diff --git a/app/tasks/cog_assets.py b/app/tasks/cog_assets.py
index 61a2e742b..6b8e6f694 100644
--- a/app/tasks/cog_assets.py
+++ b/app/tasks/cog_assets.py
@@ -9,6 +9,7 @@
 from app.models.pydantic.change_log import ChangeLog
 from app.models.pydantic.creation_options import COGCreationOptions
 from app.models.pydantic.jobs import GDALCOGJob, Job
+from app.settings.globals import DATA_LAKE_BUCKET
 from app.tasks import callback_constructor
 from app.tasks.batch import execute
 from app.tasks.raster_tile_set_assets.utils import JOB_ENV
@@ -90,6 +91,8 @@ async def create_cogify_job(
         dataset,
         "-I",
         creation_options.implementation,
+        "--prefix",
+        f"s3://{DATA_LAKE_BUCKET}/{dataset}/{version}/raster/{srid}/cog",
     ]
 
     if creation_options.export_to_gee:
diff --git a/batch/scripts/cogify.sh b/batch/scripts/cogify.sh
index d85344e6c..000a730eb 100755
--- a/batch/scripts/cogify.sh
+++ b/batch/scripts/cogify.sh
@@ -3,44 +3,50 @@
 set -e
 
 # requires arguments
-# -s | --source
-# -T | --target
 # --block_size
 # -r | --resample
 # -G | --export_to_gee
-# -d | --dataset
 # -I | --implementation
+# -t | --target
+# --prefix
 
 ME=$(basename "$0")
-. get_arguments.sh "$@"
+#. get_arguments.sh "$@"
 
 set -x
 # download all GeoTiff files
-aws s3 cp --recursive --exclude "*" --include "*.tif" "${SRC}" .
 
-# create VRT of input files so we can use gdal_translate
-if [ ! -f "merged.vrt" ]; then
+if [[ $(aws s3 ls "${PREFIX}/merged.tif") ]]; then
+  aws s3 cp "${PREFIX}/merged.tif" merged.tif
+else
+  aws s3 cp --recursive --exclude "*" --include "*.tif" "${SRC}" .
+
+  # create VRT of input files so we can use gdal_translate
   gdalbuildvrt merged.vrt *.tif
-fi
 
-# merge all rasters into one huge raster using COG block size
-if [ ! -f "merged.tif" ]; then
+  # merge all rasters into one huge raster using COG block size
   gdal_translate -of GTiff -co TILED=YES -co BLOCKXSIZE="${BLOCK_SIZE}" -co BLOCKYSIZE="${BLOCK_SIZE}" -co COMPRESS=DEFLATE -co BIGTIFF=IF_SAFER -co NUM_THREADS=ALL_CPUS --config GDAL_CACHEMAX 70% --config GDAL_NUM_THREADS ALL_CPUS merged.vrt merged.tif
+  aws s3 cp merged.tif "${PREFIX}/merged.tif"
 fi
 
-# create overviews in raster
-if ! gdalinfo "merged.tif" | grep -q "Overviews"; then
-  gdaladdo merged.tif -r "${RESAMPLE}" --config GDAL_NUM_THREADS ALL_CPUS --config GDAL_CACHEMAX 70%
+if [[ $(aws s3 ls "${PREFIX}/merged.tif.ovr") ]]; then
+  aws s3 cp "${PREFIX}/merged.tif.ovr" merged.tif.ovr
+else
+  # generate overviews externally
+  gdaladdo merged.tif -r "${RESAMPLE}" -ro --config GDAL_NUM_THREADS ALL_CPUS --config GDAL_CACHEMAX 70% --config COMPRESS_OVERVIEW DEFLATE
+  aws s3 cp merged.tif.ovr "${PREFIX}/merged.tif.ovr"
 fi
 
 # convert to COG using existing overviews, this adds some additional layout optimizations
-if [ ! -f "cog.tif" ]; then
-  gdal_translate merged.tif cog.tif -of COG -co COMPRESS=DEFLATE -co BLOCKSIZE="${BLOCK_SIZE}" -co BIGTIFF=IF_SAFER -co NUM_THREADS=ALL_CPUS --config GDAL_CACHEMAX 70% --config GDAL_NUM_THREADS ALL_CPUS
-fi
+gdal_translate merged.tif cog.tif -of COG -co COMPRESS=DEFLATE -co BLOCKSIZE="${BLOCK_SIZE}" -co BIGTIFF=IF_SAFER -co NUM_THREADS=ALL_CPUS -co OVERVIEWS=FORCE_USE_EXISTING --config GDAL_CACHEMAX 70% --config GDAL_NUM_THREADS ALL_CPUS
 
 # upload to data lake
 aws s3 cp cog.tif "${TARGET}"
 
+# delete intermediate file
+aws s3 rm "${PREFIX}/merged.tif"
+aws s3 rm "${PREFIX}/merged.tif.ovr"
+
 if [ -n "$EXPORT_TO_GEE" ]; then
   export_to_gee.py --dataset "${DATASET}" --implementation "${IMPLEMENTATION}"
 fi