Monitoring Land Cover Changes in Northwestern Iran Using Training Samples Migration Method

Document Type : Original Article

Authors

1 Ph.D. Candidate, Dep. of Remote Sensing and GIS, Faculty of Geography, University of Tehran, Iran

2 Associate Prof., Dep. of Remote Sensing and GIS, Faculty of Geography, University of Tehran, Iran

3 Head dep. of Exploration Technology, Helmholtz-Zentrum Dresden-Rossendorf (HZDR), Helmholtz Institute Freiberg for Resource Technology (HiF), Freiberg, Germany

4 Prof. of Dep. of Remote Sensing and GIS, Faculty of Geography, University of Tehran, Iran

Abstract

Introduction: Land cover maps are essential elements in geographical analysis and spatial planning. The accuracy and effectiveness of these maps rely on three factors: Satellite imagery, classification algorithms and training samples. The quality of the training dataset significantly impacts the accuracy of classification results. This study aims to generate reliable training samples using the training sample migration method to monitor land cover changes in northwestern Iran from 2002 to 2022.
Materials and Methods: The study area covers 7653 square kilometers in northwestern Iran, situated between 44°35′59′′ to 46°01′25′′ longitude and 38°38′46′′ to 38°47′48′′ latitude. Data utilized in this research include satellite images and ground truth data, specifically Landsat images. The research methodology comprises five main steps. Initially, satellite images were obtained, followed by pre-processing steps involving radiometric and geometric corrections. Subsequently, training samples were prepared using high-resolution satellite images (Google Earth images) and ground surveys. The third step involved training sample migration, where spectral similarity between training samples from reference and target years was assessed using two parameters: Euclidean distance (ED) and spectral angle distance (SAD). After determining a suitable threshold, migrated training samples were distinguished from non-migrated samples. Evaluation of the accuracy of migrated training samples was conducted using reference data derived from Google Earth. In the fourth step, classification of satellite images from different years was performed using the migrated training samples. Finally, the accuracy of the classified images was assessed through the calculation of a confusion matrix in the fifth step.
Results and Discussion: The results indicate that a threshold of 0.9 to 1.1 is optimal for distinguishing migrated training samples from non-migrated training samples across different years. It can be observed that there is an inverse relationship between the accuracy of migrated training samples and the percentage of migrated training samples, with an increase in the percentage leading to decreased accuracy. Evaluation of the accuracy of migrated training samples based on each parameter (SAD and ED) reveals that migrated training samples based on the SAD parameter exhibit higher accuracy than those based on the ED parameter. Furthermore, the use of migrated samples based on both parameters has resulted in a 10.45% increase in accuracy compared to using the ED parameter alone, and a 5% increase compared to using the SAD parameter alone. Analysis of the percentage of migrated training samples in different land cover classes demonstrates that, on average, 80.6% of water class training samples, 75.4% of bare land class samples, 71.2% of built-up class samples, 64.6% of grassland class samples, 60.2% of cropland class samples, and 54.4% of wetland class samples were migrated from the reference year (2022) to each of the target years (2002, 2007, 2012, and 2017). The accuracy assessment of migrated training samples in different land cover classes also reveals that the water, built-up, bare land, grassland, cropland, and wetland classes had the highest accuracy in the migrated training samples, in that order. Analysis of land cover changes between 2002 and 2022 indicates a decrease in the area of bare land, water, and wetland classes from 2002 to 2022, while the area of the built-up class has increased during this period. Additionally, the grassland and cropland classes did not exhibit a consistent trend of change during this period, with their trends differing in different years. However, overall, the area of these two classes increased in 2022 compared to 2002.
 Conclusion: Future studies should consider using other satellite images (including Sentinel-2) for migrating training samples to evaluate the impact of different spectral bands and satellite images on the migration process. Furthermore, investigating the effectiveness of the training sample migration method for migrating training samples of other land covers could be a potential research topic for future studies.

The study area is located in northwestern Iran with an area of 7653 square kilometers. The study area lies between 44°35′59′′ to 46°01′25′′ longitude and 38°38′46′′ to 38°47′48′′ latitude. This study used satellite images and ground truth data.

The research methodology consists of five main steps. The initial step involved obtaining satellite images and performing pre-processing steps (radiometric and geometric correction). In the second step, training samples were collected using high-resolution satellite images (Google Earth images) and ground surveys. The third step involved the migration of training samples. To do this, the spectral similarity of the training samples from the reference and target years was first calculated using two parameters: Euclidean distance (ED) and spectral angle distance (SAD). Then, adopting the specified threshold, the migrated training samples were separated from the non-migrated samples. Furthermore, the accuracy of the migrated training samples was evaluated using reference data prepared from the Google Earth. In the fourth step, using the migrated training samples, the classification of satellite images in different years was performed. Finally, using the indices obtained from the confusion matrix, the accuracy of the classified images was evaluated.

The results showed that the threshold of 0.9 to 1.1 is the optimal threshold for separating migrated training samples from non-migrated training samples in different years. We also found a reverse correlation between the accuracy of the migrated training samples and the percentage of the migrated training samples, with higher percentages resulting in lower accuracy.

The accuracy assessment of the migrated training samples based on each parameter (SAD and ED) showed that the migrated training samples using the SAD parameter have higher accuracy than the migrated training samples using the ED parameter. Moreover, employing the migrated samples considering both parameters has increased the accuracy by 10.45% compared to using the ED parameter to migrate the training samples, and by 5% compared to using the SAD parameter to migrate the training samples.

The analysis of the percentage of migrated training samples in different land cover classes showed that, on average, 80.6% of the training samples of the water class, 75.4% of the bare land, 71.2% of the built-up, 64.6% of the grassland, 60.2% of the cropland, and 54.4% of the wetland were migrated from the reference year (2022) to each of the target years (2002, 2007, 2012, and 2017). The accuracy assessment of the migrated training samples in different land cover classes also showed that the water, built-up, bare land, grassland, cropland, and wetland classes had the highest accuracy in the migrated training samples, in that order.

The classification of satellite images was performed using Landsat images between 2002 and 2022. Accordingly, satellite images were classified into six different land cover classes. The accuracy assessment results showed that the overall accuracy of the classified images in 2022, 2017, 2012, 2007, and 2002 was 94.95%, 91.93%, 90.74%, 89.45%, and 88.94%, respectively. The accuracy assessment of different land cover classes based on two parameters, producer accuracy and user accuracy, showed that the water class has the highest producer and user accuracy among different classes (98.2% and 99.34%, respectively in 2022). In contrast, the wetland class had the lowest producer and user accuracy (90.1% and 91.25%, respectively in 2022).

The analysis of land cover changes between 2002 and 2022 showed that the area of bare land, water, and wetland classes decreased from 2002 to 2022, while the area of built-up class increased during this period. Furthermore, the grassland and cropland classes did not exhibit a constant trend of change during this time period, and their trends varied by year. However, the area of these two classes increased in 2022 compared to 2002. The analysis of the changes in the area of built-up class throughout this period shows a significant increase in the area of this land cover, which has increased from 20.38 square kilometers in 2002 to 123.98 square kilometers in 2022.

It is suggested that in future studies, other satellite images, like Sentinel-2, be used to migrate training samples in order to evaluate the effect of different spectral bands and satellite images on the migration of training samples. In addition, investigating the effectiveness of the training sample migration method in migration training samples of other land covers can be one of the research topics in future studies.

Keywords

References

Arabacı, D. & Kuşçu Şimşek, Ç., 2023, Prediction of Climatic Changes Caused by Land Use Changes in Urban Area Using Artificial Neural Networks, Theoretical and Applied Climatology, 152(2), PP. 265-279, https://doi.org/10.1007/s00704-023-04386-4.
Breiman, L., 2001, Random Forests, Machine Learning, 45, PP. 5-32, https://doi.org/10.1023/A:1010933404324.
Calderón-Loor, M., Hadjikakou, M. & Bryan, B.A., 2021, High-Resolution wall-to-Wall Land-Cover Mapping and Land Change Assessment for Australia from 1985 to 2015, Remote Sensing of Environment, 252, PP. 112148-112163, https://doi.org/10.1016/j.rse.2020.112148.
Chaaban, F., El Khattabi, J. & Darwishe, H., 2022, Accuracy Assessment of ESA WorldCover 2020 and ESRI 2020 Land Cover Maps for a Region in Syria, Journal of Geovisualization and Spatial Analysis, 6(2), PP. 31-49, https://doi.org/10.1007/s41651-022-00126-w.
Chen, Q., Zhong, C., Jing, C., Li, Y., Cao, B. & Cheng, Q., 2021, Rapid Mapping and Annual Dynamic Evaluation of Quality of Urban Green Spaces on Google Earth Engine, ISPRS International Journal of Geo-Information, 10(10), PP. 670-699, https://doi.org/10.3390/ijgi10100670.
Congalton, R.G., Gu, J., Yadav, K., Thenkabail, P. & Ozdogan, M., 2014, Global Land Cover Mapping: A Review and Uncertainty Analysis, Remote Sensing, 6(12), PP. 12070-12093, https://doi.org/10.3390/rs61212070.
Ebrahimy, H., Mirbagheri, B., Matkan, A.A. & Azadbakht, M., 2021, Per-Pixel Land Cover Accuracy Prediction: A Random Forest-Based Method with Limited Reference Sample Data, ISPRS Journal of Photogrammetry and Remote Sensing, 172, PP. 17-27, https://doi.org/10.1016/j.isprsjprs.2020.11.024.
Feizizadeh, B., Lakes, T., Omarzadeh, D., Sharifi, A., Blaschke, T. & Karimzadeh, S., 2022, Scenario-Based Analysis of the Impacts of Lake Drying on Food Production in the Lake Urmia Basin of Northern Iran, Scientific reports, 12(1), PP. 6237-6253, https://doi.org/10.1038/s41598-022-10159-2.
Fonji, S.F. & Taff, G.N., 2014, Using Satellite Data to Monitor Land-Use Land-Cover Change in North-Eastern Latvia, Springerplus, 3, PP. 1-15, https://doi.org/10.1186/2193-1801-3-61.
Friedl, M.A., Sulla-Menashe, D., Tan, B., Schneider, A., Ramankutty, N., Sibley, A. & Huang, X., 2010, MODIS Collection 5 Global Land Cover: Algorithm Refinements and Characterization of New Datasets, Remote Sensing of Environment, 114(1), PP. 168-182, https://doi.org/10.1016/j.rse.2009.08.016.
Fritz, S., See, L., Perger, C., McCallum, I., Schill, C., Schepaschenko, D. et al., 2017, A Global Dataset of Crowdsourced Land Cover and Land Use Reference Data, Scientific Data, 4(1), PP. 1-8, https://doi.org/10.1038/sdata.2017.75.
Ghorbanian, A., Kakooei, M., Amani, M., Mahdavi, S., Mohammadzadeh, A. & Hasanlou, M., 2020, Improved Land Cover Map of Iran Using Sentinel Imagery within Google Earth Engine and a Novel Automatic Workflow for Land Cover Classification Using Migrated Training Samples, ISPRS Journal of Photogrammetry and Remote Sensing, 167, PP. 276-288, https://doi.org/10.1016/j.isprsjprs.2020.07.013.
Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D. & Moore, R., 2017, Google Earth Engine: Planetary-Scale Geospatial Analysis for Everyone, Remote Sensing of Environment, 202, PP. 18-27, https://doi.org/10.1016/j.rse.2017.06.031.
Huang, H., Wang, J., Liu, C., Liang, L., Li, C. & Gong, P., 2020, The Migration of Training Samples towards Dynamic Global Land Cover Mapping, ISPRS Journal of Photogrammetry and Remote Sensing, 161, PP. 27-36, https://doi.org/10.1016/j.isprsjprs.2020.01.010.
Li, C., Ma, Z., Wang, L., Yu, W., Tan, D., Gao, B. et al., 2021, Improving the Accuracy of Land Cover Mapping by Distributing Training Samples, Remote Sensing, 13(22), PP. 4594-4607, https://doi.org/10.3390/rs13224594.
Li, J., Wang, J., Zhang, J., Liu, C., He, S. & Liu, L., 2022, Growing-Season Vegetation Coverage Patterns and Driving Factors in the China-Myanmar Economic Corridor Based on Google Earth Engine and Geographic Detector, Ecological Indicators, 136, PP. 108620-108635, https://doi.org/10.1016/j.ecolind.2022.108620.
Lu, D., Mausel, P., Brondizio, E. & Moran, E., 2004, Change Detection Techniques, International Journal of Remote Sensing, 25(12), PP. 2365-2401, https://doi.org/10.1080/0143116031000139863.
Murray, N.J., Worthington, T.A., Bunting, P., Duce, S., Hagger, V., Lovelock, C.E. et al., 2022, High-Resolution Mapping of Losses and Gains of Earth’s Tidal Wetlands, Science, 376(6594), PP. 744-749, https://doi.org/10.1126/science.abm9583.
Naboureh, A., Li, A., Ebrahimy, H., Bian, J., Azadbakht, M., Amani, M. et al., 2021, Assessing the Effects of Irrigated Agricultural Expansions on Lake Urmia Using Multi-Decadal Landsat Imagery and a Sample Migration Technique within Google Earth Engine, International Journal of Applied Earth Observation and Geoinformation, 105, PP. 102607-102618, https://doi.org/10.1016/j.jag.2021.102607.
Nedd, R., Light, K., Owens, M., James, N., Johnson, E. & Anandhi, A., 2021, A Synthesis of Land Use/Land Cover Studies: Definitions, Classification Systems, Meta-Studies, Challenges and Knowledge Gaps on a Global Landscape, Land, 10(9), PP. 994-1024, https://doi.org/10.3390/land10090994.
Nian, Y., He, Z., Zhang, W. & Chen, L., 2023, Land Cover Changes of the Qilian Mountain National Park in Northwest China Based on Phenological Features and Sample Migration from 1990 to 2020, Remote Sensing, 15(4), PP. 1074-1097, https://doi.org/10.3390/rs15041074.
Olofsson, P., Foody, G.M., Herold, M., Stehman, S.V., Woodcock, C.E. & Wulder, M.A., 2014, Good Practices for Estimating Area and Assessing Accuracy of Land Change, Remote Sensing of Environment, 148, PP. 42-57, https://doi.org/10.1016/j.rse.2014.02.015.
Padial-Iglesias, M., Serra, P., Ninyerola, M. & Pons, X., 2021, A Framework of Filtering Rules over Ground Truth Samples to Achieve Higher Accuracy in Land Cover Maps, Remote Sensing, 13(14), PP. 2662-2698, https://doi.org/10.3390/rs13142662.
Patra, S., Ghosh, S. & Ghosh, A., 2011, Histogram Thresholding for Unsupervised Change Detection of Remote Sensing Images, International Journal of Remote Sensing, 32(21), PP. 6071-6089, https://doi.org/10.1080/01431161.2010.507793.
Phan, D.C., Trung, T.H., Truong, V.T., Sasagawa, T., Vu, T.P.T., Bui, D.T. et al., 2021, First Comprehensive Quantification of Annual Land Use/Cover from 1990 to 2020 across Mainland Vietnam, Scientific Reports, 11(1), PP. 1-20, https://doi.org/10.1038/s41598-021-89034-5.
Radoux, J., Lamarche, C., Van Bogaert, E., Bontemps, S., Brockmann, C. & Defourny, P., 2014, Automated Training Sample Extraction for Global Land Cover Mapping, Remote Sensing, 6(5), PP. 3965-3987, https://doi.org/10.3390/rs6053965.
Rodriguez-Galiano, V.F., Ghimire, B., Rogan, J., Chica-Olmo, M. & Rigol-Sanchez, J.P., 2012, An Assessment of the Effectiveness of a Random Forest Classifier for Land-Cover Classification, ISPRS Journal of Photogrammetry and Remote Sensing, 67, PP. 93-104, https://doi.org/10.1016/j.isprsjprs.2011.11.002.
Rujoiu-Mare, M.R. & Mihai, B.A., 2016, Mapping Land Cover Using Remote Sensing Data and GIS Techniques: A Case Study of Prahova Subcarpathians, Procedia Environmental Sciences, 32, PP. 244-255, https://doi.org/10.1016/j.proenv.2016.03.029.
Som-ard, J., Immitzer, M., Vuolo, F., Ninsawat, S. & Atzberger, C., 2022, Mapping of Crop Types in 1989, 1999, 2009 and 2019 to Assess Major Land Cover Trends of the Udon Thani Province, Thailand, Computers and Electronics in Agriculture, 198, PP. 107083-107098, https://doi.org/10.1016/j.compag.2022.107083.
Steinhausen, M.J., Wagner, P.D., Narasimhan, B. & Waske, B., 2018, Combining Sentinel-1 and Sentinel-2 Data for Improved Land Use and Land Cover Mapping of Monsoon Regions, International Journal of Applied Earth Observation and Geoinformation, 73, PP. 595-604, https://doi.org/10.1016/j.jag.2018.08.011.
Strahler, A.H., Boschetti, L., Foody, G.M., Friedl, M.A., Hansen, M.C., Herold, M. et al., 2006, Global Land Cover Validation: Recommendations for Evaluation and Accuracy Assessment of Global Land Cover Maps, European Communities, 51(4), PP. 1-60.
Varga, K., Szabó, S., Szabó, G., Dévai, G. & Tóthmérész, B., 2014, Improved Land Cover Mapping Using Aerial Photographs and Satellite Images, Open Geosciences, 7(1), PP. 15-26, https://doi.org/10.1515/geo-2015-0002.
Wang, H., Liu, Y., Wang, Y., Yao, Y. & Wang, C., 2023, Land Cover Change in Global Drylands: A Review, Science of The Total Environment, 863, PP. 160943-160956, https://doi.org/10.1016/j.scitotenv.2022.160943.
Yan, X. & Niu, Z., 2021, Reliability Evaluation and Migration of Wetland Samples, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 14, PP. 8089-8099, https://doi.org/10.1109/jstars.2021.3102866.
Yan, X., Li, J., Smith, A.R., Yang, D., Ma, T. & Su, Y., 2023, Rapid Land Cover Classification Using a 36-Year Time Series of Multi-Source Remote Sensing Data, Land, 12, PP. 2149-2163, https://doi.org/10.3390/land12122149.
Yang, C. & Everitt, J.H., 2012, Using Spectral Distance, Spectral Angle and Plant Abundance Derived from Hyperspectral Imagery to Characterize Crop Yield Variation, Precision Agriculture, 13, PP. 62-75, https://doi.org/10.1007/s11119-011-9248-z.
Yu, L., Du, Z., Dong, R., Zheng, J., Tu, Y., Chen, X. et al., 2022, FROM-GLC Plus: Toward Near Real-Time and Multi-Resolution Land Cover Mapping, GIScience & Remote Sensing, 59(1), PP. 1026-1047, https://doi.org/10.1080/15481603.2022.2096184.
Zhu, Z., 2017, Change Detection Using Landsat Time Series: A Review of Frequencies, Preprocessing, Algorithms, and Applications, ISPRS Journal of Photogrammetry and Remote Sensing, 130, PP. 370-384, https://doi.org/10.1016/j.isprsjprs.2017.06.013.
Zhu, Q., Wang, Y., Liu, J., Li, X., Pan, H. & Jia, M., 2021, Tracking Historical Wetland Changes in the China Side of the Amur River Basin Based on Landsat Imagery and Training Samples Migration, Remote Sensing, 13(11), PP. 2161-2176, https://doi.org/10.3390/rs13112161.
 

Files

Share

How to cite

Statistics