Since ancient times, people have used maps to find their way, divide territories as well as learn about distant places. As technology progressed, maps transformed from scrolls and murals into digital repositories. In their high-tech form, such records can hold different layers of information, thus overlapping more items into one. The entire system necessary to assemble, store, update and display such information is called a Geographical Information System (GIS). A GIS includes the hardware, software, and know-how to combine geography and information management.
Why and When to Use GIS
As all GIS systems contain a wealth of information classified by geographical locations, these make excellent training datasets for AI systems. This connection to AI is natural in the context of recent advances in computer vision and image recognition. By now, there have been some successful attempts to use GIS and AI for pollution management or to control disease spreading.
Satellite images provide data at different levels of depth, which is still severely underused. Thanks to GIS, these can be transformed into actionable solutions in a few hours. There are some challenges though when it comes to identifying buildings, roads, and other objects.
Another straightforward use is integrating GIS with IoT data and creating real-time solutions to industrial and urban problems. Such solutions can also get data from public institutions, such as health reports, pollution measurements, and more.
Although this is a tempting opportunity, there are a few limitations and challenges of using GIS and AI together, and most of them have to do with infrastructure, costs, data structure, and human-related barriers.
Infrastructure and cost challenges
Maintaining a GIS system requires complex hardware and software components since it processes a lot of visual data and requires significant computational and storage resources. To function correctly, the elements need to be interconnected.
Such hardware and largely cloud-based solutions come with a considerable price tag, and most projects attempting GIS require a reasonable budget, which can be an obstacle for smaller companies. As GIS becomes more popular, it will be included in SaaS solutions, which democratizes the costs.
Data Structure challenges
Most of the times, a GIS is hard to integrate with traditional maps as it is self-supporting. This means that instead of building a GIS on top of existing routes, like in the Google Maps system, it needs to be created from scratch, thus leading to additional workloads.
Even if a GIS system is created by “reading” existing maps, it still raises some challenges related to interpreting text and contours. Advancements in OCR have made this more accessible, but it is not yet flawless and still needs human supervision.
To work efficiently with AI, GIS systems require a large number of data points to learn from. In the case of rare but devastating processes like disease outbreaks, there is not enough information to serve as a learning material. This is also the case when GIS is used to evaluate historical methods when records are used as a proxy for future developments.
The diversity of data is another problem, as very similar data can lead to overfitting the model, with no new insights. Deep learning relies on spectral data and provides high accuracy. The algorithm goes over different data sources and populates the GIS. Spectral data means that you can use only one dataset with different spatial and spectral resolution.
A problem highlighted in a recent GIS Lounge article also mentions a language barrier. Most data image sources are in English, and this limits the application of AI to countries which don’t have the necessary input to provide comprehensive datasets.
There are still problems related to how GIS software handles spatial statistics, as this domain is in its early development stages. Since there are few GIS providers and the software is proprietary, there has been no satisfactory development of solutions, plugins, and add-ons. Every problem needs to be solved by the AI software developers from scratch.
Additionally, GIS relies on significant data generalizations and information simplification. Although necessary for fast computing and map generation, such assumptions can hurt the result.
The people using and developing a GIS system need to have some know-how in data science, machine learning, and geography. On top of that, they need to understand the basics of the industry they are working with and be able to collaborate with specialists from other areas.
Another challenge is related to decision-makers. Most of the times, these people don’t have a thorough understanding of GIS systems, so they can’t offer the necessary support and don’t usually provide sufficient budgets. There is a significant gap between GIS users and decision-makers, and this is made even more profound by using specific jargon. The key takeaway from this and a quick solution would be to focus on the ability of GIS to produce maps that are easy to understand and use, compared to raw data in tables.
Future outlook and solutions
Machine learning and deep learning can help organizations make good decisions fast. These methods can offer comprehensive solutions related to location services like fleet localization, location intelligence, asset tracking and management and other real-time surveillance.
The downside is that until now there have been different challenges to it. The first one is related to the infrastructure and associated costs, but as technology progresses this will soon be an issue of the past.
The next obstacle and probably the hardest to overcome is the data structure. Every AI system needs reliable data to learn from, and so far there haven’t been enough reliable sources to provide the necessary material.
To overcome the software challenges, a possible solution would be to create some open-source GIS projects where the core code belongs to the community so that developers can extend it.
Last but not least, there is still a lack of qualified personnel. As AI and ML are in high demand, universities started offering programs, but it will be at least 3-5 years until graduates reach a proficient level.
Note: This is a guest blog by Jasmine Morgan, who is a Technology Consultant with a software engineering academic background.