Geo-informatics

My PROJECTS

Software Development project : Dynamic Visualization of Tweets in Boston

Collaborators: Boqin Cai, Tamara Loran(University of Salzburg)

Introduction

In this project, we created an application that can dynamically visualize tweets from a database, as well as physiological measurements of a bike trip from a Web Feature Service overtime on top of a map of Boston. The project should show how to integrate data sets from different sources, especially with regards to the Open Geospatial Consortium (OGC) compliant standardized web services interfaces: Web Feature Service and Web Map Services. Java programming language was used for implementation in combination with Processing.

Implementation

The implementation will be illustrated by use of a UML Class diagram.

Figure 1 shows the UML Class Diagram of our application, which was created with the ObjectAid UML Explorer. ObjectAid is a code visualization tool, which can be implemented into the Eclipse IDE
as a plugin. We created eight classes, which are dependent on each other. First, we created three classes to connect to our different data sources: ”DBConnector”,“WMSConnector”, and “WFSConnector”. These classes are connected to the “TimeSeriesVisualiser” class, which is our executable class. The ”DBConnector” class reference the “Tweet” class in order
to pass data from the database to the “TimeSeriesVisualiser”. Map coordinates from the “WFSConnector” and from the “DBConnector” are transformed to screen coordinates by using the
“convertor()” method of the “CoordinateTransformer” class. Moreover, we decided to use an inheritance relationship to reduce code redundancy. Therefore, we created the superclass “SpatialTimeSeriesDataset” which is extended by the subclasses: “Tweet” and “PhysioMeasurement”
with their attributes and methods.

Figure 1. UML class diagram for our application

Figure 2. Output of the application

Figure 2 shows the output of our application. Our application is finally able to visualize the WFS,WMS, and the data from the database dynamically over time. As we can see, there is still room for
improvement, such as including a legend for the WFS and the database or add a time slide. Moreover, it would be desirable to manage the overlapping of the popups as well as remove old popups.

Software Development

Comparison of Random Forest and Support Vector Machine for Land use land cover classification

Figure 1. results

In this project, the performance of Support Vector Machine and Random Forest was compared for Land Use Land Cover classification. Sentinel 2 data was used for analysis and the study area selected was Linz in Austria.

The classification was performed using Arc GIS Pro. The Sentinel imagery data was first segmented in to segments then afterwards the SVM and RF classifiers were used to classify the land uses into five classes.

Support Vector Machine achieved a higher overall accuracy of 91% as compared to random forest with OA of 87% .

Application development: Visualizing Covid19 cases over the world

Collaborator: Bismarck Ade(University of Salzburg)

Introduction

The Web-Map seeks to visualize the spatial and temporal spread of Covid-19 pandemic. To achieve this, we implemented the following:

1. Visualization of the virus by country using the circle geometry to represent the number of cases.

2. An onclick pop-up to display the number of confirmed cases, deaths and recovered by country.

3. A search option to filter the cases by country

4. A line chart to visualize the temporal variation.

Data

The Covid-19 data was accessed using an API and integrated into the application. Data on the
reported, recovered and death cases were retrieved from https://corona.lmao.ninja/countries
whereas a time series data from https://coronavirus-tracker-api.herokuapp.com/all was used
for the line chart since its in time series format. The API incorporates data from John
Hopkins University CoronaVirus Resource Center and Worldometer.

Implementation

The design premise of the application is to visualize the covid-19 data in an interactive fashion. Open layer API was used for the web mapping functionality and D3.js used for displaying the line chart information. We used HTML and CSS to set the framework for the application by defining its structure and style. Using the grid container, the web map application interface was divided into a header, two columns and a footer. The header
element contains the title whilst the left-grid column holds the line chart and user search input. The right column contains the main web-map display as shown in figure 1.

Figure 1. Web application

Project: Detecting settlements in Salzburg City

Collaborator: Wei Tseng(University of Salzburg)

Settlements have their importance to the urban planners and is fundamental for designing urban policies at the city government bureau. In this project, we want to try the ArcGIS. Learn API on detecting settlement area in Salzburg. We use the Sentinel-2 imagery as our data source and manually select the settlement area.

Transfer learning was used for to detect the settlement. In the Arcgis.learn library, in this experiment, we used the U-net model. The most important characteristic of U-net is it's combination of the encoding and decoding phase and it predicts pixelwise segmentation.

The model was trained and the validation dataset was used to examine the performance of the model by calling the function mIOU. The return dictionary gives both the accuracy on the training set and validation set. We got a difference between two figures which indicates the model is overfitting our training set and fail to generalize the feature of the settlement. Some solutions to the poor performance include taking a better ground truth of settlement map. Or trying different hyperparameters (tile size, batch size, learning rate...etc)

Figure 2. Learning curve