| Title: | Spatial Downscaling using the Dissever Algorithm |
|---|---|
| Description: | Spatial downscaling of coarse grid mapping to fine grid mapping using predictive covariates and a model fitted using the 'caret' package. The original dissever algorithm was published by Malone et al. (2012) <doi:10.1016/j.cageo.2011.08.021>, and extended by Roudier et al. (2017) <doi:10.1016/j.compag.2017.08.021>. |
| Authors: | Pierre Roudier [cre, aut], Brendan Malone [ctb] |
| Maintainer: | Pierre Roudier <[email protected]> |
| License: | GPL-2 |
| Version: | 0.2-3.9000 |
| Built: | 2024-11-04 04:04:40 UTC |
| Source: | https://github.com/pierreroudier/dissever |
Dissever: algorithm for spatial downscaling
Algorithm described in: [DOI: 10.1016/j.cageo.2011.08.021]
Brendan Malone ([email protected]);
Performs spatial downscaling of coarse grid mapping to fine grid mapping using predictive covariates and a model fitted using the caret package.
## S4 method for signature 'RasterLayer,RasterLayer' dissever(coarse, fine, method = "rf", p = 0.5, nmax = NULL, thresh = 0.01, min_iter = 5, max_iter = 20, boot = NULL, level = 0.9, tune_length = 3, tune_grid = .create_tune_grid(model = method, tune_length = tune_length), train_control_init = .default_control_init, train_control_iter = .default_control_iter, verbose = FALSE)## S4 method for signature 'RasterLayer,RasterLayer' dissever(coarse, fine, method = "rf", p = 0.5, nmax = NULL, thresh = 0.01, min_iter = 5, max_iter = 20, boot = NULL, level = 0.9, tune_length = 3, tune_grid = .create_tune_grid(model = method, tune_length = tune_length), train_control_init = .default_control_init, train_control_iter = .default_control_iter, verbose = FALSE)
coarse |
object of class |
fine |
object of class |
method |
a string specifying which classification or regression model to use (via the caret package). Possible values are found using names(caret::getModelInfo()). |
p |
numeric, proportion of the fine map that is sampled for fitting the dissever model (between 0 and 1, defaults to 0.5) |
nmax |
numeric maximum number of pixels selected for fitting the dissever model. It will override the number of pixels chosen by the |
thresh |
numeric, dissever iterations will proceed until the RMSE of the dissever model reaches this value, or until the maximum number of iterations is met (defaults to 0.01) |
min_iter |
numeric, minimum number of iterations (defaults to 5) |
max_iter |
numeric, maximum number of iterations (defaults to 20) |
boot |
numeric, if not NULL (default), the number of bootstrap replicates used to derive the confidence intervals. |
level |
If this is a numeric value, it is used to derive confidence intervals using quantiles. If it is a function, it is used to derive the uncertainty using this function. |
tune_length |
numeric, the number of parameters to test to find the optimal parametrisation of the caret model (defaults to 3) |
tune_grid |
a data frame with possible tuning values |
train_control_init |
Control parameters for finding the optimal parameters of the caret model (see trainControl) |
train_control_iter |
Control parameters for fitting the caret model during the iteration phase (see trainControl) |
verbose |
controls the verbosity of the output (TRUE or FALSE) |
Brendan Malone, Pierre Roudier
Malone, B.P, McBratney, A.B., Minasny, B., Wheeler, I., (2011) A general method for downscaling earth resource information. Computers & Geosciences, 41: 119-125. http://dx.doi.org/10.1016/j.cageo.2011.08.021
# Load the Edgeroi dataset (see ?edgeroi) data(edgeroi) # Plot the Edgeroi dataset (using the raster package) library(raster) plot(edgeroi$carbon) # coarse resolution layer plot(edgeroi$predictors) # fine resolution predictors # Run dissever using a simple linear model. # In this instance we are subsampling heavily (p = 0.05) to keep # run time short res_lm <- dissever( coarse = edgeroi$carbon, fine = edgeroi$predictors, method = "lm", min_iter = 5, max_iter = 10, p = 0.05 ) # A lot more models are available through caret: ## Not run: subset(caret::modelLookup(), forReg == TRUE, select = 'model') ## End(Not run) # Plot dissever results plot(res_lm, type = 'map', main = "Dissever using GAM") plot(res_lm, type = 'perf', main = "Dissever using GAM")# Load the Edgeroi dataset (see ?edgeroi) data(edgeroi) # Plot the Edgeroi dataset (using the raster package) library(raster) plot(edgeroi$carbon) # coarse resolution layer plot(edgeroi$predictors) # fine resolution predictors # Run dissever using a simple linear model. # In this instance we are subsampling heavily (p = 0.05) to keep # run time short res_lm <- dissever( coarse = edgeroi$carbon, fine = edgeroi$predictors, method = "lm", min_iter = 5, max_iter = 10, p = 0.05 ) # A lot more models are available through caret: ## Not run: subset(caret::modelLookup(), forReg == TRUE, select = 'model') ## End(Not run) # Plot dissever results plot(res_lm, type = 'map', main = "Dissever using GAM") plot(res_lm, type = 'perf', main = "Dissever using GAM")
Subset of the Australian 1-km resolution soil carbon map, along woth selected environmental covariates, for a small area of the Edgeroi district in NSW, Australia.
data(edgeroi)data(edgeroi)
The edgeroi dataset is a list with two elements:
carbona 7 row, 10 column RasterLayer of soil carbon stocks for a small area in the Edgeroi district NSW, Australia. The grid has a pixel resolution of 1 x 1 km. It contains the predicted soil carbon stock for the 0-30cm depth
predictorsan 87 row, 117 column, 5 layer RasterStack of selected environmental covariates from a small area in the Edgeroi district NSW, Australia. The grids have a pixel resolution of 90m x 90m. It contains the following layers:
Doseratenumeric; gamma radiometric data
Elevationnumeric; topographic variable of bare earth ground elevation. Derived from digital elevation model
Panchromatnumeric; panchromatic band of the Landsat 7 satelite
Slopenumeric; Slope gradient of the land surface. Derived from digital elevation model
TWInumeric; topographic wetness index. Secondary derivative of the digital elevation model
The Edgeroi District, NSW is an intensive cropping area upon the fertile alluvial Namoi River plain. The District has been the subject of many soil invetigations, namely McGarry et al. (1989) whom describe an extensive soil data set collected from the area. More recently, digital soil mapping studies of the area have been conducted, for example, Malone et al. (2009). The 1km mapping of soil carbon stock was performed using digital soil mapping methods McBratney et al. (2003) using the soil data from McGarry et al. (1989) and available environmental covariates.
The raw spatial data that contributed to the creation of edgeGrids were sourced from publically accessable repositories hosted by various Australian Government and international agencies including CSIRO (for the DEM), Geosciences Australia (for the radiometric data) and NASA (for the Landsat 7 ETM+ data). The projection for the RasterStack is WGS 84 Zone 55.
McGarry, D., Ward, W.T., McBratney, A.B. (1989) Soil Studies in the Lower Namoi Valley: Methods and Data. The Edgeroi Data Set. (2 vols) (CSIRO Division of Soils: Adelaide).
McBratney, A.B., Mendonca Santos, M.L., Minasny, B. (2003) On Digtial Soil mapping. Geoderma 117: 3-52.
Malone, B.P., McBratney, A.B., Minasny, B. (2009) Mapping continuous depth functions of soil carbon storage and available water capacity. Geoderma 154, 138-152.
@examples data(edgeroi) summary(edgeroi$carbon) summary(edgeroi$predictors)
Generates confidence intervals of a dissever output using bootstraping
## S4 method for signature 'list,RasterStack' generate_ci(object, covariates, level = 0.9, n = 50L)## S4 method for signature 'list,RasterStack' generate_ci(object, covariates, level = 0.9, n = 50L)
object |
object of class |
covariates |
object of class |
level |
If this is a numeric value, it is used to derive confidence intervals using quantiles. If it is a function, it is used to derive the uncertainty using this function. |
n |
the number of bootstrap replicates used to derive the confidence intervals |
Pierre Roudier
# Load the Edgeroi dataset (see ?edgeroi) data(edgeroi) # Create a dissever output diss <- dissever( coarse = edgeroi$carbon, fine = edgeroi$predictors, method = "lm", min_iter = 5, max_iter = 10, p = 0.05 ) # Generate the confidence intervals ## Not run: ci <- generate_ci(diss, edgeroi$predictors, n = 5) plot(ci) ## End(Not run)# Load the Edgeroi dataset (see ?edgeroi) data(edgeroi) # Create a dissever output diss <- dissever( coarse = edgeroi$carbon, fine = edgeroi$predictors, method = "lm", min_iter = 5, max_iter = 10, p = 0.05 ) # Generate the confidence intervals ## Not run: ci <- generate_ci(diss, edgeroi$predictors, n = 5) plot(ci) ## End(Not run)
Plots a dissever result. Two modes are possible to visualise either the resulting map or the convergence of the disseveration.
## S3 method for class 'dissever' plot(x, type = "map", ...)## S3 method for class 'dissever' plot(x, type = "map", ...)
x |
object of class |
type |
character, type of visualisation to produce. Options are "map", to produce a map of the dissevered coarse map, or "perf", to show the convergence of the RMSE during the disseveration procedure. |
... |
Additional arguments passed to plot |
Pierre Roudier
# See ?dissever# See ?dissever
Prints the performance of the model used to do the dissever procedure.
## S3 method for class 'dissever' print(x, ...)## S3 method for class 'dissever' print(x, ...)
x |
object of class |
... |
Additional arguments passed to print |
Pierre Roudier
Prints summary of the model used in the dissever procedure.
## S3 method for class 'dissever' summary(object, ...)## S3 method for class 'dissever' summary(object, ...)
object |
object of class |
... |
Additional arguments passed to summary |
Pierre Roudier