ABSTRACT
Is the production of national, continental, or global landmass fine-scale mapping using high-resolution remotely sensed imageries feasible with today’s remote sensing technology? In particular, is fine-scale large-area mapping of built-up areas feasible? Even if these questions may be considered trivial and already demonstrated for data scenarios including low-and moderate-resolution images, they are still far from being solved if applied to high-and very-high-resolution (HR/VHR) optical images. The reasons
8.1 Introduction .................................................................................................. 143 8.2 Rationale ....................................................................................................... 144 8.3 On the Complexity of HR/VHR Human Settlement Description ................ 147
8.3.1 Data Complexity ............................................................................... 148 8.3.2 Thematic Complexity ....................................................................... 148
8.4 Possible Solving Strategy.............................................................................. 150 8.4.1 Features ............................................................................................. 151 8.4.2 Data Representation .......................................................................... 152 8.4.3 Classification Schema ....................................................................... 153
8.5 GHSL Processing Workflow ......................................................................... 155 8.5.1 Input Image Data Available .............................................................. 155 8.5.2 Global Reference Data ...................................................................... 156 8.5.3 General Workflow ............................................................................. 156 8.5.4 Learning Approach ........................................................................... 158 8.5.5 Validation Strategy ........................................................................... 159
8.6 Description of the First GHSL Results ......................................................... 159 8.7 Conclusions ................................................................................................... 165 References .............................................................................................................. 166
behind this are linked to the specific characteristics of the input data scenarios including HR/VHR data and to the specific characteristics of the physical targets associated with the human settlement information to be recognized and analyzed. Quadratic increase of input data volume, exponential increase of computational complexity due to the necessity to process multiscale structural (shape/size, morphological, and textural) image descriptors and the necessity of spatial and thematic uncertainty management, and increase of thematic complexity and automatic learning strategies are some of the challenges that must be addressed in order to solve the two questions regarding the class of HR/VHR input image data. These two main questions were explored during the first operational test made by the Joint Research Center (JRC) during 2012 within the framework of the Global Human Settlement Layer (GHSL) production. The extent of the test area was 24.3 millions of square kilometers and covered parts of four continents. The imagery was collected by a variety of optical satellite and airborne sensors with spatial resolution ranging from 0.5 to 10 m (Pesaresi et al. 2013). It is the largest known automatic image classification involving this kind of image input (Figure 8.1).
