Road intersection data have been used across a range of geospatial analyses. However, many datasets dating from before the advent of GIS are only available as historical printed maps. To be analyzed by GIS software, they need to be scanned and transformed into a usable (vector-based) format. Because the number of scanned historical maps is voluminous, automated methods of digitization and transformation are needed. Frequently, these processes are based on computer vision algorithms. However, the key challenges to this are (1) the low conversion accuracy for low quality and visually complex maps, and (2) the selection of optimal parameters. In this paper, we used a region-based deep convolutional neural network-based framework (RCNN) for object detection, in order to automatically identify road intersections in historical maps of several cities in the United States of America. We found that the RCNN approach is more accurate than traditional computer vision algorithms for double-line cartographic representation of the roads, though its accuracy does not surpass all traditional methods used for single-line symbols. The results suggest that the number of errors in the outputs is sensitive to complexity and blurriness of the maps, and to the number of distinct red-green-blue (RGB) combinations within them.

Machine learning (ML) is becoming an ever more important tool in hydrologic modeling. Previous studies have shown the higher prediction accuracy of those ML models over traditional process-based ones. However, there is another advantage of ML which is its lower computational demand. This is important for the applications such as hydraulic soil erosion estimation over a large area and at a finer spatial scale. Using traditional models like Rangeland Hydrology and Erosion Model (RHEM) requires too much computation time and resources. In this study, we designed an Artificial Neural Network that is able to recreate the RHEM outputs (annual average runoff, soil loss, and sediment yield and not the daily storm event-based values) with high accuracy (Nash-Sutcliffe Efficiency ≈ 1.0) and a very low computational time (13 billion times faster on average using a GPU). We ran the RHEM for more than a million synthetic scenarios and train the Emulator with them. We also, fine-tuned the trained Emulator with the RHEM runs of the real-world scenarios (more than 32,000) so the Emulator remains comprehensive while it works specifically accurately for the real-world cases. We also showed that the sensitivity of the Emulator to the input variables is similar to the RHEM and it can effectively capture the changes in the RHEM outputs when an input variable varies. Finally, the dynamic prediction behavior of the Emulator is statistically similar to the RHEM.