Urban land cover and land use mapping plays an important role in urban planning and management. In this paper, novel multi-scale deep learning models, namely ASPP-Unet and ResASPP-Unet are proposed for urban land cover classification based on very high resolution (VHR) satellite imagery. The proposed ASPP-Unet model consists of a contracting path which extracts the high-level features, and an expansive path, which up-samples the features to create a high-resolution output. The atrous spatial pyramid pooling (ASPP) technique is utilized in the bottom layer in order to incorporate multi-scale deep features into a discriminative feature. The ResASPP-Unet model further improves the architecture by replacing each layer with residual unit. The models were trained and tested based on WorldView-2 (WV2) and WorldView-3 (WV3) imageries over the city of Beijing. Model parameters including layer depth and the number of initial feature maps (IFMs) as well as the input image bands were evaluated in terms of their impact on the model performances. It is shown that the ResASPP-Unet model with 11 layers and 64 IFMs based on 8-band WV2 imagery produced the highest classification accuracy (87.1% for WV2 imagery and 84.0% for WV3 imagery). The ASPP-Unet model with the same parameter setting produced slightly lower accuracy, with overall accuracy of 85.2% for WV2 imagery and 83.2% for WV3 imagery. Overall, the proposed models outperformed the state-of-the-art models, e.g., U-Net, convolutional neural network (CNN) and Support Vector Machine (SVM) model over both WV2 and WV3 images, and yielded robust and efficient urban land cover classification results.

This study presented a MODIS 8-day 1 km evapotranspiration (ET) downscaling method based on Landsat 8 data (30 m) and machine learning approaches. Eleven indicators including albedo, land surface temperature (LST), and vegetation indices (VIs) derived from Landsat 8 data were first upscaled to 1 km resolution. Machine learning algorithms including Support Vector Regression (SVR), Cubist, and Random Forest (RF) were used to model the relationship between the Landsat indicators and MODIS 8-day 1 km ET. The models were then used to predict 30 m ET based on Landsat 8 indicators. A total of thirty-two pairs of Landsat 8 images/MODIS ET data were evaluated at four study sites including two in United States and two in South Korea. Among the three models, RF produced the lowest error, with relative Root Mean Square Error (rRMSE) less than 20%. Vegetation greenness related indicators such as Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Soil Adjusted Vegetation Index (SAVI), and vegetation moisture related indicators such as Normalized Difference Infrared Index—Landsat 8 OLI band 7 (NDIIb7) and Normalized Difference Water Index (NDWI) were the five most important features used in RF model. Temperature-based indicators were less important than vegetation greenness and moisture-related indicators because LST could have considerable variation during each 8-day period. The predicted Landsat downscaled ET had good overall agreement with MODIS ET (average rRMSE = 22%) and showed a similar temporal trend as MODIS ET. Compared to the MODIS ET product, the downscaled product demonstrated more spatial details, and had better agreement with in situ ET observations (R² = 0.56). However, we found that the accuracy of MODIS ET was the main control factor of the accuracy of the downscaled product. Improved coarse-resolution ET estimation would result in better finer-resolution estimation. This study proved the potential of using machine learning approaches for ET downscaling considering their effectiveness and ease of implementation. Future research includes development of the spatial-temporal fusion models of Landsat data and MODIS ET in order to increase temporal resolution of downscaled ET.

Urban tree species mapping is an important prerequisite to understanding the value of urban vegetation in ecological services. In this study, we explored the potential of bi-temporal WorldView-2 (WV2, acquired on 14 September 2012) and WorldView-3 images (WV3, acquired on 18 October 2014) for identifying five dominant urban tree species with the object-based Support Vector Machine (SVM) and Random Forest (RF) methods. Two study areas in Beijing, China, Capital Normal University (CNU) and Beijing Normal University (BNU), representing the typical urban environment, were evaluated. Three classification schemes—classification based solely on WV2; WV3; and bi-temporal WV2 and WV3 images—were examined. Our study showed that the single-date image did not produce satisfying classification results as both producer and user accuracies of tree species were relatively low (44.7%–82.5%), whereas those derived from bi-temporal images were on average 10.7% higher. In addition, the overall accuracy increased substantially (9.7%–20.2% for the CNU area and 4.7%–12% for BNU). A thorough analysis concluded that near-infrared 2, red-edge and green bands are always more important than the other bands to classification, and spectral features always contribute more than textural features. Our results also showed that the scattered distribution of trees and a more complex surrounding environment reduced classification accuracy. Comparisons between SVM and RF classifiers suggested that SVM is more effective for urban tree species classification as it outperforms RF when working with a smaller amount and imbalanced distribution of samples.

Land subsidence induced by excessive groundwater withdrawal has caused serious social, geological, and environmental problems in Beijing. Rapid increases in population and economic development have aggravated the situation. Monitoring and prediction of ground settlement is important to mitigate these hazards. In this study, we combined persistent-scatterer interferometric synthetic aperture radar with Grey system theory to monitor and predict land subsidence in the Beijing plain. Land subsidence during 2003–2014 was determined based on 39 ENVISAT advanced synthetic aperture radar (ASAR) images and 27 RadarSat-2 images. Results were consistent with global positioning system, leveling measurements at the point level and TerraSAR-X subsidence maps at the regional level. The average deformation rate in the line-of-sight was from −124 to 7 mm/year. To predict future subsidence, the time-series deformation was used to build a prediction model based on an improved Grey-Markov model (IGMM), which adapted the conventional GM(1,1) model by utilizing rolling mechanism and integrating a k-means clustering method in Markov-chain state interval partitioning. Evaluation of the IGMM at both point level and regional scale showed good accuracy (root-mean-square error <3 mm; R² = 0.94 and 0.91). Finally, land subsidence in 2015–2016 was predicted, and the maximum cumulative deformation will reach 1717 mm by the end of 2016. The promising results indicate that this method can be used as an alternative to the conventional numerical and empirical models for short-term prediction when there is lack of detailed geological or hydraulic information.

Accurate paddy rice mapping with fine spatial detail is significant for ensuring food security and maintaining sustainable environmental development. In northeastern China, rice is planted in fragmented and patchy fields and its production has reached over 10% of the total amount of rice production in China, which has brought the increasing need for updated paddy rice maps in the region. Existing methods for mapping paddy rice are often based on remote sensing techniques by using optical images. However, it is difficult to obtain high quality time series remote sensing data due to the frequent cloud cover in rice planting area and low temporal sampling frequency of satellite imagery. Therefore, paddy rice maps are often developed using few Landsat or time series MODIS images, which has limited the accuracy of paddy rice mapping. To overcome these limitations, we presented a new strategy by integrating a spatiotemporal fusion algorithm and phenology-based algorithm to map paddy rice fields. First, we applied the spatial and temporal adaptive reflectance fusion model (STARFM) to fuse the Landsat and MODIS data and obtain multi-temporal Landsat-like images. From the fused Landsat-like images and the original Landsat images, we derived time series vegetation indices (VIs) with high temporal and high spatial resolution. Then, the phenology-based algorithm, considering the unique physical features of paddy rice during the flooding and transplanting phases/open-canopy period, was used to map paddy rice fields. In order to prove the effectiveness of the proposed strategy, we compared our results with those from other three classification strategies: (1) phenology-based classification based on original Landsat images only, (2) phenology-based classification based on original MODIS images only and (3) random forest (RF) classification based on both Landsat and Landsat-like images. The validation experiments indicate that our fusion-and phenology-based strategy could improve the overall accuracy of classification by 6.07% (from 92.12% to 98.19%) compared to using Landsat data only, and 8.96% (from 89.23% to 98.19%) compared to using MODIS data, and 4.66% (from93.53% to 98.19%) compared to using the RF algorithm. The results show that our new strategy, by integrating the spatiotemporal fusion algorithm and phenology-based algorithm, can provide an effective and robust approach to map paddy rice fields in regions with limited available images, as well as the areas with patchy and fragmented fields.