CropSense - Satellite, model and AI-based yield forecasting for agricultural crops
Abstract
The aim of this project is the development of methods for satellite, model and AI-based yield forecasting of agricultural crops in the context of Austrian agriculture. In order to achieve the project goal, the following technologies and methods will be combined or further developed: - Sentinel 2 spectral data (satellite images) will be used as input for the AI component, the reflection model PROSAIL and the crop growth model iCrop. - The AI component will be trained to recognize agricultural crops with manually marked and already available training data from the Austria and USA. The trained model should be able to correctly classify the crop plants visible on the satellite images to 90%. With the reflection model PROSAIL, possible reflection values are calculated by discrete variation of the input values and stored in a database. The reflection values shown in the satellite images are then compared with the values in the database in order to be able to draw conclusions about possible input value combinations and crop plants. - The results of the AI component and the reflection model will be transferred to the fuzzy logic classifier in order to finally determine the crop species. - The identified crop species and growth parameters derived from the spectral data (e.g. Leaf Area Index) are used to calibrate the iCrop model to generate yield forecasts (as well as harvest time, phenology, fertilizer and water requirements).
Publikationen
Project staff
Ahmad M. Manschadi
Assoc. Prof. Dr. Ahmad M. Manschadi
manschadi@boku.ac.at
Tel: +43 1 47654-95112
Project Leader
01.01.2021 - 30.06.2022