Transferability of Machine Learning Models for Crop Classification in Remote Sensing Imagery - A Study on Phenological, Temporal, and Spatial Influences

Aug 5, 2026

Upload by Ezra

Machine learning models can classify crops from satellite imagery with high accuracy, yet their transferability to new regions often remains limited. This study evaluates model transfer using Sentinel-2 data and introduces a new testing methodology that systematically measures spatial transfer quality. Random Forest, XGBoost, SGD, MLP, SVM, and majority voting are tested across 18 scenarios covering phenological, temporal, spatial, and data-quantity influences, showing accuracy declines as time gaps increase.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages20

Reading Time50 min

AccessFree to download

Abstract

Introduction

Background and motivation

Methods and test methodology

Models evaluated

Test scenarios and influence factors

Results and findings

Accuracy and transferability trends

4.2 8.3 Article Transferability of Machine Learning Models for Crop Classification in Remote Sensing Imagery Using a New Test Methodology: A Study on Phenological, Temporal, and Spatial Influences Hauke Hoppe, Peter Dietrich, Philip Marzahn, Thomas Weiß, Christian Nitzsche, Uwe Freiherr von Lukas, Thomas Wengerek and Erik Borg Special Issue In Situ Data in the Interplay of Remote Sensing II Edited by Prof. Dr. Peter Dietrich, Prof. Dr. Erik Borg, Dr. Mona Ahmad Mahmoud Morsy and Dr. Mahmud Haghshenas Haghighi [https://doi.org/10.3390/rs16091493](https://doi.org/10.3390/rs16091493) remote sensing Article Transferability of Machine Learning Models for Crop Classi􀀂cation in Remote Sensing Imagery Using a New Test Methodology: A Study on Phenological, Temporal, and Spatial In􀀃uences Hauke Hoppe 1, *, Peter Dietrich 2,3, Philip Marzahn 4, Thomas Weiß 1,4, Christian Nitzsche 1, Uwe Freiherr von Lukas 1,5, Thomas Wengerek 6 and Erik Borg 7,8 Citation: Hoppe, H.; Dietrich, P.; Marzahn, P.; Weiß, T.; Nitzsche, C.; Freiherr von Lukas, U.; Wengerek, T.; Borg, E. Transferability of Machine Learning Models for Crop Classi􀀂cation in Remote Sensing Imagery Using a New Test Methodology: A Study on Phenological, Temporal, and Spatial In􀀃uences. Remote Sens. 2024, 16, 1493 . [https://doi.org/10.3390/rs16091493](https://doi.org/10.3390/rs16091493) Academic Editor: Peng Fu Received: 27 February 2024 Revised: 5 April 2024 Accepted: 18 April 2024 Published: 23 April 2024 Copyright: © 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license ([https://](https://)[ ](https://)[creativecommons.org/licenses/by/](creativecommons.org/licenses/by/)[ ](creativecommons.org/licenses/by/)[4.0/](4.0/)) . 1 Fraunhofer Institute for Computer Graphics Research (IGD), Joachim-Jungius-Str. 11, D-18059 Rostock, Germany; [thomas.weiss@igd-r.fraunhofer.de](thomas.weiss@igd-r.fraunhofer.de) (T.W.); [christian.nitzsche@igd-r.fraunhofer.de](christian.nitzsche@igd-r.fraunhofer.de) (C.N.); [uwe.freiherr.von.lukas@igd-r.fraunhofer.de](uwe.freiherr.von.lukas@igd-r.fraunhofer.de) (U.F.v.L.) 2 Environmental and Engineering Geophysics, Eberhard Karls University Tübingen, Schnarrenbergstr. 94-96, D-72076 Tübingen, Germany; [peter.dietrich@ufz.de](peter.dietrich@ufz.de) 3 Department of Monitoring and Exploration Technologies, Helmholtz Center for Environmental Research, D-04318 Leipzig, Germany 4 Geodesy and Geoinformatics, University of Rostock, D-18059 Rostock, Germany; [philip.marzahn@uni-rostock.de](philip.marzahn@uni-rostock.de) 5 Institute for Visual and Analytic Computing, University of Rostock, D-18059 Rostock, Germany 6 Faculty of Economics, Hochschule Stralsund, University of Applied Sciences, D-18435 Stralsund, Germany; [thomas.wengerek@hochschule-stralsund.de](thomas.wengerek@hochschule-stralsund.de) 7 German Aerospace Center, German Remote Sensing Data Center, National Ground Segment, D-17235 Neustrelitz, Germany; [erik.borg@dlr.de](erik.borg@dlr.de) 8 Geoinformatics and Geodesy, Neubrandenburg University of Applied Sciences, D-17033 Neubrandenburg, Germany * Correspondence: hauke.hoppe@igd-r.fraunhofer.de Abstract: Machine learning models are used to identify crops in satellite data, which achieve high classi􀀂cation accuracy but do not necessarily have a high degree of transferability to new regions. This paper investigates the use of machine learning models for crop classi􀀂cation using Sentinel-2 imagery. It proposes a new testing methodology that systematically analyzes the quality of the spatial transfer of trained models. In this study, the classi􀀂cation results of Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Stochastic Gradient Descent (SGD), Multilayer Perceptron (MLP), Support Vector Machines (SVM), and a Majority Voting of all models and their spatial trans

FAQs

Why do crop classification models need transferability evaluation across regions?

High classification accuracy on a given region does not guarantee reliable performance in new regions. Transferability assessment addresses this gap by measuring how models generalize spatially.

What testing methodology is proposed in the study?

The methodology systematically analyzes the spatial transfer quality of trained machine learning models using Sentinel-2 imagery and defines 18 test scenarios.

How do results change when models are trained on different temporal extents?

Model accuracies tend to decrease with increasing time differences between regions; the study reports varying combined F1-scores depending on whether training uses a single day, half-season, or the full growing season.

Aug 5, 2026

Upload by Ezra