Harvest Date Optimization Using Satellite Crop Phenology

Pinpoint optimal harvest windows from crop phenology trajectories. Klarety tracks NDVI progression and thermal accumulation from Sentinel-2 time-series to identify peak maturity dates and recommend harvest windows within 3-day precision. Reduces post-harvest losses and improves market timing for commodity producers.

Optimize harvest

Phenology-based harvest windows from satellite data

Klarety AI

Harvest date recommendations from one AOI prompt

Klarety agents track NDVI phenology curves and return 3-day precision harvest window recommendations with confidence ranges.

Phenology - Sentinel-2

NDVI trajectory and thermal accumulation modeling

Agents model growing degree days from Sentinel-2 and thermal indices to identify peak maturity across field polygons.

GIS Output

Phenology outputs for precision farming GIS

Export maturity stage rasters and harvest window polygons for farm management and agronomy decision system integration.

NDVI phenology curve peak maturity detection

Klarety agents fit a double-logistic phenology curve to the Sentinel-2 NDVI time series for each field polygon, identify the inflection point marking peak NDVI, and estimate the senescence onset date as the harvest window. Growing degree day accumulation is cross-referenced to validate crop maturity stage. Output is a per-field harvest recommendation table with 3-day window and confidence score.

Klarety AIGet harvest window recommendations for your fields

analysis/harvest_phenology_optimizer.pyAgent code

python

import eeimport numpy as npfrom scipy.optimize import curve_fitfrom datetime import datetime, timedelta
ee.Initialize()
# Kansas winter wheat AOIaoi = ee.Geometry.Rectangle([-98.5, 37.8, -97.5, 38.4])
# Sentinel-2 NDVI time series (Jan to Jul 2026)def get_ndvi_series():    collection = (ee.ImageCollection('COPERNICUS/S2_SR_HARMONIZED')                  .filterBounds(aoi)                  .filterDate('2026-01-01', '2026-07-15')                  .filter(ee.Filter.lt('CLOUDY_PIXEL_PERCENTAGE', 20))                  .map(lambda img: img.normalizedDifference(['B8','B4'])                                      .rename('ndvi')                                      .set('date', img.date().millis())))    return collection
series = get_ndvi_series()dates  = series.aggregate_array('date').getInfo()ndvi_vals = [series.filter(ee.Filter.eq('date', d)).first()             .reduceRegion(ee.Reducer.mean(), aoi, 10).get('ndvi').getInfo()             for d in dates]
# Day-of-year for fittingdoys = [(datetime.fromtimestamp(d/1000) - datetime(2026,1,1)).days for d in dates]ndvi_arr = np.array([v if v else 0 for v in ndvi_vals])doy_arr  = np.array(doys, dtype=float)
# Double-logistic phenology modeldef double_logistic(t, vmin, vmax, t_green, k_green, t_senesc, k_senesc):    green   = 1 / (1 + np.exp(-k_green   * (t - t_green)))    senesce = 1 / (1 + np.exp( k_senesc  * (t - t_senesc)))    return vmin + (vmax - vmin) * green * senesce
p0 = [0.15, 0.85, 80, 0.15, 170, 0.12]try:    popt, _ = curve_fit(double_logistic, doy_arr, ndvi_arr, p0=p0, maxfev=10000)    vmin, vmax, t_green, k_green, t_senesc, k_senesc = popt    # Peak NDVI = midpoint between green-up and senescence    peak_doy = (t_green + t_senesc) / 2    harvest_doy = t_senesc - 7  # 1 week before full senescence onset    harvest_date = datetime(2026,1,1) + timedelta(days=int(harvest_doy))    print(f"Green-up DOY: {t_green:.0f}")    print(f"Peak NDVI DOY: {peak_doy:.0f}")    print(f"Senescence onset DOY: {t_senesc:.0f}")    print(f"Recommended harvest window: {harvest_date.strftime('%b %d')} ± 3 days")except Exception as e:    print(f"Fit error: {e}")

output/harvest_window_report.mdOutput report