MAXENT preventing the EnsembleForecasting function of running

[beardant]

I'm trying to run an ensemble model whereby I'm not making any real projections in time or space,
I just have presences and pseudo-absences and want to model habitat suitability using an ensemble model.

The problem is related to MAXENT, as running the code without it works.

I first format the data;

myBiomodData1 <- BIOMOD_FormatingData(resp.name= myRespName1,
resp.var = myResp1,
expl.var = stack1,
resp.xy = myRespXY1,
filter.raster=TRUE,
PA.nb.rep = 1

I then use BIOMOD_modeling to create the 'submodels', three runs for each.

myBiomodModelOut1 <- BIOMOD_Modeling(
myBiomodData1,
models = c('GLM','GBM','FDA', 'RF', 'MARS','MAXENT','CTA','ANN','GAM','SRE'),
bm.options = myBiomodOption,
nb.rep=3,
data.split.perc=70,
var.import=3,
metric.eval = c('ROC', "TSS"),
save.output = TRUE,
scale.models = FALSE,
do.full.models=FALSE,
do.progress=TRUE)

Besides a warning (see below), all models are computed.
Warning message:
In newton(lsp = lsp, X = G$X, y = G$y, Eb = G$Eb, UrS = G$UrS, L = G$L, :
Iteration limit reached without full convergence - check carefully

The evaluations for the MAXENT submodels are adequate.

I then create an ensemble model using the following criteria;
myBiomodEM1 <- BIOMOD_EnsembleModeling(
bm.mod = myBiomodModelOut1,
em.by = 'all',
metric.eval = 'TSS',
metric.select='TSS',
metric.select.thresh = 0.5,
em.algo=c('EMca', 'EMwmean'))

To then attain a raster predicting habitat, I use the BIOMOD_EnsembleForecasting function.
Even though my desired output is not really a 'forecast', in so much that I have no new environment to project to, so the 'new.env' is the same RasterStack the 'expl.var' in the BIOMOD_formatingdata function. I suspect this may be causing an issue? I'm not sure if I would call the type of model output I want 'projections', or 'forecasts'.

myBiomodEF <- BIOMOD_EnsembleForecasting(
proj.name="ensemble",
bm.em = myBiomodEM1,
output.format='.img',
metric.binary='TSS',
new.env=stack1
)

The forecasting function runs, but at each projection of one of the three MAXENT runs, I get a pop-up error from outside of R saying :

Error in file C\Layers.\Species\models\1681839621\Species_PA1_RUN1_MAXENT_outputs\agriprop.asc:java.lang.NumberFormatException: For input string:"NODATA_value"

The final error in R , after all single models are projected, reads as follows:

Error: [rast] extents do not match
In addition: Warning messages: 1: In system(command = maxent.command, wait = TRUE, intern = TRUE) : running command 'java -mx512m -cp "C:/layers/maxent.jar" density.Project "./Species/models/1681839621/Species_PA1_RUN1_MAXENT_outputs/Species_PA1_RUN1.lambdas" "./Species/models/1681839621/Species_PA1_RUN1_MAXENT_outputs" "./Species/models/1681839621/Species_PA1_RUN1_MAXENT_outputs/projMaxent.asc" doclamp=false visible=false autorun nowarnings notooltips' had status 1 2: In system(command = maxent.command, wait = TRUE, intern = TRUE) : running command 'java -mx512m -cp "C:/layers/maxent.jar" density.Project "./Species/models/1681839621/Species_PA1_RUN2_MAXENT_outputs/Species_PA1_RUN2.lambdas" "./Species/models/1681839621/Species_PA1_RUN2_MAXENT_outputs" "./Species/models/1681839621/Species_PA1_RUN2_MAXENT_outputs/projMaxent.asc" doclamp=false visible=false autorun nowarnings notooltips' had status 1 3: In system(command = maxent.command, wait = TRUE, intern = TRUE) : running command 'java -mx512m -cp "C:/layers/maxent.jar" density.Project "./Species/models/1681839621/Species_PA1_RUN3_MAXENT_outputs/Species_PA1_RUN3.lambdas" "./Species/models/1681839621/Species_PA1_RUN3_MAXENT_outputs" "./Species/models/1681839621/Species_PA1_RUN3_MAXENT_outputs/projMaxent.asc" doclamp=false visible=false autorun nowarnings notooltips' had status 1

[rpatin]

Hello @beardant,
This is likely an issue with the environmental raster used for prediction, which probably have rectangular cells (more explanations here: #221 (comment)).

1. You can check that with show(stack1) which will show something like that:

class       : SpatRaster 
dimensions  : 47, 120, 5  (nrow, ncol, nlyr)
resolution  : 3, 3  (x, y)
[...]

If your resolution have different values for x and y then this is the problem. Only MAXENT is concerned as internally it uses .asc files and do not read them properly.

2. To solve that, you can:
   1. do not project MAXENT models by using models.chosen argument in BIOMOD_Projection
   2. use MAXNET instead of MAXENT, both algorithm are highly similar
   3. use only rasters with square resolution if you want to keep using MAXENT

In any case we are aware of the issue but this depends on MAXENT development and we will try to submit the issue there.
Best regards,
Rémi