Spatial Predictive Modelling with R

Cover
Half Title
Title Page
Copyright Page
Contents
Preface
Author Bio
1. Data acquisition, data quality control, and spatial reference systems
	1.1. Acquiring data for spatial predictive modeling
		1.1.1. Non-random sampling
		1.1.2. Unstratified random sampling
		1.1.3. Stratified random sampling design
		1.1.4. Stratified random sampling design with prior information
	1.2. Data quality control
		1.2.1. Accuracy of location information
		1.2.2. Sampling methods
		1.2.3. Sample duplications at the same location
		1.2.4. Sample quality
		1.2.5. Samples with missing values
		1.2.6. Data accuracy
	1.3. Spatial data types and spatial reference systems
		1.3.1. Spatial data types
		1.3.2. Spatial reference systems
2. Predictive variables and exploratory analysis
	2.1. Principles for pre-selection of predictive variables and limitations
		2.1.1. Principles
		2.1.2. Availability of causal variables
		2.1.3. Hidden predictive variables
		2.1.4. Limitations
	2.2. Predictive variables
		2.2.1. Predictive variables in terrestrial environmental sciences
		2.2.2. Predictive variables in marine environmental sciences
	2.3. Exploratory analysis
		2.3.1. Exploratory analysis for non-machine learning methods
		2.3.2. Exploratory analysis for machine learning methods
		2.3.3. Exploratory analysis for hybrid methods
3. Model evaluation and validation
	3.1. Predictive errors, observational errors, and true predictive errors
		3.1.1. Observed values and predicted values
		3.1.2. Relationships of predictive error with observational error and true predictive error
	3.2. Accuracy and error measures for predictive models
		3.2.1. Accuracy and error measures for numerical data
		3.2.2. Accuracy and error measures for categorical data
	3.3. R functions for accuracy and error measures
		3.3.1. Function pred.acc
		3.3.2. Function vecv
		3.3.3. Tovecv
	3.4. Model validation
		3.4.1. Validation methods
		3.4.2. Validation functions in R
		3.4.3. Effects of randomness associated of cross-validation methods on predictive accuracy assessments
		3.4.4. Procedure for the assessment of the performance of predictive models
4. Mathematical spatial interpolation methods
	4.1. Inverse distance weighted
		4.1.1. Implementation of IDW in gstat
		4.1.2. Parameter optimization for IDW
		4.1.3. Predictive accuracy of IDW with the optimal parameters
		4.1.4. Predictions of IDW
	4.2. Nearest neighbors
		4.2.1. Implementation of NN in gstat
		4.2.2. Predictive accuracy of NN
		4.2.3. Predictions of NN
	4.3. K nearest neighbors
		4.3.1. Parameter optimization for KNN
		4.3.2. Predictive accuracy of KNN with the optimal parameter
		4.3.3. Predictions of KNN
5. Univariate geostatistical methods
	5.1. Variogram modeling
		5.1.1. Concepts for variogram modeling
		5.1.2. Variogram modeling and variogram model selection
	5.2. Simple Kriging
		5.2.1. Implementation of SK in krige
		5.2.2. Parameter optimization for SK
		5.2.3. Predictive accuracy of SK with the optimal parameters
		5.2.4. SK predictions and variances
	5.3. Ordinary kriging
		5.3.1. Implementation of OK in gstat
		5.3.2. Implementation of OK in krige
		5.3.3. Parameter optimization for OK
		5.3.4. Predictive accuracy of OK with the optimal parameters
		5.3.5. OK predictions and variances
	5.4. Universal kriging
		5.4.1. Variogram modeling without anisotropy for UK
		5.4.2. Variogram modeling with anisotropy for UK
		5.4.3. Implementation of UK in krige with anisotropy
		5.4.4. Parameter optimization for UK
		5.4.5. Predictive accuracy of UK with the optimal parameters
		5.4.6. UK predictions and variances
	5.5. Block kriging
6. Multivariate geostatistical methods
	6.1. Simple cokriging
		6.1.1. Data normality and correlation
		6.1.2. Parameter optimization for SCK
		6.1.3. Predictive accuracy of SCK with the optimal parameters
		6.1.4. SCK predictions and variances
	6.2. Ordinary cokriging
		6.2.1. Data requirements
		6.2.2. Parameter optimization for OCK
		6.2.3. Predictive accuracy of OCK with the optimal parameters
		6.2.4. OCK predictions and variances
	6.3. Kriging with an external drift
		6.3.1. Application of KED
		6.3.2. Variable selection and parameter optimization for KED
		6.3.3. Predictive accuracy of KED
		6.3.4. KED predictions and variances
7. Modern statistical methods
	7.1. Linear models
		7.1.1. Relationships of response variable with predictive variables
		7.1.2. Implementation of LM in lm
		7.1.3. Model selection based on likelihood methods
		7.1.4. Variable selection based on predictive accuracy
		7.1.5. Predictive accuracy
		7.1.6. Predictions and standard errors
	7.2. Trend surface analysis
		7.2.1. Implementation of TSA in lm
		7.2.2. Variable selection
		7.2.3. Predictive accuracy
		7.2.4. Predictions and standard errors
	7.3. Thin plate splines
		7.3.1. Estimation of smoothing parameter lambda
		7.3.2. Implementation of TPS in Tps
		7.3.3. Varible selection and parameter optimization for TPS
		7.3.4. Predictive accuracy
		7.3.5. Predictions
	7.4. Generalized linear models
		7.4.1. Implementation of GLM in glm
		7.4.2. Implementation of GLM in glmnet
		7.4.3. Variable selection
		7.4.4. Parameter estimation for glmnet
		7.4.5. Predictive accuracy
		7.4.6. Spatial predictions and standard errors
	7.5. Generalized least squares
		7.5.1. Implementation of GLS in gls
		7.5.2. Variable selection for GLS
		7.5.3. Predictive accuracy
		7.5.4. Predictions
8. Tree-based machine learning methods
	8.1. Classification and regression trees
		8.1.1. Implementation of CART in the function rpart
		8.1.2. Implementation of CART in the function tree
	8.2. Random forest
		8.2.1. Application of RF
		8.2.2. Variable selection for RF
		8.2.3. Predictive accuracy of the RF models developed from variable selection methods
		8.2.4. Comparison of variable selection methods
		8.2.5. Predictions of RF
		8.2.6. Notes on RF
	8.3. Generalized boosted regression modeling
		8.3.1. Application of GBM
		8.3.2. Variable selection for GBM
		8.3.3. Parameter optimization for GBM models
		8.3.4. Predictive accuracy of GBM
		8.3.5. Partial dependence plots for GBM
		8.3.6. Predictions of GBM
9. Support vector machines
	9.1. Application of SVM
	9.2. Variable selection for SVM
	9.3. Parameter optimization for SVM models
	9.4. Predictive accuracy of SVM models
	9.5. Predictions of SVM
	9.6. Further modeling methods
10. Hybrids of modern statistical methods with geostatistical methods
	10.1. Hybrid method of LM and IDW
		10.1.1. Variable selection and parameter optimization based on predictive accuracy
		10.1.2. Predictive accuracy
		10.1.3. Predictions
	10.2. Hybrid method of LM and OK
		10.2.1. Variable selection and parameter optimization based on predictive accuracy
		10.2.2. Predictive accuracy
		10.2.3. Predictions
	10.3. Hybrid methods of LM, OK, and IDW
		10.3.1. Variable selection and parameter optimization based on predictive accuracy
		10.3.2. Predictive accuracy
		10.3.3. Predictions
	10.4. Hybrid method of GLM and IDW
		10.4.1. Variable selection and parameter optimization based on predictive accuracy
		10.4.2. Predictive accuracy
		10.4.3. Predictions
	10.5. Hybrid method of GLM and OK
		10.5.1. Variable selection and parameter optimization based on predictive accuracy
		10.5.2. Predictive accuracy
		10.5.3. Predictions
	10.6. Hybrid methods of GLM, OK, and IDW
		10.6.1. Variable selection and parameter optimization based on predictive accuracy
		10.6.2. Predictive accuracy
		10.6.3. Predictions
	10.7. Hybrid method of GLS and IDW
		10.7.1. Variable selection and parameter optimization based on predictive accuracy
		10.7.2. Predictive accuracy
		10.7.3. Predictions
	10.8. Hybrid method of GLS and OK
		10.8.1. Variable selection and parameter optimization based on predictive accuracy
		10.8.2. Predictive accuracy
		10.8.3. Predictions
	10.9. Hybrid methods of GLS, OK, and IDW
		10.9.1. Variable selection and parameter optimization based on predictive accuracy
		10.9.2. Predictive accuracy
		10.9.3. Predictions
11. Hybrids of machine learning methods with geostatistical methods
	11.1. Hybrid method of RF and IDW
		11.1.1. Variable selection and parameter optimization based on predictive accuracy
		11.1.2. Predictive accuracy
		11.1.3. Predictions
		11.1.4. A note on RFIDW
	11.2. Hybrid method of RF and OK
		11.2.1. Variable selection and parameter optimization based on predictive accuracy
		11.2.2. Predictive accuracy
		11.2.3. Predictions
		11.2.4. Notes on RFOK
	11.3. Hybrid methods of RF, OK, and IDW
		11.3.1. Variable selection and parameter optimization based on predictive accuracy
		11.3.2. Predictive accuracy
		11.3.3. Predictions
		11.3.4. A note on RFOKRFIDW
	11.4. Hybrid method of GBM and IDW
		11.4.1. Variable selection and parameter optimization based on predictive accuracy
		11.4.2. Predictive accuracy
		11.4.3. Predictions
	11.5. Hybrid method of GBM and OK
		11.5.1. Variable selection and parameter optimization based on predictive accuracy
		11.5.2. Predictive accuracy
		11.5.3. Predictions
	11.6. Hybrid methods of GBM, OK, and IDW
		11.6.1. Variable selection and parameter optimization based on predictive accuracy
		11.6.2. Predictive accuracy
		11.6.3. Predictions
	11.7. Hybrid method of SVM and IDW
		11.7.1. Variable selection and parameter optimization based on predictive accuracy
		11.7.2. Predictive accuracy
		11.7.3. Predictions
	11.8. Hybrid method of SVM and OK
		11.8.1. Variable selection and parameter optimization based on predictive accuracy
		11.8.2. Predictive accuracy
		11.8.3. Predictions
	11.9. Hybrid methods of SVM, OK, and IDW
		11.9.1. Variable selection and parameter optimization based on predictive accuracy
		11.9.2. Predictive accuracy
		11.9.3. Predictions
12. Applications and comparisons of spatial predictive methods
	12.1. Parameter optimization and/or variable selection
		12.1.1. IDW
		12.1.2. OK
		12.1.3. KED
		12.1.4. GLM
		12.1.5. GLMkrigingIDW
		12.1.6. RF
		12.1.7. RFkrigingIDW
		12.1.8. GBM
		12.1.9. GBMkrigingIDW
		12.1.10. SVM
		12.1.11. SVMkrigingIDW
	12.2. Predictive accuracy of spatial predictive methods
	12.3. Notes on prediction uncertainty
Appendix
	A. Data sets used in this book
		A.1. Location data for Petrel area
			A.1.1. Data source
			A.1.2. Location data
		A.2. Bathymetry data for Petrel area
			A.2.1. Data source
			A.2.2. Bathymetry data for Australian EEZ
			A.2.3. Bathymetry data for Petrel area
		A.3. Sample data for Petrel area
		A.4. Bee count data
			A.4.1. Data source
			A.4.2. Bee count data and relevant predictive variables
		A.5. Hard data
			A.5.1. Data source
			A.5.2. Hard data and relevant predictive variables
		A.6. Sponge data
			A.6.1. Data Source
			A.6.2. Sponge data and relevant predictive variables
References
Index