Sofia: Satellite Observations for Fire Activity

In SOfireA: Satellite Observations for Fire Activity

Description

SOFIA (Satellite Observations for Fire Activity) is an empirical modelling concept to predict burned area based on satellite and climate data. Thereby several logistic functions are multiplicatively combined.

Usage

Sofia(x, area = rep(1, nrow(x)), per.group = rep(FALSE, ncol(x)), 
    sofiapar = NULL, par = NULL, return.all = TRUE, ...)

Arguments

x	data.frame with independent variables
area	a vector or data.frame/matrix with fractional coverage of grid cell area. If 'area' is a vector, it represents the maximal fractional burned area of a grid cell (e.g. the maximum vegetated area). If 'area' is a data.frame or matrix, it represents fractional coverage of groups (e.g. PFTs). Columns should represent groups and rows should be observations (grid cells and time steps).
per.group	a boolean vector that indicates if a column in x acts per group (e.g. PFTs)
sofiapar	object of class SofiaPar which is used for the fit. If NULL, the argument 'par' is used to create sofiapar using the function SofiaPar
par	vector of parameters of logistic functions. If NULL, default parameters are used (that are usually physically not plausible)
return.all	return all input and results? The function returns an object of class 'Sofia'. If TRUE, this object includes in the 'data' slot the fitted values, the fits per group, the response functions, the inputs 'x' and 'area'. If FALSE, only the fitted values are included.
...	further arguments

Details

No details.

Value

an object of class 'Sofia' which is actually a list.

Author(s)

Matthias Forkel <[e-mail]> [aut, cre]

References

No reference.

See Also

SofiaOpt, SofiaLogistic

Examples

# example based on aritifical data
#---------------------------------

# explanatory variables
sm <- 1:100
temp <- rnorm(100, 12, 10)
x <- cbind(sm, temp)

# fractional coverage of groups, e.g. plant functional types
tree <- runif(100, 0, 0.8)
grass <- 1 - tree
area <- cbind(tree, grass)

# calculate Sofia with some dummy parameters:
sf <- Sofia(x, area, per.group=c(TRUE, FALSE))
sf$eq
summary(sf$data)
plot(sf)

# with some more realisitc parameters:
par <- SofiaPar(colnames(x), per.group=c(TRUE, FALSE), group.names=c("tree", "grass"))
par
par$par <- c(1, 1, 20, 2, 1, -0.2, -0.1, 13, 10)
sf <- Sofia(x, area, per.group=c(TRUE, FALSE), sofiapar=par)
plot(sf)

sm <- 1:100
sm.2 <- sm
temp <- rnorm(100, 12, 10)
x <- cbind(sm, sm.2, temp)
par <- SofiaPar(colnames(x), per.group=c(TRUE, TRUE, FALSE), 
   group.names=c("tree", "grass"))
par
par$par <- c(2, 1, 20, 2, 2, 0.3, 0.2, 20, 40, 1, 1, -0.2, -0.1, 20, 10)  
sf <- Sofia(x, area, per.group=c(TRUE, TRUE, FALSE), sofiapar=par)
plot(sf)


# example based on real data
#---------------------------

require(ModelDataComp)

# get data
data(firedata)

# predictor variables
train <- firedata$train == 1 # use training data
xvars.df <- data.frame(
   NLDI = firedata$NLDI[train],
   CRU.WET.orig = firedata$CRU.WET.orig[train],
   Liu.VOD.annual = firedata$Liu.VOD.annual[train],
   GIMMS.FAPAR.pre = firedata$GIMMS.FAPAR.pre[train],
   CRU.DTR.orig = firedata$CRU.DTR.orig[train]
)

# observed data
obs <- firedata$GFED.BA.obs[train]
regid <- firedata$regid[train]

# Which x variable should depend on land cover?
per.group <- c(FALSE, TRUE, TRUE, TRUE, TRUE)

# land cover
area <- data.frame(
   Tree = firedata$CCI.LC.Tree[train],
   Shrub = firedata$CCI.LC.Shrub[train],
   HrbCrp = firedata$CCI.LC.HrbCrp[train]
)

# define parameters (from Forkel et al. 2016, Fig. 1)
sofiapar <- SofiaPar(colnames(xvars.df), colnames(area), per.group=per.group, 
   par.act=c(1.9, 780, 1, # for NLDI
   0.3, 1.1, -5.3, 8.9, 0.54, -23, -39, 13, -16, # for CRU.DTR
   0.13, 3, 0.53, 0.35, -0.44, 0.36, -1.2, -4.8, -45, # for CRU.WET
   -0.7, 18, -1.5, 22, 11, -17, -2.3, 0.64, 1,  # for GIMMS.FAPAR
   1.9, 3, -0.36, -21, 68, -38, 0.35, 0.31, 0.11) # for Liu.VOD
   )

# run model
sf <- Sofia(xvars.df, area, per.group=per.group, sofiapar=sofiapar)
plot(sf) 

SOfireA documentation built on May 2, 2019, 6:07 p.m.