Calculating the harvest of a specific crop type on a regional, national or global scale is an important input variable for a large number of industries. Farmers, agricultural commodity traders, food brands and grocery chains depend on the supply of crops to determine product pricing. To mitigate the risks of fluctuating prices, commodity futures at fixed prices are negotiated on crops long before the harvest.
Crop yield per acre can be calculated based on phenological formulas, weather and crop observation. However, all relevant parameters for estimating crop growth, e.g. meteorological data on precipitation and temperatures are seldomly tracked at large scale. Interdependencies between crop yields across regions and nations are not taken into account, although having a significant influence on market prices. Moreover, reliable information on crop growth and expected harvests are seldomly monitored throughout the year and made available to the public.
Through optical and radar satellite data (Sentinel-1 and -2), crops can be classified per parcel by applying convolutional neural networks. Moreover, its biomass index (NDVI), a standard normalized vegetation index, can be calculated. Creating a time series observation throughout the year, e.g. by pixel-set encoders and temporal self-attention, spanning over all major cultivation areas provides information about the total amount of crops cultivated and its current vegetation status. Taking into account meteorological data and modelling growth behaviour based on growth patterns in the past, the amount of harvested crops can be estimated.
Receive news about Machine Learning and news around dida.
Successfully signed up.
Valid email address required.
Email already signed up.
Something went wrong. Please try again.