NDVI Crop Stress Detection Across Agricultural Regions

Detect crop stress and irrigation deficits before yield loss occurs. Klarety agents compute normalized difference vegetation index from Sentinel-2 10-meter bands, threshold anomaly zones against seasonal baselines, and generate field-level risk maps. Farmers and agronomists receive stress alerts every 5 days — no GIS skills, no satellite expertise.

Monitor crop health

Crop stress detection for agronomists and traders

Klarety AI

Field-level stress maps from one satellite query

Klarety agents compute NDVI anomaly zones from Sentinel-2 10m bands and return field-level risk maps every 5 days.

NDVI - Sentinel-2

Seasonal baseline anomaly thresholding

Agents compute NDVI deviations against seasonal baselines and classify stress severity by zone with confidence scores.

GIS Output

Stress layers for precision agriculture GIS

Export stress zone polygons with severity classifications for farm management systems and GEE workflow integration.

Sentinel-2 NDVI anomaly detection for crop stress

Klarety agents compute a 5-year seasonal NDVI baseline from Sentinel-2 for each field polygon, then flag current NDVI values more than 1.5 standard deviations below the baseline as stress zones. Severity is classified as mild (1.5-2 SD), moderate (2-3 SD), or severe (>3 SD). Output is a field-level GeoJSON with stress severity, confidence, and affected area per field.

Klarety AIMonitor crop stress in your region

analysis/ndvi_crop_stress_monitor.pyAgent code

python

import eeimport numpy as np
ee.Initialize()
# Central Valley California AOIaoi = ee.Geometry.Rectangle([-120.8, 36.4, -119.8, 37.2])
# 5-year NDVI baseline (same DOY window)def ndvi_baseline(year):    return (ee.ImageCollection('COPERNICUS/S2_SR_HARMONIZED')            .filterBounds(aoi)            .filterDate(f'{year}-04-01', f'{year}-04-30')            .filter(ee.Filter.lt('CLOUDY_PIXEL_PERCENTAGE', 20))            .map(lambda img: img.normalizedDifference(['B8','B4']).rename('ndvi'))            .mean())
baseline_years = [2020, 2021, 2022, 2023, 2024]baseline_stack = ee.ImageCollection([ndvi_baseline(y) for y in baseline_years])ndvi_mean = baseline_stack.mean().rename('ndvi_mean')ndvi_std  = baseline_stack.reduce(ee.Reducer.stdDev()).rename('ndvi_std')
# Current NDVI (April 2026)current = (ee.ImageCollection('COPERNICUS/S2_SR_HARMONIZED')           .filterBounds(aoi)           .filterDate('2026-04-01', '2026-04-30')           .filter(ee.Filter.lt('CLOUDY_PIXEL_PERCENTAGE', 20))           .map(lambda img: img.normalizedDifference(['B8','B4']).rename('ndvi'))           .mean())
# Z-score anomalyz_score = ndvi_mean.subtract(current).divide(ndvi_std.add(0.001)).rename('z_score')
# Stress classificationmild_stress   = z_score.gte(1.5).And(z_score.lt(2.0))mod_stress    = z_score.gte(2.0).And(z_score.lt(3.0))severe_stress = z_score.gte(3.0)
def area_ha(mask):    a = mask.multiply(ee.Image.pixelArea()).reduceRegion(        ee.Reducer.sum(), aoi, 10, maxPixels=1e9    ).values().get(0).getInfo()    return round((a or 0) / 10000, 1)
print(f"Mild stress (1.5-2.0 SD):  {area_ha(mild_stress):>8.0f} ha")print(f"Moderate stress (2-3 SD):  {area_ha(mod_stress):>8.0f} ha")print(f"Severe stress (>3 SD):     {area_ha(severe_stress):>8.0f} ha")

output/crop_stress_report.mdOutput report