Spatial Predictive Modelling with R

Length: 416 pages
Edition: 1
Language: English
Publisher: Chapman and Hall/CRC
Publication Date: 2022-01-31
ISBN-10: 0367550547
ISBN-13: 9780367550547
Sales Rank: #0 (See Top 100 Books)

Spatial predictive modeling (SPM) is an emerging discipline in applied sciences, playing a key role in the generation of spatial predictions in various disciplines. SPM refers to preparing relevant data, developing optimal predictive models based on point data, and then generating spatial predictions. This book aims to systematically introduce the entire process of SPM as a discipline. The process contains data acquisition, spatial predictive methods and variable selection, parameter optimization, accuracy assessment, and the generation and visualization of spatial predictions, where spatial predictive methods are from geostatistics, modern statistics, and machine learning.

The key features of this book are:

Systematically introducing major components of SPM process.
Novel hybrid methods (228 hybrids plus numerous variants) of modern statistical methods or machine learning methods with mathematical and/or univariate geostatistical methods.
Novel predictive accuracy-based variable selection techniques for spatial predictive methods.
Predictive accuracy-based parameter/model optimization.
Reproducible examples for SPM of various data types in R.

This book provides guidelines, recommendations, and reproducible examples for developing optimal predictive models by considering various components and associated factors for quality-improved spatial predictions. It provides valuable tools for researchers, modelers, and university students not only in SPM field but also in other predictive modeling fields.

Dr Li has produced over 100 various publications in spatial predictive modelling, statistical computing, ecological and environmental modelling, and ecology, developed a number of hybrid methods for SPM, and published four R packages for variable selections as well as SPM.

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