GovTech News

Today we are excited to present you a solution co-created by Statistics Lithuania and Kaunas University of Technology (KTU) scientists. Statistics Lithuania has submitted a challenge for the creation of a real-time satellite image of Lithuania for agriculture land recognition. A group of scientists from KTU has been developing a solution that closely meet their Statistics Lithuania needs and during the GovTech Co-create Accelerator they have applied the solution. We talked with Jan Golubovič from Statistics Lithuania and Valentas Gružauskas from KTU – let’s read more about their co-creation!

Tell us more about your submitted challenge for GovTech Challenge Series 2.0 and why is it important to solve it?

Jan: Having to prepare official statistics in the old way, by using questionnaires, is a problem for two reasons:

1. We have to bother respondents and waste their precious time,

2. Because information acquired this way is not ideally accurate or timely.

To mitigate these two problems, we are searching for the ways to use modern technologies to acquire information for statistical process.

For this reason, we started this challenge to have a primal source: a clean, cloudless, fogless map of Lithuania with highest possible periodicity. Having that we will work on solving problems of official statistics and maybe creating some experimental statistics along the way. Our first task in this work is to identify crop fields and then crop types on those fields so we don’t have to ask farmers for the information.

It is important to solve this problem and other problems of the same sort, to create even more reliable, accurate and timely statistics.

Why is it important to look for innovative solutions that could be applied for the public sector use?

Jan: World is changing and becoming more and more digitalised, and it is in the interest of Department of Statistics to harness this digital information and create official statistics from it. With each year, our user base is changing and demand for timely statistics is growing. To stay relevant and useful, we also have to change our work methods and look for the new and innovative solutions.

Do you have any tips for other institutions who want to work with startups and apply innovations?

Jan: Just try it. If it doesn’t go well, all you lost is your time. If it goes well, you might find a surprising solution. But in any case, you will get some experience and a wider range of view.

Tell us more about your submitted challenge for GovTech Challenge Series 2.0 and why is it important to solve it?

Valentas: We are a group of researchers from KTU working with various artificial intelligence applications. The team consists of Andrius Kriščiūnas and Tautvydas Fyleris from Informatics faculty, and Valentas Gružauskas from School of Economics and Business. Previously, we have worked with object recognition from aerial images, during which we gained practical experience with computer vision application for large-scale data analysis. Because of our experience, we applied to GovTech Lab project and work on Statistics Lithuania challenge – „How to create a real time satellite image of Lithuania?“

The challenge requires developing a real time Lithuania map by using open access data from European union – Copernicus. The map comprises Copernicus 1 and Copernicus 2 images. Copernicus 1 images are radar images, which can be used to analyse earth surface reflection signals in radio spectrum, while Copernicus 2 data is optical satellite images. The challenge in developing such a map is related to noise in data and data size. For example, radar images can have noise in them, which reduces the quality of analysis. While satellite images consist of clouds and needs additional pre-processing. In this case, the term real-time is relative, since it means a time period which can be used to obtain the whole country view. This period depending on weather conditions might be 1 week, or even several months. For example, aerial images of Lithuania are usually developed in 2-year period. Thus, the main goal is to develop a program which could automatically download, store and provide an interface for consumers to further work.

The second challenge is data size. For example, Lithuania’s map of 1 areal period can take up to 1 – 3 terabytes of space. The described pre-proceeding steps and technical issues are not easily implemented due to required expertise and technical requirements, thus such data could be used in studies as well. For example, students might want to analyse Lithuania from a satellite image perspective, and they would need to reproduce the whole described process. However, if they use our already completed system, they can directly focus on analysis, rather on trivial tasks. Of course, the solution might not be limited to the public sector, but also designed for private companies to help plan infrastructure, investment strategies, valuation of real estate and used in other activities.

Why is it important to look for innovative solutions that could be applied for the public sector use?

Valentas: To understand better for what purpose the map could be used, it is important to know the traditional approach towards statistical indicator development. In this case, Statistics Lithuania is mostly interested in agriculture land recognition to have a possibility to know the actual situation in Lithuania more precisely. Currently, surveys are being used to collect this information, however surveys have their limitations.

Firstly, statistical indicator accuracy is crucial for practical use. Surveys in this case can lead to inaccurate data because of human error. Another issue is that surveys are being applied irregularly, and because of this the information renewal period is high. Lastly, for better analysis, more precise data is needed. In this case, from satellite images we can obtain precise coordinates, shapes and identify a different agriculture land types. Of course, the satellite images can be used to identify other type of information not limited to agriculture land. For instance, in our previously conducted project with areal images, we identified buildings. However, the quality of the images is crucial for accuracy and what kind of objects can be identified.

In this case, Copernicus has 10 x 10 meter per 1 pixel, while areal images have 0.25 x 0.25 meter per 1 pixel, which can lead to higher accuracy. However, there is a trade-off between frequency, costs, and accuracy, which needs additional analysis to determine what kind of data sources can be used for which specific purposes. Also, the whole country image size of 1 period can take up to 1 terabyte of disk space. If long period needs to be analysed, it can lead to high computational costs due to disk space and processing time. At the same time, you need to think about the functionality and scalability of such a system. If the product must be universal and have a possibility to support various needs of customers, the development of the system should be different.

Do you have any tips for other institutions who want to work with start-ups and apply innovations?

Valentas: When government works with commercial companies, it is important to understand the required solution and market size of that solution. If the government wants to reduce expenses for product development, they can help the startup to develop and help obtain more customers.

However, if the product is unique to their needs, the product might not have a larger market. In this case, the government should plan a larger budget for such product purchasing. This statement would lead to my second recommendations, which is related to funding and bureaucracy of government institutes. Since the processes are slow in government, it would be better to plan and prepare the documents in advance and to better understand the needs of your own organisation. To do that, the government institutions could try to answer several questions before searching for solution providers. Will the solution be a separate product or will it be included in the existing system? If the solution will be integrated in the existing product, it is important to know the integration possibilities.

Lastly, the institutions need to consider all possible users and their needs. In this project case, if the solution will be applied only for agriculture, and we might not think about possibilities to provide data to other organisations. However, we could plan in advance and develop the system from the beginning for multiple users, for different purposes for data application.