Low-cost air quality sensors can fill gaps between the sparse measurements done with high-quality national monitoring grids and might contribute to creating a more complete understanding of air pollution in an urban area. However, until there is no agreement on what degree of sensor accuracy is acceptable, the sensor data quality should be validated before governmental bodies use it as input for decision-making (Lewis2017).

This research proposes a method to assess and improve the data quality of low-cost air quality sensors measuring Particulate Matter (PM). To answer the research question "How can accuracy and precision of Particulate Matter measurement results from a low-cost outdoor sensor network be improved by using a correction model, using data from reference sensors and additional sensors measuring inferencing phenomena?" an experiment setup with sensors operating under real-world conditions is applied.

Two low-cost sensor nodes, both containing a microcontroller, two low-cost PM sensors, and a temperature and humidity sensor, are placed at two locations in the city of Rotterdam. At those two locations, they are placed next to a high-quality air quality monitoring station from the environmental agency of Rotterdam. These monitoring stations provide benchmark data for the low-cost sensor nodes. A third data source provides data on air pressure and wind speed for the whole city of Rotterdam.

The data that originates from both sensor nodes and monitoring stations are matched and correlated with each other. Subsequently, the measurements from the low-cost sensor nodes are evaluated. Correlations and cross inferences of PM with other independent variables such as humidity, ambient temperature, wind speed and air pressure are investigated. Thereafter, utilizing the Stepwise Multiple Linear Regression method, various correction models are created that take various combinations of external variables into account. The correction models vary with respect to the amount of included external environmental variables and the polynomial degree. From all those possible correction models, the best correction model per location is selected by evaluating the Root Mean Square Error (RMSE) of the corrected dataset.

Consequently, the results of the chosen correction model are validated. It is found that the best performing correction models are those that include only the original PM data and the effect of adding more independent variables is limited. The best correction models for the four low-cost PM sensors are able to decrease the RMSE of the observations: the original normalized RMSE ranged from 0.0918 to 0.1249, while the corrected normalized RMSE range from 0.03110 to 0.03759. So, it is possible to improve the data quality of low-cost PM sensors with the stepwise MLR method and setup as shown in this research. However, including parameters for independent variables humidity, temperature, air pressure or wind speed does not improve the data quality significantly.

Besides, when an extra sensor node is placed in an air quality monitoring network as described in this research, it is necessary to create a correction model for that specific sensor. Like Castell (2017) and Mukherjee (2017) also found, it is necessary to calibrate each individual low-cost sensor before adding it to an air quality measuring network of the type as described in this research. Namely, it is found that for each low-cost PM sensor in the network different correction models are created.

Landslides are destructive and recurrent natural disasters that cost annually significant social and economic losses all over the world. These events can be induced by natural factors as earthquakes and extreme rainfall, as well as by human intervention, including construction and mining. A primary resource to conduct landslides studies for prediction, risk assessment, and mitigation are historical databases with accurate location of individual events. To increase the location accuracy of those past landslide events, and optimize conventional time- and cost- consuming mapping routines, this study aims to develop an automatic landslide detection method from free-of-charge optical satellite imagery (Sentinel-2) and global Digital Elevation Model (ALOS World3D-30m DEM) using Object-based Image Analysis (OBIA) in combination with Machine Learning (ML). Existing works have successfully used earth-observation datasets for the generation of landslides databases. Most of them apply rule-based techniques using features thresholds that are not global and therefore perform poorly when applied to new regions where the method was not developed. This study presents a first attempt of an automatic method that generalizes to landslides occurring over the entire world without knowledge of their cause or triggering factor. To obtain a robust method that can deal with the complex characteristics of landslides (e.g. diversity of shapes/sizes, land cover, illumination and spectral variability), we explored OBIA, an image processing technique that has demonstrated better performance than the pixel-based approach, specially when the target objects are bigger than the cell resolution. The developed method consists in cloud-free images acquisition and determination of suitable features for image segmentation and image classification. For the image segmentation, we developed a two-step approach that consists in an initial segmentation using k-means and the Red/Green Difference (RGD) as input feature to create homogeneous segments and isolate landslides from non-landslides. This first approach leads to oversegmentation of non-landslide areas and, consequently, to an imbalanced dataset. The second step consists in a merging algorithm using Normalized Difference Vegetation Index (NDVI) as input feature to merge homogeneous non-landslide segments and balance the dataset. These two-stages include the setting of parameters as the number of clusters (K) and NDVI thresholds that were experimentally derived. Once the segments are created and the dataset is balanced, a non-parametric supervised classification using Random Forest (RF) was applied to identify landslide segments; the main advantage of this classifier is that it can deal with different statistical distributions of features and can handle imbalanced datasets. Using a training and testing set of 70% and 30%, our method achieved a precision of 83%, recall of 83%, and f1-score of 83%. We found that topographic features have less influence than spectral ones; however, their exclusion decreases the model performance in about 10%. Our method is built using entirely open source technologies allowing its applicability and re-usability. For future work, we propose to use our method to detect new landslides and increase the number of training samples. Additionally, we recommend to explore a complementary approach to the merging algorithm to reduce the number of non-landslide segments, balance the dataset, and keep accurate classification results while more training images are added to the model.

In the western part of the Netherlands, the soil contains mainly sand, peat, and clay which are known as soft soil layers. The buildings and infrastructures, such as roads, constructed on these soil layers are usually associated with substantial construction measures during the execution of the project and might suffer from damages induced by the post-construction deformations. In practice, one of the primary stages of road construction involves geotechnical in-situ investigations for determining the soil properties based on which the settlement is predicted through empirical models. There are several techniques for monitoring the post-construction deformation on roads, among which the most time and cost-efficient technique is advanced Differential InSAR (D-InSAR). Since no research has been dedicated to establishing a direct link between the geotechnical in situ measurements and deformation measurements, in this research, the main focus is to develop a fully data-driven methodology to model road deformation based on loading/unloading conditions and soil properties. The study area is the newly constructed part of the A4 highway (Delft-Schiedam) in the Netherlands.
The proposed methodology in this research consists of three steps. In the first step of the methodology, the measurements that represent soil properties, loading/unloading and deformation measurements should be determined and gathered. Cone Penetration Testing (CPT) measurements and boreholes are two freely available data sets that represent soil properties. Another important soil property is the variations in soil water content can be characterized by temperature and precipitation. The latest stage of loading/unloading history can be determined by comparing the elevation of the study area before and after the construction. Deformation time series produced by D-InSAR techniques are suitable measurements for investigating spatiotemporal deformations on roads. After determining pre-processing steps for each of the raw data sets, the relevant parameters from each data source are extracted. In the next step, the correlations and similarities between the soil properties, loading/unloading condition, and deformation are investigated. The last step deals with extracting suitable features from CPT profiles in order to use machine learning to model the relationship between soil properties, loading/unloading conditions, and deformation. To this end, the CPT profiles are segmented, then qualitative (soil types) or quantitative descriptors of the segments are used as features. To determine the soil classes, Support Vector Machines (SVM) classifier is used. The relationship between soil properties, loading/unloading and the linear rate of deformation is modeled through two tree-based algorithms, i.e. Random Forests and Gradient tree-boosting. The Pearson correlation and the coefficient of determination between soil properties, loading/unloading and the linear rate of deformation are 0.6 and 0.4, respectively. The correlation of deformation time series and temperature and precipitation is quite low and no consistent pattern could be found between the time delays. The soil classification by SVM classifier is more accurate compared to empirical charts. For the deformation modeling, the best performance metrics are obtained through the Gradient Boosting algorithm with quantitative descriptors as features, (Mean Absolute Error (MAE) is 1.1 mm/year, Root Mean Squared Error (RMSE) is 1.5 mm/year and the coefficient of determination is 0.5). In conclusion, the resulting models with different algorithms and different sets of features are of moderate accuracy. The uncertainty of the models is due to three main reasons: 1. The complexity of the study area in terms of construction history 2. Lack of other necessary data 3. The uncertainties caused by the proposed methodology.

There is a paradigm shift from two- to three-dimensional data, from maps to information dense models. Self-driving cars, digitization of historic buildings or maintenance of highway infrastructure are a small selection of many applications that use laser scanning to acquire three-dimensional data of our physical surroundings. Most of these applications require more than the shape of their surroundings. For example, a self-driving car needs to identify pedestrians, road signs and traffic lights in order to navigate safely. Therefore the point cloud acquired by laser scanning needs to be enriched with additional information. Automatic assignment of the object type a point belongs to is called classification.

This research focuses on deep learning for point cloud classification, because it revolutionized classification of imagery. PointNet is used to enable deep learning directly on point cloud data sets. To date PointNet is proven for indoor point clouds, this research explores the application of PointNet on an outdoor highway scene. The methodology creates point cloud training data efficiently by reusing known 2D object locations. Different spatial representations and sampling methods for the points are tested.

On average the method classifies 50% of the points correctly in four object classes. In combination with clustering of the point-wise predictions, the method predicts 60% of the 2D object locations successfully. The performance is comparable to the 47% average class accuracy PointNet achieves for 13 classes on the indoor data set.

Every day, terabytes of information is generated, filling storage devices around the world. However,the human brain have limited capacities to read and understand raw data from a computer screen.That is why data specialists need to ingeniously create better ways to display, process and analyzemassive amounts of data.Our research project is not about avoiding subsidence, not even about cracks on buildings; it ispurely data analysis and interpretation. This study will help professionals understand and fightagainst building subsidence. Our task was to create, manipulate and make sense of charts like theone below (a real line graph from InSAR data), then translate them into useful information forstakeholders in the local, national and global community.The aim of the project was to understand if ground sensor technologies are comparable to othersources of information. In our analysis different strategies to analyze building subsidence wereimplemented, e.g. homogeneous subsidence, heterogeneous subsidence and for water levels,interpolation and cross correlation methods. In addition, other techniques like sensor fusing wereimplemented to compare data from different sources.As a result from all these strategies, we can say that the water level sensors placed in our researchbuilding, have a high similarity with citizens and municipality data. In contrast, InSAR data is notcomparable with the subsidence sensors placed in the building because they have differentreferences and the period of study was too short to get accurate results from satellite data. Finally,an idea for future implementation strategies was proposed. On this idea, measurements of levelscan be carried out taking as a reference the NAP level and comparing the subsidence between ahealthy-foundations building and another one with wooden-piles foundation.