Satellites map which crops are growing where, how healthy they are and how dry the land is, so India can forecast harvests and respond early to drought.
From space, a healthy green field looks very different from a dry or bare one, especially in near-infrared light that our eyes cannot see. Satellites can survey every district of India in days. That makes them a powerful tool for agriculture.
ISRO began working on satellite crop forecasting in the early 1980s, which led to a project called CAPE. Today the programme is called FASAL: Forecasting Agricultural output using Space, Agro-meteorology and Land based observations. ISRO says it gives in-season production forecasts for nine field crops, including wheat, rice, jute, mustard, cotton, sugarcane and rabi sorghum.
The forecasts are made by the Mahalanobis National Crop Forecast Centre, set up in New Delhi in April 2012 by the Ministry of Agriculture and Farmers' Welfare. It works with the India Meteorological Department, state agriculture departments and state remote sensing centres.
Take sugarcane. A 2019 paper from the centre describes two forecasts. The first, at the end of June, uses several AWiFS images (56 m) from Resourcesat to estimate the area under sugarcane for the country and each state. The second, at the end of November, uses sharper LISS-3 images (23 m) along with Europe's Sentinel-2 and the US Landsat-8, down to district level.
Satellites cannot do it alone. Field teams record crop type, sowing date and health at known locations using Bhuvan FASAL, a phone app built by ISRO's National Remote Sensing Centre. Those ground checks train the computer to recognise each crop in the images.
ISRO also lists an operational agricultural drought assessment that uses satellite data and helps the government decide on relief. Other uses include horticulture surveys, soil mapping and experimental crop insurance.
Search for Resourcesat-2 and Resourcesat-2A: they pass over India in the morning, when the light suits crop imaging. Find the Resourcesat satellites
The workhorse number is NDVI, the Normalized Difference Vegetation Index:
NDVI = (NIR − Red) / (NIR + Red)
where NIR and Red are the reflectances in the near-infrared and red bands. On AWiFS these are band B4 (0.77–0.86 μm) and band B3 (0.62–0.68 μm).
Healthy leaves absorb red light for photosynthesis and strongly reflect near-infrared. Bare soil reflects both about equally. Worked examples with typical-looking values:
Healthy crop: NIR = 0.45, Red = 0.05 → (0.45 − 0.05) / (0.45 + 0.05) = 0.40 / 0.50 = 0.80 Stressed crop: NIR = 0.30, Red = 0.10 → 0.20 / 0.40 = 0.50 Bare soil: NIR = 0.25, Red = 0.20 → 0.05 / 0.45 ≈ 0.11
Clouds spoil this, because they reflect red and near-infrared almost equally, which drags NDVI down. NRSC's AWiFS product handles this by maximum-value compositing: over each 15-day period it keeps, for every pixel, the highest NDVI seen. Cloudy days give low values and are dropped automatically.
Area estimation then works by classification. Using ground-truth fields, the software learns the spectral signature of, say, sugarcane through the season, labels every pixel, and counts them. With 56 m pixels, each pixel is 3,136 m², about 0.31 hectare; a district with 500,000 sugarcane pixels would have about 157,000 hectares under the crop.
The 2019 paper compared FASAL estimates with official statistics and found correlations of 0.94 for area and 0.95 for production, but only 0.60 for yield, showing that yield remains the hardest part.
Forecasting Agricultural output using Space, Agro-meteorology and Land based observations.
They absorb red light for photosynthesis and reflect near-infrared strongly, so (NIR − Red)/(NIR + Red) is large.
Ground truth on crop type and condition at known spots is needed to train and check the classification of satellite images.