The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Share
Publications Copernicus
Download
Citation
Share
Articles | Volume XLVIII-G-2025
https://doi.org/10.5194/isprs-archives-XLVIII-G-2025-1315-2025
https://doi.org/10.5194/isprs-archives-XLVIII-G-2025-1315-2025
31 Jul 2025
 | 31 Jul 2025

AI4EO hyperview challenge: combination of machine learning methods on hyperspectral images to predict the soil parameters

Marsia Sanità, Eva Savina Malinverni, Roberto Pierdicca, Adriano Mancini, Ewa Glowienka, and Lindo Nepi

Keywords: Remote Sensing, Machine Learning Multi-output Regression Model, Hyperspectral, Soil Parameters

Abstract. In the AI4EO educational challenge "Seeing Beyond the Visible", hyperspectral images are used to predict the chemical parameters on the soil (K, Mg, P2O5, pH) in anticipation of the correct use of fertilisers. The challenge is set in an agricultural area of Poland and the available data are hyperspectral images (150 contiguous hyperspectral bands) and in situ samples for soil parameter measurements. The aim of this challenge was to advance the state of art of soil parameter analysis by hyperspectral images. Having a good knowledge of the chemical characteristics of the soil is important in order to be able to identify which types of crops are most suitable in that area to optimise production and reduce the use of fertilisers. In the face of ongoing climate change and the disastrous calamitous events that follow, the idea of a sustainable agriculture becomes a necessity. Artificial intelligence (AI) through Machine Learning (ML) and Deep Learning (DL) techniques can be a great support for farmers in optimising the use of natural resources and ensuring better land management. In this paper, a group of engineers in the field of data science and geomatics carries out this research topic accepting the challenge proposed by AI4EO. A variety of AI techniques were applied by the authors of this paper with respect to the other participants in the challenge methods. The proposed approach is based on the novelty of a dataset filtering and on the use of a Random Forest Multi-Output Regressor.

Share