Crop Mapping without Labels: Investigating Temporal and Spatial Transferability of Crop Classification Models Using a 5-Year Sentinel-2 Series and Machine Learning
<p>Study areas and distribution of the croplands.</p> "> Figure 2
<p>Typical crop proportions (% on <span class="html-italic">y</span>-axis) in the two studied regions based on official agricultural statistics. Available online <a href="http://www.statistics.sk" target="_blank">www.statistics.sk</a> (accessed on 25 May 2023).</p> "> Figure 3
<p>Overall workflow of the study.</p> "> Figure 4
<p>Observational quality of input monthly median composites. Number of observations used for median compositing and proportion of total coverage (% on <span class="html-italic">x</span>-axis) of the study regions.</p> "> Figure 5
<p>Long-term (2018–2022) means and Standard Deviations (SD) of spectral bands for Danubian lowland cereal crops, rapeseed, and grasslands. <span class="html-italic">y</span>-axis represents reflectance (0–1) values multiplied by 10,000. X axis represents Sentinel-2 spectral band for specific month.</p> "> Figure 6
<p>Long-term (2018–2022) means and standard deviations (SD) of spectral bands for East Slovakian lowland cereal crops, rapeseed, and grasslands. <span class="html-italic">y</span>-axis represents reflectance (0–1) values multiplied by 10,000. X axis represents Sentinel-2 spectral band for specific month.</p> "> Figure 7
<p>Long-term (2018–2022) means and Standard Deviations (SD) of spectral bands for Danubian lowland summer crops. <span class="html-italic">y</span>-axis represents reflectance (0–1) values multiplied by 10,000. X axis represents Sentinel-2 spectral band for specific month.</p> "> Figure 8
<p>Long-term (2018–2022) means and Standard Deviations (SD) of spectral bands for East Slovakian lowland summer crops. <span class="html-italic">y</span>-axis represents reflectance (0–1) values multiplied by 10,000. X axis represents Sentinel-2 spectral band for specific month.</p> "> Figure 9
<p>Variability in the spectral band B6 distributions of soybeans during July and August across different years and regions. <span class="html-italic">x</span>-axis represents reflectance (0–1) values multiplied by 10,000.</p> "> Figure 10
<p>Variability in the spectral band B6 distributions of grasslands during July and August across different years and regions. <span class="html-italic">x</span>-axis represents reflectance values (0–1) multiplied by 10,000.</p> "> Figure 11
<p>Overall accuracies (OA in % on Y Axis) across years: Scenario 1 represents the temporal transferability analysis in the Danubian lowland, while Scenario 2 pertains to the East Slovakian lowland region.</p> "> Figure 12
<p>Crop-Specific Performance Using F1 Statistics (in % on Y Axis) for All Scenarios: Scenario 1 represents the temporal transferability analysis in the Danubian Lowland, while Scenario 2 focuses on the temporal transferability analysis in the East Slovakian Lowland region. Scenario 3 involves the spatial transferability scenario, using 1 year of data from the Danubian Lowland for training and the same year of data from the East Slovakian Lowland for testing. Scenario 4 encompasses all years (2018–2022) of data from the Danubian Lowland for training while testing each individual year of data from the East Slovakian Lowland separately. Tabulated data are provided in <a href="#app1-remotesensing-15-03414" class="html-app">Table S2, Supplement S2</a>.</p> "> Figure 13
<p>Classification performance on different scenarios.</p> "> Figure 14
<p>Frequency at which specific features are identified among the top 10 most important features in all considered classifiers for Scenario 1 and 2022 tested year (A) and their aggregation according to spectral bands (B) and monthly period (C).</p> "> Figure 15
<p>Evolution of overall accuracies (OA in % on Y Axis) with sequential increases in input features within the season 2022.</p> "> Figure 16
<p>Evolution of overall accuracies (OA in % on Y Axis) with sequential increases in input features within the season 2020.</p> "> Figure 17
<p>Spatial distribution of crops for 2022 in Danubian lowlands and Eastern Slovakia. The classification model for Danubian lowlands followed the Scenario 1, e.g., the train data included reference crop labels in Danubian region spanning years 2018, 2019, 2020, and 2021. The classification model for eastern Slovakia followed Scenario 2, e.g., the train data included reference crop labels in eastern Slovakia’s lowlands spanning years 2018, 2019, 2020, and 2021. Hence, year 2022, with prolonged summer drought, was not used for the training.</p> "> Figure 18
<p>Spatial distribution of misclassification errors and parcel-based confidence levels for Scenario 1 in tested year 2022 in Danubian lowland region. The confidence levels range from lowest to highest, where the lowest confidence level indicates that only one model correctly classified the parcel, the medium confidence level indicates that two or three models correctly classified the parcel, and the highest confidence level indicates that all models correctly classified the parcel. The misclassification errors are displayed in parcels where none of the models correctly classified the given parcel. Spatial distribution of misclassification errors and parcel-based confidence levels for Scenario 1 in tested year 2022 in Danubian lowland region. The confidence levels range from lowest to highest, where the lowest confidence level indicates that only one model correctly classified the parcel, the medium confidence level indicates that two or three models correctly classified the parcel, and the highest confidence level indicates that all models correctly classified the parcel. The misclassification errors are displayed in parcels where none of the models correctly classified the given parcel.</p> ">
Abstract
:1. Introduction
- Developing an effective workflow for broad-scale crop mapping using machine learning techniques that can be easily deployed for nationwide agricultural monitoring.
- Investigating the transferability capacities of the developed crop classification models in both temporal and spatial aspects.
- Providing analysis-ready datasets to the remote sensing community for further testing and supporting the on-going development of improved methods.
2. Materials and Methods
2.1. Study Area
2.2. Satellite Data
2.3. Methodology
- Preprocessing of satellite products.
- Creating a reference dataset by extracting spectro-temporal information.
- Training different classification models.
- Analyzing performance and accuracy assessment.
- Applying models to images.
2.3.1. Preprocessing of Satellite Products
2.3.2. Creating Training and Test Datasets
2.3.3. Training Different Models
2.3.4. Analyzing Performance and Accuracy Assessment
2.3.5. Applying Models to Images
3. Results
3.1. Observational Quality
3.2. Crop Specific Spectral Response across Regions and Seasons
3.3. Temporal Transferability
3.4. Spatial Transferability
3.5. Feature Importance
3.6. Confidence Map
4. Discussion
4.1. Spatiotemporal Generalization
4.2. Observation Quality Consideration
4.3. Effect of Anomaly Seasons
4.4. Regional Variability Consideration
4.5. Feature Importance
4.6. Future Prospect: Multi-Sensor Synergies
4.7. Future Prospect: Reference Datasets
4.8. Future Prospect: Multi-Model Consideration
5. Conclusions
Supplementary Materials
Author Contributions
Funding
Data Availability Statement
Acknowledgments
Conflicts of Interest
References
- Vuolo, F.; Neuwirth, M.; Immitzer, M.; Atzberger, C.; Ng, W.-T. How Much Does Multi-Temporal Sentinel-2 Data Improve Crop Type Classification? Int. J. Appl. Earth Obs. Geoinf. 2018, 72, 122–130. [Google Scholar] [CrossRef]
- Tuia, D.; Persello, C.; Bruzzone, L. Recent Advances in Domain Adaptation for the Classification of Remote Sensing Data. IEEE Geosci. Remote Sens. Mag. 2016, 4, 41–57. [Google Scholar] [CrossRef]
- Belgiu, M.; Bijker, W.; Csillik, O.; Stein, A. Phenology-Based Sample Generation for Supervised Crop Type Classification. Int. J. Appl. Earth Obs. Geoinf. 2021, 95, 102264. [Google Scholar] [CrossRef]
- Yang, Z.; Diao, C.; Gao, F. Towards Scalable within-Season Crop Mapping with Phenology Normalization and Deep Learning. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2023, 16, 1390–1402. [Google Scholar] [CrossRef]
- Lin, C.; Zhong, L.; Song, X.-P.; Dong, J.; Lobell, D.B.; Jin, Z. Early- and in-Season Crop Type Mapping without Current-Year Ground Truth: Generating Labels from Historical Information via a Topology-Based Approach. Remote Sens. Environ. 2022, 274, 112994. [Google Scholar] [CrossRef]
- Sexton, J.O.; Urban, D.L.; Donohue, M.J.; Song, C. Long-Term Land Cover Dynamics by Multi-Temporal Classification across the Landsat-5 Record. Remote Sens. Environ. 2013, 128, 246–258. [Google Scholar] [CrossRef]
- Luo, Y.; Zhang, Z.; Zhang, L.; Han, J.; Cao, J.; Zhang, J. Developing High-Resolution Crop Maps for Major Crops in the European Union Based on Transductive Transfer Learning and Limited Ground Data. Remote Sens. 2022, 14, 1809. [Google Scholar] [CrossRef]
- Wang, S.; Azzari, G.; Lobell, D.B. Crop Type Mapping without Field-Level Labels: Random Forest Transfer and Unsupervised Clustering Techniques. Remote Sens. Environ. 2019, 222, 303–317. [Google Scholar] [CrossRef]
- Zhong, L.; Gong, P.; Biging, G.S. Efficient Corn and Soybean Mapping with Temporal Extendability: A Multi-Year Experiment Using Landsat Imagery. Remote Sens. Environ. 2014, 140, 1–13. [Google Scholar] [CrossRef]
- Qin, R.; Liu, T. A Review of Landcover Classification with Very-High Resolution Remotely Sensed Optical Images—Analysis Unit, Model Scalability and Transferability. Remote Sens. 2022, 14, 646. [Google Scholar] [CrossRef]
- Hu, Y.; Zeng, H.; Tian, F.; Zhang, M.; Wu, B.; Gilliams, S.; Li, S.; Li, Y.; Lu, Y.; Yang, H. An Interannual Transfer Learning Approach for Crop Classification in the Hetao Irrigation District, China. Remote Sens. 2022, 14, 1208. [Google Scholar] [CrossRef]
- You, N.; Dong, J. Examining Earliest Identifiable Timing of Crops Using All Available Sentinel 1/2 Imagery and Google Earth Engine. ISPRS J. Photogramm. Remote Sens. 2020, 161, 109–123. [Google Scholar] [CrossRef]
- Johnson, D.M.; Mueller, R. Pre- and within-Season Crop Type Classification Trained with Archival Land Cover Information. Remote Sens. Environ. 2021, 264, 112576. [Google Scholar] [CrossRef]
- Cai, Y.; Guan, K.; Peng, J.; Wang, S.; Seifert, C.; Wardlow, B.; Li, Z. A High-Performance and in-Season Classification System of Field-Level Crop Types Using Time-Series Landsat Data and a Machine Learning Approach. Remote Sens. Environ. 2018, 210, 35–47. [Google Scholar] [CrossRef]
- Lapin, M.; Faako, P.; Melo, M.; Stastny, P.; Tomlain, J. Climatic Regions; 1:1,000,000; 27. Klimaticke Oblasti; 1:1,000,000; Ministry of Environment of the Slovak Republic Bratislava: Bratislava, Slovakia, 2002; ISBN 80-88833-27-2. [Google Scholar]
- Michaeli, E.; Vilček, J.; Ivanová, M. Characteristics of Agricultural Soils in the East-Slovak Lowland and the Possibilities of Improving of Their Productive Potential. Zivotn. Prostr. 2013, 47, 242–246. [Google Scholar]
- Miklós, L.; Izakovičová, Z. Atlas of Representative Geoecosystems of Slovakia; Slovak Academy of Sciences, Ministry of Environment and Ministry of Education of the Slovak Republik: Bratislava, Slovakia, 2006; pp. 95–123. ISBN 80-969272-4-8. [Google Scholar]
- Griffiths, P.; Van Der Linden, S.; Kuemmerle, T.; Hostert, P. A Pixel-Based Landsat Compositing Algorithm for Large Area Land Cover Mapping. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2013, 6, 2088–2101. [Google Scholar] [CrossRef]
- Bull, A.D. Convergence Rates of Efficient Global Optimization Algorithms. J. Mach. Learn. Res. 2011, 12, 2879–2904. [Google Scholar]
- Breiman, L. Random Forests. Mach. Learn. 2001, 45, 5–32. [Google Scholar] [CrossRef] [Green Version]
- Phiri, D.; Morgenroth, J. Developments in Landsat Land Cover Classification Methods: A Review. Remote Sens. 2017, 9, 967. [Google Scholar] [CrossRef] [Green Version]
- Mountrakis, G.; Im, J.; Ogole, C. Support Vector Machines in Remote Sensing: A Review. ISPRS J. Photogramm. Remote Sens. 2011, 66, 247–259. [Google Scholar] [CrossRef]
- Mas, J.F.; Flores, J.J. The Application of Artificial Neural Networks to the Analysis of Remotely Sensed Data. Int. J. Remote Sens. 2008, 29, 617–663. [Google Scholar] [CrossRef]
- Foody, G.M. Status of Land Cover Classification Accuracy Assessment. Remote Sens. Environ. 2002, 80, 185–201. [Google Scholar] [CrossRef]
- Fisher, A.; Rudin, C.; Dominici, F. All Models Are Wrong, but Many Are Useful: Learning a Variable’s Importance by Studying an Entire Class of Prediction Models Simultaneously. J. Mach. Learn. Res. 2019, 20, 1–81. [Google Scholar]
- Liu, W.; Gopal, S.; Woodcock, C.E. Uncertainty and Confidence in Land Cover Classification Using a Hybrid Classifier Approach. Photogramm. Eng. Remote Sens. 2004, 70, 963–971. [Google Scholar] [CrossRef]
- Sykas, D.; Sdraka, M.; Zografakis, D.; Papoutsis, I. A Sentinel-2 Multiyear, Multicountry Benchmark Dataset for Crop Classification and Segmentation with Deep Learning. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2022, 15, 3323–3339. [Google Scholar] [CrossRef]
- Weikmann, G.; Paris, C.; Bruzzone, L. TimeSen2Crop: A Million Labeled Samples Dataset of Sentinel 2 Image Time Series for Crop-Type Classification. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2021, 14, 4699–4708. [Google Scholar] [CrossRef]
- Flood, N. Seasonal Composite Landsat TM/ETM+ Images Using the Medoid (a Multi-Dimensional Median). Remote Sens. 2013, 5, 6481–6500. [Google Scholar] [CrossRef] [Green Version]
- Frantz, D.; Röder, A.; Stellmes, M.; Hill, J. Phenology-Adaptive Pixel-Based Compositing Using Optical Earth Observation Imagery. Remote Sens. Environ. 2017, 190, 331–347. [Google Scholar] [CrossRef] [Green Version]
- Roy, D.P.; Yan, L. Robust Landsat-Based Crop Time Series Modelling. Remote Sens. Environ. 2020, 238, 110810. [Google Scholar] [CrossRef]
- Zhang, L.; Gao, L.; Huang, C.; Wang, N.; Wang, S.; Peng, M.; Zhang, X.; Tong, Q. Crop Classification Based on the Spectrotemporal Signature Derived from Vegetation Indices and Accumulated Temperature. Int. J. Digit. Earth 2022, 15, 626–652. [Google Scholar] [CrossRef]
- Wei, M.; Wang, H.; Zhang, Y.; Li, Q.; Du, X.; Shi, G.; Ren, Y. Investigating the Potential of Crop Discrimination in Early Growing Stage of Change Analysis in Remote Sensing Crop Profiles. Remote Sens. 2023, 15, 853. [Google Scholar] [CrossRef]
- Immitzer, M.; Vuolo, F.; Atzberger, C. First Experience with Sentinel-2 Data for Crop and Tree Species Classifications in Central Europe. Remote Sens. 2016, 8, 166. [Google Scholar] [CrossRef]
- Yi, Z.; Jia, L.; Chen, Q. Crop Classification Using Multi-Temporal Sentinel-2 Data in the Shiyang River Basin of China. Remote Sens. 2020, 12, 4052. [Google Scholar] [CrossRef]
- Claverie, M.; Ju, J.; Masek, J.G.; Dungan, J.L.; Vermote, E.F.; Roger, J.-C.; Skakun, S.V.; Justice, C. The Harmonized Landsat and Sentinel-2 Surface Reflectance Data Set. Remote Sens. Environ. 2018, 219, 145–161. [Google Scholar] [CrossRef]
- Saunier, S.; Pflug, B.; Lobos, I.M.; Franch, B.; Louis, J.; De Los Reyes, R.; Debaecker, V.; Cadau, E.G.; Boccia, V.; Gascon, F.; et al. Sen2Like: Paving the Way towards Harmonization and Fusion of Optical Data. Remote Sens. 2022, 14, 3855. [Google Scholar] [CrossRef]
- Griffiths, P.; Nendel, C.; Hostert, P. Intra-Annual Reflectance Composites from Sentinel-2 and Landsat for National-Scale Crop and Land Cover Mapping. Remote Sens. Environ. 2019, 220, 135–151. [Google Scholar] [CrossRef]
- Kussul, N.; Mykola, L.; Shelestov, A.; Skakun, S. Crop Inventory at Regional Scale in Ukraine: Developing in Season and End of Season Crop Maps with Multi-Temporal Optical and SAR Satellite Imagery. Eur. J. Remote Sens. 2018, 51, 627–636. [Google Scholar] [CrossRef] [Green Version]
- Liu, Z.; Zhang, L.; Yu, Y.; Xi, X.; Ren, T.; Zhao, Y.; Zhu, D.; Zhu, A. Cross-Year Reuse of Historical Samples for Crop Mapping Based on Environmental Similarity. Front. Plant Sci. 2022, 12, 761148. [Google Scholar] [CrossRef]
- Rußwurm, M.; Pelletier, C.; Zollner, M.; Lefèvre, S.; Körner, M. BreizhCrops: A Time Series Dataset for Crop Type Mapping. arXiv 2020, arXiv:1905.11893. [Google Scholar] [CrossRef]
- Aghababaei, M.; Ebrahimi, A.; Naghipour, A.A.; Asadi, E.; Pérez-Suay, A.; Morata, M.; Garcia, J.L.; Rivera Caicedo, J.P.; Verrelst, J. Introducing ARTMO’s Machine-Learning Classification Algorithms Toolbox: Application to Plant-Type Detection in a Semi-Steppe Iranian Landscape. Remote Sens. 2022, 14, 4452. [Google Scholar] [CrossRef]
- Tang, B.; Bi, Y.; Li, Z.-L.; Xia, J. Generalized Split-Window Algorithm for Estimate of Land Surface Temperature from Chinese Geostationary FengYun Meteorological Satellite (FY-2C) Data. Sensors 2008, 8, 933–951. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- Peña-Barragán, J.M.; Ngugi, M.K.; Plant, R.E.; Six, J. Object-Based Crop Identification Using Multiple Vegetation Indices, Textural Features and Crop Phenology. Remote Sens. Environ. 2011, 115, 1301–1316. [Google Scholar] [CrossRef]
- Gilcher, M.; Udelhoven, T. Field Geometry and the Spatial and Temporal Generalization of Crop Classification Algorithms—A Randomized Approach to Compare Pixel Based and Convolution Based Methods. Remote Sens. 2021, 13, 775. [Google Scholar] [CrossRef]
- Yuan, Y.; Lin, L. Self-Supervised Pretraining of Transformers for Satellite Image Time Series Classification. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2021, 14, 474–487. [Google Scholar] [CrossRef]
- Nowakowski, A.; Mrziglod, J.; Spiller, D.; Bonifacio, R.; Ferrari, I.; Mathieu, P.P.; Garcia-Herranz, M.; Kim, D.-H. Crop Type Mapping by Using Transfer Learning. Int. J. Appl. Earth Obs. Geoinf. 2021, 98, 102313. [Google Scholar] [CrossRef]
- Krichen, M.; Mihoub, A.; Alzahrani, M.Y.; Adoni, W.Y.H.; Nahhal, T. Are Formal Methods Applicable to Machine Learning and Artificial Intelligence? In Proceedings of the 2022 2nd International Conference of Smart Systems and Emerging Technologies (SMARTTECH), Riyadh, Saudi Arabia, 9–11 May 2022; pp. 48–53. [Google Scholar]
- Raman, R.; Gupta, N.; Jeppu, Y. Framework for Formal Verification of Machine Learning Based Complex System-of-Systems. Insight 2023, 26, 91–102. [Google Scholar] [CrossRef]
Transfer Scenario | Satellite Platform | Source Domain | Target Domain | Geographical Location | Crop Configuration/ Nomenclature | Classifier/ Method | Accuracy | Reference |
---|---|---|---|---|---|---|---|---|
Spatial | Sentinel-2 | England, France | 10 Countries | Europe | 4 crops | RF | 89% | [7] |
Spatial | Sentinel-2 | Zeeland region | Flevo-land, Friesland | Netherland | 10 crops | Dynamic Time Warping | 69–75% | [3] |
Temporal | Landsat | 2006–2010 a | 2006–2010 | Kansas | 3 crops | RF | 83.4% | [9] |
Temporal | Sentinel-2 | 2017–2018 | 2019 | Midwest US, NE China, Hauts-de-France | 3, 4, 8 crops | RF | 90.7; 89.8; 83.7% | [5] |
Temporal | Sentinel-1/ Sentinel-2 | 2020 | 2019 | Hetao Irrigation District | 6 crops | RF | 92% | [11] |
Temporal | Sentinel-1/ Sentinel-2 | 2017 | 2018 | Heilongjang | 4 crops | RF | 91% | [12] |
Temporal | Sentinel-2 | 2016–2019 | 2020 | 16 States across USA | 3 crops | RF | 71.3 b | [13] |
Temporal | Landsat | 2010–2015 | 2016 c | 9 States across USA | 3 crops | RF | 70% | [8] |
Temporal | Landsat | 2000–2014 | 2015 | Illinois | 2 crops | DNN | 96% | [14] |
Train Region | Train Years | Test Region | Test Year | |
---|---|---|---|---|
Scenario 1 | Danube | 2018, 2019, 2020, 2021 | Danube | 2022 |
Scenario 2 | East | 2018, 2019, 2020, 2021 | East | 2022 |
Scenario 3 | Danube | 2022 | East | 2022 |
Scenario 4 | Danube | 2018, 2019, 2020, 2021, 2022 | East | 2022 |
Algorithm | Hyperparameter | Type/Statistic | Frequency/Value |
---|---|---|---|
SVM | Kernel function | Gaussian Quadratic Cubic | 10 * 7 * 3 * |
Box constraint level 1 | Mean Min Max | 445.14 4.36 961.49 | |
Kernel scale | Mean Min Max | 13.02 5.26 23.83 | |
Neural Network | Fully connected layers | Layer 1 Layer 2 Layer 3 | 9 * 9 * 2 * |
Activation function 2 | Tanh Relu Sigmo | 12 * 6 * 2 * | |
First layer size | Mean Min Max | 154 10 298 | |
Second layer size | Mean Min Max | 64 11 176 | |
Third layer size | Mean Min Max | 23 16 29 | |
Regularization strength (Lambda) | Mean Min Max | 2.72 × 10−6 4.49 × 10−8 7.62 × 10−6 | |
RF | Number of learners | Mean Min Max | 360 32 500 |
Number of predictors to sample | Mean Min Max | 16 3 46 | |
Max. number of splits | Mean Min Max | 13,837 1104 39,619 |
Scenario 1 | Train Danube without 2018 | Train Danube without 2019 | Train Danube without 2020 | Train Danube without 2021 | Train Danube without 2022 | Mean |
---|---|---|---|---|---|---|
Test Danube 2018 | Test Danube 2019 | Test Danube 2020 | Test Danube 2021 | Test Danube 2022 | ||
QDA | 93.30 | 92.44 | 93.74 | 90.28 | 90.22 | 92.00 |
SVM | 91.76 | 93.88 | 94.76 | 92.22 | 91.53 | 92.83 |
NN | 91.57 | 92.22 | 94.14 | 92.01 | 91.39 | 92.27 |
RF | 92.26 | 87.61 | 91.56 | 86.64 | 82.50 | 88.11 |
Mean | 92.22 | 91.54 | 93.55 | 90.29 | 88.91 |
Scenario 2 | Train East without 2018 | Train East without 2019 | Train East without 2020 | Train East without 2021 | Train East without 2022 | Mean |
---|---|---|---|---|---|---|
Test East 2018 | Test East 2019 | Test East 2020 | Test East 2021 | Test East 2022 | ||
QDA | 87.70 | 91.10 | 86.00 | 86.60 | 85.00 | 87.28 |
SVM | 84.40 | 87.60 | 88.90 | 85.90 | 80.50 | 85.40 |
NN | 88.20 | 89.10 | 91.90 | 85.40 | 81.10 | 87.14 |
RF | 85.80 | 89.70 | 90.80 | 82.80 | 74.10 | 84.64 |
Mean | 86.53 | 89.38 | 89.40 | 85.18 | 80.18 |
Prediction | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|
SVM | Barley | Rapeseed | Maize | Wheat | Sugar Beet | Sunflower | Soybean | Grass | ∑ | |
Reference | Barley | 975 | 2 | 2 | 18 | 1 | 0 | 0 | 2 | 1000 |
Rapeseed | 21 | 977 | 0 | 1 | 1 | 0 | 0 | 0 | 1000 | |
Maize | 26 | 2 | 922 | 13 | 1 | 10 | 11 | 15 | 1000 | |
Wheat | 80 | 1 | 0 | 913 | 0 | 1 | 0 | 5 | 1000 | |
Sugar beet | 5 | 15 | 4 | 0 | 971 | 5 | 0 | 0 | 1000 | |
Sunflower | 8 | 0 | 4 | 0 | 1 | 967 | 8 | 12 | 1000 | |
Soybean | 16 | 0 | 193 | 18 | 1 | 160 | 603 | 9 | 1000 | |
Grass | 3 | 0 | 1 | 0 | 2 | 0 | 0 | 994 | 1000 | |
∑ | 1134 | 997 | 1126 | 963 | 978 | 1143 | 622 | 1037 | ||
OA | 91.5 | |||||||||
KIA | 0.90 |
Prediction | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|
SVM | Barley | Rapeseed | Maize | Wheat | Sugar Beet | Sunflower | Soybean | Grass | ∑ | |
Reference | Barley | 794 | 25 | 9 | 142 | 6 | 1 | 13 | 10 | 1000 |
Rapeseed | 0 | 995 | 0 | 0 | 4 | 1 | 0 | 0 | 1000 | |
Maize | 0 | 0 | 971 | 6 | 3 | 1 | 16 | 0 | 1000 | |
Wheat | 9 | 10 | 3 | 976 | 1 | 0 | 1 | 0 | 1000 | |
Sugar beet | 0 | 0 | 5 | 0 | 987 | 3 | 5 | 0 | 1000 | |
Sunflower | 4 | 0 | 4 | 0 | 18 | 949 | 23 | 2 | 1000 | |
Soybean | 2 | 0 | 25 | 0 | 6 | 13 | 952 | 2 | 1000 | |
Grass | 3 | 6 | 5 | 15 | 2 | 4 | 8 | 957 | 1000 | |
∑ | 812 | 1036 | 1022 | 1139 | 1027 | 972 | 1021 | 971 | ||
OA | 94.8 | |||||||||
KIA | 0.90 |
Scenario 3 | Train Danube 2018 | Train Danube 2019 | Train Danube 2020 | Train Danube 2021 | Train Danube 2022 | Mean |
---|---|---|---|---|---|---|
Test East 2018 | Test East 2019 | Test East 2020 | Test East 2021 | Test East 2022 | ||
QDA | 81.10 | 88.60 | 88.75 | 87.35 | 81.34 | 85.43 |
SVM | 84.90 | 88.66 | 90.30 | 87.41 | 85.29 | 87.31 |
NN | 82.16 | 91.54 | 90.27 | 87.34 | 83.80 | 87.02 |
RF | 76.90 | 89.29 | 89.15 | 80.65 | 72.10 | 81.62 |
Mean | 81.27 | 89.52 | 89.62 | 85.69 | 80.63 |
Scenario 4 | Train Danube 2018–2022 | Train Danube 2018–2022 | Train Danube 2018–2022 | Train Danube 2018–2022 | Train Danube 2018–2022 | Mean |
---|---|---|---|---|---|---|
Test East 2018 | Test East 2019 | Test East 2020 | Test East 2021 | Test East 2022 | ||
QDA | 85.70 | 91.40 | 89.50 | 84.00 | 84.30 | 86.98 |
SVM | 89.90 | 93.10 | 92.70 | 86.52 | 88.70 | 90.18 |
NN | 90.10 | 92.80 | 89.00 | 84.24 | 86.20 | 88.47 |
RF | 85.20 | 90.30 | 89.60 | 82.50 | 78.80 | 85.28 |
Mean | 87.73 | 91.90 | 90.20 | 84.32 | 84.50 |
Prediction | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|
SVM | Barley | Rapeseed | Maize | Wheat | Sugar Beet | Sunflower | Soybean | Grass | ∑ | |
Reference | Barley | 889 | 0 | 24 | 0 | 2 | 15 | 10 | 60 | 1000 |
Rapeseed | 16 | 956 | 2 | 1 | 0 | 17 | 4 | 4 | 1000 | |
Maize | 7 | 0 | 748 | 0 | 9 | 24 | 200 | 12 | 1000 | |
Wheat | 172 | 2 | 2 | 791 | 1 | 4 | 4 | 24 | 1000 | |
Sugar beet | 0 | 0 | 0 | 0 | 1000 | 0 | 0 | 0 | 1000 | |
Sunflower | 1 | 0 | 19 | 0 | 66 | 616 | 266 | 32 | 1000 | |
Soybean | 11 | 3 | 40 | 1 | 3 | 9 | 928 | 5 | 1000 | |
Grass | 0 | 0 | 0 | 0 | 0 | 0 | 6 | 994 | 1000 | |
∑ | 1096 | 961 | 835 | 793 | 1081 | 685 | 1418 | 1131 | ||
OA | 86.53 | |||||||||
KIA | 0.85 |
Prediction | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|
SVM | Barley | Rapeseed | Maize | Wheat | Sugar Beet | Sunflower | Soybean | Grass | ∑ | |
Reference | Barley | 832 | 2 | 7 | 2 | 3 | 1 | 2 | 0 | 1000 |
Rapeseed | 12 | 836 | 0 | 0 | 0 | 0 | 0 | 1 | 1000 | |
Maize | 4 | 2 | 735 | 0 | 41 | 26 | 38 | 3 | 1000 | |
Wheat | 11 | 2 | 0 | 832 | 0 | 0 | 1 | 3 | 1000 | |
Sugar beet | 0 | 0 | 0 | 0 | 849 | 0 | 0 | 0 | 1000 | |
Sunflower | 4 | 0 | 10 | 8 | 154 | 606 | 50 | 17 | 1000 | |
Soybean | 6 | 0 | 11 | 4 | 39 | 3 | 765 | 21 | 1000 | |
Grass | 1 | 0 | 2 | 0 | 2 | 0 | 0 | 844 | 1000 | |
∑ | 870 | 842 | 765 | 846 | 1088 | 636 | 856 | 889 | ||
OA | 92.74 | |||||||||
KIA | 0.92 |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2023 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://creativecommons.org/licenses/by/4.0/).
Share and Cite
Rusňák, T.; Kasanický, T.; Malík, P.; Mojžiš, J.; Zelenka, J.; Sviček, M.; Abrahám, D.; Halabuk, A. Crop Mapping without Labels: Investigating Temporal and Spatial Transferability of Crop Classification Models Using a 5-Year Sentinel-2 Series and Machine Learning. Remote Sens. 2023, 15, 3414. https://doi.org/10.3390/rs15133414
Rusňák T, Kasanický T, Malík P, Mojžiš J, Zelenka J, Sviček M, Abrahám D, Halabuk A. Crop Mapping without Labels: Investigating Temporal and Spatial Transferability of Crop Classification Models Using a 5-Year Sentinel-2 Series and Machine Learning. Remote Sensing. 2023; 15(13):3414. https://doi.org/10.3390/rs15133414
Chicago/Turabian StyleRusňák, Tomáš, Tomáš Kasanický, Peter Malík, Ján Mojžiš, Ján Zelenka, Michal Sviček, Dominik Abrahám, and Andrej Halabuk. 2023. "Crop Mapping without Labels: Investigating Temporal and Spatial Transferability of Crop Classification Models Using a 5-Year Sentinel-2 Series and Machine Learning" Remote Sensing 15, no. 13: 3414. https://doi.org/10.3390/rs15133414
APA StyleRusňák, T., Kasanický, T., Malík, P., Mojžiš, J., Zelenka, J., Sviček, M., Abrahám, D., & Halabuk, A. (2023). Crop Mapping without Labels: Investigating Temporal and Spatial Transferability of Crop Classification Models Using a 5-Year Sentinel-2 Series and Machine Learning. Remote Sensing, 15(13), 3414. https://doi.org/10.3390/rs15133414