[Python] Google Earth Engine: Exported NDVI Rasters are Empty Despite Correct Visualization

I am using Google Earth Engine (GEE) to export monthly median NDVI rasters from the MODIS MOD13Q1 dataset, clipped to a specific fire polygon.

Issue

• NDVI visualizes correctly in the GEE Console (with expected values).
• Exported GeoTIFFs are completely empty (NaN values) despite no errors in the logs.
• The script successfully initiates exports, but the files contain no valid NDVI data when opened in QGIS or Python (rasterio).

What I've Tried

1️⃣ Visualizing NDVI in GEE Console (Works)


I confirmed that MODIS NDVI data exists for my region by displaying the median NDVI for a month:

var fireRegion = ee.Geometry.Polygon([
[[-121.974, 37.634], [-121.974, 36.972], [-120.700, 36.972], [-120.700, 37.634]]
]);

var modis = ee.ImageCollection("MODIS/061/MOD13Q1")
.select("NDVI")
.filterBounds(fireRegion)
.filterDate("2020-08-01", "2020-08-31")
.median()
.multiply(0.0001) // Scale NDVI correctly
.clip(fireRegion);

var ndviVis = {min: -0.2, max: 1.0, palette: ["red", "yellow", "green"]};

Map.centerObject(fireRegion, 8);
Map.addLayer(modis, ndviVis, "MODIS NDVI (August 2020)");
print("NDVI Image:", modis);


✅ NDVI renders correctly on the GEE map. 2️⃣ Batch Export in Python (Fails)

I wrote a script in Python to export NDVI for each month from 2019-2022.

import ee
import geopandas as gpd
import logging
from datetime import datetime, timedelta

# Set up logging
logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s")

# Initialize Earth Engine
try:
ee.Initialize(project="sound-of-resiliency")
except Exception as e:
ee.Authenticate()
ee.Initialize(project="sound-of-resiliency")

# Load the shapefile
shapefile = gpd.read_file("fire_poly.shp")

# Ensure the shapefile is in WGS84 (EPSG:4326)
if shapefile.crs != "EPSG:4326":
shapefile = shapefile.to_crs("EPSG:4326")

# Convert shapefile to an EE geometry
fire_region_geom = shapefile.geometry.union_all()
fire_region = ee.Geometry(fire_region_geom.__geo_interface__)

# Load MODIS NDVI dataset
modis = ee.ImageCollection("MODIS/061/MOD13Q1").select("NDVI").filterBounds(fire_region)

def get_monthly_ndvi(year, month):
"""Compute median NDVI for all images within a month."""
start = datetime(year, month, 1)
end = (start + timedelta(days=32)).replace(day=1) # First day of next month

monthly_collection = modis.filterDate(start.strftime("%Y-%m-%d"), end.strftime("%Y-%m-%d"))

# Debugging: Print number of images found
image_count = monthly_collection.size().getInfo()
logging.info(f"MODIS images found for {year}-{month}: {image_count}")

if image_count == 0:
logging.warning(f"No MODIS NDVI images found for {year}-{month}. Skipping.")
return None

monthly_ndvi = (
monthly_collection.median()
.multiply(0.0001) # Convert to NDVI scale (-1 to 1)
.clip(fire_region)
)

return monthly_ndvi

def export_ndvi_raster(image, year, month):
"""Export NDVI raster to Google Drive if valid."""
if image is None:
logging.warning(f"Skipping export for {year}-{str(month).zfill(2)} (No NDVI data)")
return

image = image.reproject(crs="EPSG:4326", scale=250)

stats = image.reduceRegion(
reducer=ee.Reducer.mean(),
geometry=fire_region,
scale=250,
maxPixels=1e13
)

ndvi_value = stats.get("NDVI")

if ndvi_value is not None:
task = ee.batch.Export.image.toDrive(
image=image,
description=f"NDVI_{year}_{str(month).zfill(2)}",
folder="GEE_Exports",
fileNamePrefix=f"NDVI_{year}_{str(month).zfill(2)}",
scale=250,
region=fire_region,
fileFormat="GeoTIFF",
maxPixels=1e13
)
task.start()
logging.info(f"Export started for NDVI {year}-{str(month).zfill(2)}")
else:
logging.warning(f"Skipping export for {year}-{str(month).zfill(2)} (No valid NDVI pixels)")

for year in range(2019, 2023):
for month in range(1, 13):
logging.info(f"Processing NDVI for {year}-{str(month).zfill(2)}")
monthly_ndvi = get_monthly_ndvi(year, month)
export_ndvi_raster(monthly_ndvi, year, month)

logging.info("All NDVI exports have been started. Check GEE tasks panel.")


3️⃣ Checking the Exported .tif Files

After downloading the .tif files from Google Drive, I analyzed them using Python (rasterio):

import rasterio
import numpy as np

with rasterio.open("NDVI_2022_12.tif") as src:
data = src.read(1)
print(f"Min: {np.nanmin(data)}, Max: {np.nanmax(data)}, Mean: {np.nanmean(data)}")

Output:

Min: nan, Max: nan, Mean: nan


✅ The raster has data, but all values are NaN. Things I Have Checked

✅ Fire polygon (fire_region) is correctly defined (since NDVI visualizes correctly).
✅ MODIS NDVI images exist for each month (verified with size().getInfo()).
✅ NDVI is properly scaled (multiply(0.0001)).
✅ Reprojection to EPSG:4326 is applied before export.
✅ The export is correctly initiated and appears in GEE "Tasks".
❌ But the exported rasters contain only NaN values.


❓ Questions

Why does the NDVI render fine in GEE but export as NaN when using .median()?
Is there a better way to compute and export monthly NDVI from MOD13Q1?
Could the .median() operation be causing an issue with missing pixels?
Should scale=250 be modified during export, or is MODIS using an incompatible projection?


Any help would be greatly appreciated! Thanks in advance.

Continue reading...