by the way, the R package "dismo", how to select fitting model? like setting the feature class is "qp", reg is "2.4".

thanks very much

hey, just use the predict function. if you use the raster::predict then 'object' is the environmental stack or brick for future/past if i remember correctly. in addition just put the fitted model. the latter sometimes hidden as a slot (@model) or a named list entry ($model) dep. on what package orfunction fits the model. this is the same proces as for getting the current prediction, ..just different data. make sure to also add the parameter type = "cloglog" if it is a maxnet model. Of course you can specify clamp parameter, too.

you could also use the base predict function but be aware that object then refers to the model while newdata requires the novel environm. data as a dataframe. the result would then be numbers to be filled into a raster. if it doesnt work, 99% of the time you are confusing which predict function you are or should be using or the newdata/object misses a variable or even more likely the spelling differs.

Rainer Ferdinand Wunderlich thanks a lot!! in my other question, after getting your advice to using R and read the vignette, I think that I understand the Maxent more deeply than before although still there are some questions I can't solve.

thanks a lot again!!

Alessandro Ferrarini yes, the Maxent software is so easy, the reason why I use the R package instead of Maxent software is I need some evaluation to evaluate the models produce, not only AUC but also like Kappa, TSS, AICc et cetera. So I have to learn the R to realize the same functions in Maxent software.

Rainer Ferdinand Wunderlich

sorry when I use the future climate, but the function can't run and the error showing

#import occurrence data

p_coords

9. stop("undefined columns selected")

8. `[.data.frame`(data, vars)

7. data[vars]

6. .local(object, ...)

5. fun(model, blockvals, ...)

4. fun(model, blockvals, ...)

3. .local(object, ...)

2. predict(future, model, type = "cloglog")

1. predict(future, model, type = "cloglog")

what is 'future?

if raster* then it should work UNLESS a) you miss a variable in Future or

b) you mistyped their labels or

c) train returns an object that includes but is not a model

you can try to be specific and run raster::predict().

Cannot promise to always check back here but you are welcome to post an update for further exploration.

lol as i said likely mistyped....

check those

names(predictors)

names(future)

Rainer Ferdinand Wunderlich

these

> names(predictors)

[1] "bio01" "bio03" "bio06" "bio07" "bio13" "bio14"

[7] "bio15" "elevation"

> names(future)

[1] "bio01_2021_2040" "bio02_2021_2040" "bio03_2021_2040" "bio04_2021_2040"

[5] "bio05_2021_2040" "bio06_2021_2040" "bio07_2021_2040" "bio08_2021_2040"

I just to try the function predict(). so the future factors are not the same as current.

may it is problem? and another question I want to ask is if I use the resolution of about 1 km (30 arc seconds by 30 arc seconds), but the future factors are not high resolution like current factors, for example, 2.5 minutes. is there any error in using predict() function?

thank!

still can't work. when I select the same factors in the current and future.

names(predictors)

[1] "bio01" "bio03" "bio06" "bio07"

> names(future)

[1] "bio01_2021_2040" "bio03_2021_2040" "bio06_2021_2040" "bio07_2021_2040"

this is the introduction of model.

model

Object of class SDMmodel

Method: Maxent

Species: species

Presence locations: 113

Absence locations: 9998

Model configurations:

--------------------

fc: qh

reg: 1.2

iter: 500

Variables:

---------

Continuous: bio01 bio03 bio06 bio07

Categorical: NA

maybe the problem is the tables are mistyped because I use the predictor instead of the future, this function can work. but I am not sure.

map5 map5

class : RasterLayer

dimensions : 1781, 2748, 4894188 (nrow, ncol, ncell)

resolution : 0.008333333, 0.008333333 (x, y)

extent : 73.48781, 96.38781, 34.33413, 49.1758 (xmin, xmax, ymin, ymax)

crs : +proj=longlat +datum=WGS84 +no_defs

source : memory

names : layer

values : 2.93849e-05, 1 (min, max)

and if the tables are mistyped, the question is the future Bioclimatic variables in Worldclim like [ Future climate, 2.5 minutes spatial resolution — WorldClim 1 documentation ] 2021-2040 GCMs ssp126, select the BCC-CSM2-MR and bc file.

the download file is one tif file that includes 19 Bioclimatic variables, so I have to use the Arcgis Windows---Image analysis---Identity function---Extract Band Function one by one from band1 to band 19. so if the problem is the tables are mistyped, I can't find a better way to avoid it.

bio01_2021_2040 is not the same character string as bio01...

all variables used in 'train' or generally calibrating are required in predict.

Rainer Ferdinand Wunderlich thanks, but there is still a question, the 'train' function should use the 'data, and the data is based on current presence species data and current factors.

so if all variables are used in 'train', the data is the current or the future? and the current species presence and directly is used in future variables 'train'?

code:

data

of course you calibrate the model with the current data (i doubt you have presence data for the future ^u^). you can then use predict() to get the spatial distribution for any period of time (past, present, future) or any region you have complete information of the very same variables used for training.

you do know that you can change the names of variables, right?

try this after making sure the order is identical and no var is missing.

names(future)

Rainer Ferdinand Wunderlich thanks, but I still get in trouble with using current occurrence data coupe with future factors to calibrate the model. In my mind, the current data is based on current occurrence data and factors, which simulates the plant distribution in current. and predicting the future environment is still under the current simulation data because the current factors are mostly suitable for current occurrence.

if using future factors and current to calibrate the model, are there some mistakes?

and I think predicting the future by using the response curve for every factor in current and then conjecture the future distribution under the future changed factors.

it is my confusion. maybe I think too complicated.

"if using future factors and current to calibrate the model, are there some mistakes?"

this is simply absurd. the main mistake would be to even think about it.

again: for future prediction you use the variables representing future conditions. their names or labels, however, must be identical to those of the variables representing current conditions because the predict function otherwise rightfully throws an error that the variable(s) does/do not exist in the newdata (i.e., the object parameter in the raster::predict function).

Rainer Ferdinand Wunderlich thanks! I think I make a mistake in my thinking because of my incomprehension of Maxent, I get a lot from your advice and your opinions,!!! sooooo thanks!!