The purpose of this paper is to explore the dynamics of neural networks in forecasting crop (wheat) yield using remote sensing and other data. We use the Adaptive Neuro-Fuzzy Inference System (ANFIS). The input to ANFIS are several parameters derived from the crop growth simulation model (CGMS) including soil moisture content, above ground biomass, and storage organs biomass. In addition we use remote sensing information in the form of the Normalized Difference Vegetation Index (NDVI). ANFIS has only one output node, the yield. In other words a single number is sought. An additional difficulty in predicting yield is that remote sensing data do not go long back in time. Hence any predicting effort is forced to use a very limited number of past years in order to construct a model to forecast future values. The system is trained by leaving one year out and using all the other data. We then evaluate the deviation of our estimate compared to the yield of the year that is left out. The procedure is applied to all the years and the average forecasting accuracy is given.
STATHAKIS Dimitrios;
SAVIN Igor;
NEGRE Thierry;
2007-06-01
International Society for Photogrammetry and Remote Sensing (ISPRS)
JRC36837
https://publications.jrc.ec.europa.eu/repository/handle/JRC36837,
| Name | Country | City | Type |
|---|
This document is only visible at the Commission level.
You are not authorized to publish or distribute it outside the European Commission.
This is a public document. You can share this publication.
Datasets
| ID | Title | Public URL |
|---|
Dataset collections
| ID | Acronym | Title | Public URL |
|---|
Scripts / source codes
| Description | Public URL |
|---|
Additional supporting files
| File name | Description | File type |
|---|