The Loess positive and negative terrains (P-N terrains), which are widely distributed on the Loess Plateau, are discussed for the first time by introducing its characteristic, demarcation as well as extraction method from high-resolution Digital Elevation Models. Using 5 m-resolution DEMs as original test data, P-N terrains of 48 geomorphological units in different parts of Shaanxi Loess Plateau are extracted accurately. Then six indicators for depicting the geomorphologic landscape and spatial configuration characteristic of P-N terrains are proposed. The spatial distribution rules of these indicators and the relationship between the P-N terrains and Loess relief are discussed for further understanding of Loess landforms. Finally, with the integration of P-N terrains and traditional terrain indices, a series of un-supervised classification methods are applied to make a proper landform classification in northern Shaanxi. Results show that P-N terrains are an effect clue to reveal energy and substance distribution rules on the Loess Plateau. A continuous change of P-N terrains from south to north in Shaanxi Loess Plateau shows an obvious spatial difference of Loess landforms and the positive terrain area only accounted for 60.5% in this region. The P-N terrains participant landform classification method increases validity of the result, especially in the Loess tableland, Loess tableland-ridge and the Loess low-hill area. This research is significant on the study of Loess landforms with the Digital Terrains Analysis methods.

No abstract

Resolution has been playing a significant role in landslide susceptibility mapping and hazard assessment. Based on geographical information system (GIS) and information model, the effects of raster resolution on landslide susceptibility mapping are studied in a central area of Shenzhen, China. Eight factors are selected to calculate landslide susceptibility with eleven groups of different resolutions (5 to 190 m). It has been found that a finer resolution does not necessarily lead to a higher accuracy of landslide susceptibility mappings, while the result of 90 m-resolution has the best accuracy and the 150 m-resolution has the worst one. The accuracy curve is in a shape of "W" along with resolution decreasing: 1) The accuracy decreases from 5 to 70 m; 2) and then the best accuracy appears at 90 m, which is almost the same as the mean size of landslides in study area; 3) the accuracy decreases again from 110 to 150 m; 4) and finally the accuracy increases from 150 to 190 m. The sensitivity analysis indicates that the effects of raster resolution are mainly caused by the resolution impact on landform parameter derivation, while factors like geology and human activity are very insensitive to resolutions. A further study shows that in flat, ridge, and slope foot terrains, the susceptibility mapping result is sensitive to resolution, but in the sloping surface area the sensitivity is much less sensitive to resolution. At last, by choosing study areas with different sizes, it has also been found that the optimal resolutions are variable due to size of study area. But the study area is larger than a threshold, which is 135 km 2 in this study, and the optimal resolution is almost fixed. resolution sensitivity, landslide, susceptibility mapping, digital elevation model (DEM), information model Landslide susceptibility mapping is very important in landslide study and risk management engineering. In order to depict the spatial distribution of landslide susceptibility, many influential factors, as well as a properly selected mathematical or statistical model are needed. This kind of

This paper presents the scientific outcomes of the 2020 Data Fusion Contest organized by the Image Analysis and Data Fusion Technical Committee of the IEEE Geoscience and Remote Sensing Society. The 2020 Contest addressed the problem of automatic global land-cover mapping with weak supervision, i.e. estimating high-resolution semantic maps while only low-resolution reference data is available during training. Two separate competitions were organized to assess two different scenarios: 1) high-resolution labels are not available at all and 2) a small amount of high-resolution labels are available additionally to low-resolution reference data. In this paper we describe the DFC2020 dataset that remains available for further evaluation of corresponding approaches and report the results of the bestperforming methods during the contest. Index Terms-Image analysis and data fusion, multimodal, land cover mapping, weak supervision, deep learning, convolutional neural networks, random forests.