One way to do this, perhaps:

1) Convert your gene annotations-of-interest to a sorted BED file via convert2bed from BEDOPS, e.g.:

$ convert2bed --input=gff < genes.gff > genes.bed

2) Make a BED file of the regulatory or promoter regions, using the BED-formatted gene file and bedops --range -N:0 for forward-strand oriented annotations and bedops --range 0:N for reverse-strand oriented annotations, e.g. for a 1kb region upstream of the TSS:

$ awk '$6=="+"' genes.bed > genes.for.bed
$ awk '$6=="-"' genes.bed > genes.rev.bed
$ bedops --range -1000:0 --everything genes.for.bed > promoters.for.bed
$ bedops --range 0:1000 --everything genes.rev.bed  > promoters.rev.bed
$ bedops --everything promoters.*.bed > promoters.bed

Adjust this window or the set operations, depending on what parts of the genome you decide you want to call a promoter, with respect to the gene or genes of interest.

3) Use samtools-indexed FASTA of your build of Arabidopsis with a script like bed2faidx.pl to convert the BED-formatted promoter regions to sequences:

$ bed2faidx.pl < promoters.bed > promoters.fa

4) Use web or command-line MEME to query the promoter sequences for putative TF motifs.

5) Use command-line TOMTOM to query the MEME TF motifs for matches against known/published TFs from databases like JASPAR Plantae.

If you want a more fine-grained answer, you might split the gene annotations from step 1 into per-gene files, and run steps 2-5 on each gene separately. You can then get a pool of published TFs per-gene and determine from there where there are similarities or unique hits.