Table 2 Examples of application of classification methods towards SDGs using EO data
From: Monitoring sustainable development by means of earth observation data and machine learning: a review
SDGs | Field | Main finding | References |
|---|---|---|---|
SDG 2 (Zero Hunger) | Agriculture | Multi-temporal crop classification reduces the unfavourable effects of using single-date acquisition | [81] |
The proposed method performed similar to SVM and RF in the classification of crops with similar phenology | [57] | ||
Developed an efficient framework for multi-temporal crops classification | [82] | ||
SDG 6 (Clean Water and Sanitation) | Wetland | The developed framework for coastal plain wetlands classification had high accuracy. | [83] |
SDG 8 (Decent Work and Economic Growth) | Slavery | The approach was used to help to liberate slaves by mapping brick kilns. | [58] |
SDG 11 (Sustainable Cities and Communities) | Land use | The approach based on CNN achieved an accuracy of ≅ 98% for land use and land cover analysis | [84] |
The proposed approach confirmed its suitability for urban planning because it had a superior performance compared to the global one | [56] | ||
Living conditions | Deep learning demonstrated a high potential to map areas of deprived living conditions | [85] | |
Land cover | The multivariate time series algorithm showed high accuracy for rare land cover classes | [86] | |
SDG 13 (Climate Action) | Climate | The model based on decision trees, and used to classify local climate zones, achieved a good performance | [75] |
SDG 14 (Life Below Water) | Marine habitat | SVM and K-NN classifiers achieved an accuracy higher than 90% on mapping coastal marine habitat | [50] |
SDG 15 (Life on Land) | Land cover | The approach used allowed to differentiate the hyperspectral subclasses from the classes | [87] |
Forest | Sentinel-2 is considered a powerful source of data for forest monitoring and mapping | [52] | |
RF was the best method to predict and map the area and volume of eucalyptus | [88] |