International Journal of Remote Sensing · 2011 · 102 citations · 29 references
AgricultureCrop SciencesEarth SciencesCrop SensingCropping SystemsRemote SensingCrop ClassificationLand Use Change DetectionMulti-temporal Image AnalysisRemote Sensing Change DetectionVegetation IndicesClassification Accuracy AssessmentCrop MappingNormalized Difference Vegetation Index MappingRemote Sensing Crop ClassificationSatellite Image Time Series AnalysisSegment-based FeaturesTemporal AggregationTemporal Segmentation
Crop cover and crop rotation mapping is an important and still evolving field in remote sensing science for which robust and highly automated processing chains are required. This study presents an improved mapping procedure for crop rotations of irrigated areas in Central Asia by using classification and regression trees (CARTs) applied to transformations of 250 m Moderate Resolution Imaging Spectroradiometer (MODIS) Normalized Difference Vegetation Index (NDVI) time series. The time series were divided into several temporal segments, from which metrics were derived as input features for classification. This temporal aggregation was applied to suppress within-class temporal variability. Various lengths of temporal segments were tested for their potential to increase classification accuracy. In addition, tests of enhancing the classification accuracy were done by combining different classification results using the majority rule for voting. These different processing strategies were applied to four annual time series (2004–2007) of the Khorezm region, where 270 000 ha of irrigated land is dominated by rotations of cotton, wheat and rice. Improved classification results were obtained for CARTs applied to metrics derived from a mixture of different segment lengths. The sole use of either long or short temporal segments was inferior. CART prioritized segments representing active phases of the phenological development. The best result, the optimized segment-based approach, achieved an overall accuracy between 83 and 85% for classifications between 2004 and 2007; in particular, the small range demonstrated the robustness regarding inter-annual variations. These accuracies exceeded those of the original time series without temporal segmentation by 6–7%. With some adjustments to other crops and field heterogeneity influencing the usefulness of a respective sensor, the approach can be applied to other irrigation systems in Central Asia.
29
Classification and Regression Trees.
Alexander Gordon, Leo Breiman, Jerome H. Friedman et al. · Biometrics · 1984 · 23.9K citations
Classification and Regression Trees.
John Van Ryzin, Leo Breiman, Jerome H. Friedman et al. · Journal of the American Statistical Association · 1986 · 21K citations
Global land cover mapping from MODIS: algorithms and early results
M. A. Friedl, D.K. McIver, J.C.F. Hodges et al. · Remote Sensing of Environment · 2002 · 2.8K citations