.  Greedy function approximation: A gradient boosting machine. , 2001 .

[22]  Michael J. Oimoen,et al.  The National Elevation Dataset , 2002 .

[23]  J. Townshend,et al.  Global land cover classi(cid:142) cation at 1 km spatial resolution using a classi(cid:142) cation tree approach , 2004 .

[24]  G. Foody,et al.  Predictive relations of tropical forest biomass from Landsat TM data and their transferability between regions , 2003 .

[25]  J. Friedman.  Stochastic gradient boosting , 2002 .

[26]  P. McCullagh,et al.  Generalized Linear Models , 1972, Predictive Analytics.

[27]  J. Townshend,et al.  A stepwise regression tree for nonlinear approximation: Applications to estimating subpixel land cover , 2003 .

[28]  Limin Yang,et al.  COMPLETION OF THE 1990S NATIONAL LAND COVER DATA SET FOR THE CONTERMINOUS UNITED STATES FROM LANDSAT THEMATIC MAPPER DATA AND ANCILLARY DATA SOURCES , 2001 .

[29]  Jesús Muñoz,et al.  Comparison of statistical methods commonly used in predictive modelling , 2004 .

[30]  John Bell,et al.  A review of methods for the assessment of prediction errors in conservation presence/absence models , 1997, Environmental Conservation.

[31]  W. Cohen,et al.  Modelling forest cover attributes as continuous variables in a regional context with Thematic Mapper data , 2001 .

[32]  Eric P. Crist,et al.  A Physically-Based Transformation of Thematic Mapper Data---The TM Tasseled Cap , 1984, IEEE Transactions on Geoscience and Remote Sensing.

[33]  W. González-Manteiga,et al.  Support vector machines and gradient boosting for graphical estimation of a slate deposit , 2004 .

[34]  R. Tibshirani,et al.  Generalized Additive Models , 1991 .

[35]  E. Tomppo,et al.  Satellite image-based national forest inventory of finland for publication in the igarss'91 digest , 1991, [Proceedings] IGARSS'91 Remote Sensing: Global Monitoring for Earth Management.

[36]  Janet L. Ohmann,et al.  Predictive mapping of forest composition and structure with direct gradient analysis and nearest- neighbor imputation in coastal Oregon, U.S.A. , 2002 .

[37]  C. Brodley,et al.  Decision tree classification of land cover from remotely sensed data , 1997 .

[38]  W. Thuiller.  BIOMOD – optimizing predictions of species distributions and projecting potential future shifts under global change , 2003 .

[39]  A. Prasad,et al.  Newer Classification and Regression Tree Techniques: Bagging and Random Forests for Ecological Prediction , 2006, Ecosystems.

[40]  J. Townshend,et al.  An operational atmospheric correction algorithm for Landsat Thematic Mapper imagery over the land , 1997 .

[41]  Ross Ihaka,et al.  Gentleman R: R: A language for data analysis and graphics , 1996 .

[42]  S. Manel,et al.  Evaluating presence-absence models in ecology: the need to account for prevalence , 2001 .

[43]  Gretchen G. Moisen,et al.  Comparing five modelling techniques for predicting forest characteristics , 2002 .

[44]  M. Austin.  Spatial prediction of species distribution: an interface between ecological theory and statistical modelling , 2002 .

[45]  E. DeLong,et al.  Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. , 1988, Biometrics.

[46]  Limin Yang,et al.  Development of a 2001 National land-cover database for the United States , 2004 .

[47]  M. Bauer,et al.  Estimation and mapping of forest stand density, volume, and cover type using the k-nearest neighbors method , 2001 .

[48]  Thomas C. Edwards,et al.  Modeling spatially explicit forest structural attributes using Generalized Additive Models , 2001 .

[49]  G. De’ath,et al.  Development of a robust classifier of freshwater residence in barramundi (Lates calcarifer) life histories using elemental ratios in scales and boosted regression trees , 2005 .

[50]  Scott J. Goetz,et al.  Observed and predicted responses of plant growth to climate across Canada , 2005 .

[51]  J. Ross Quinlan,et al.  Induction of Decision Trees , 1986, Machine Learning.

[52]  Chengquan Huang,et al.  Enhanced algorithm performance for land cover classification from remotely sensed data using bagging and boosting , 2001, IEEE Trans. Geosci. Remote. Sens..

[53]  J. Ross Quinlan,et al.  C4.5: Programs for Machine Learning , 1992 .

[54]  L. Breiman.  Arcing classifier (with discussion and a rejoinder by the author) , 1998 .

[55]  Limin Yang,et al.  Derivation of a tasselled cap transformation based on Landsat 7 at-satellite reflectance , 2002 .

[56]  P. McCullagh,et al.  Generalized Linear Models , 1992 .

[57]  Alan H. Strahler,et al.  Maximizing land cover classification accuracies produced by decision trees at continental to global scales , 1999, IEEE Trans. Geosci. Remote. Sens..

[58]  A. Prasad,et al.  Potential Changes in Tree Species Richness and Forest Community Types following Climate Change , 2001, Ecosystems.

[59]  R. Dennis Cook,et al.  Generalized Linear Models , 2008 .

[60]  Leo Breiman,et al.  Classification and Regression Trees , 1984 .