Gaia

https://drive.google.com/drive/folders/1YT4iToYEzDL6ID19xtaF0SJTRq3q5ojg?usp=sharing

https://drive.google.com/file/d/1ewhoJaPnsWg-ZeZLgnOgcWmfaEzrlI8a/view?usp=sharing

-What exactly does it do?

The solution proposed by Gaia conceptually allows the images to be processed before interferometry using a machine learning model trained to detect tropospheric water vapor (TWV). 

-How does it work?

The machine learning model will analyze sets of images containing TWV and compare them with images of the same site but in another season of the year when the TWV concentration is lower. The objective is to recognize the TWV, to characterize and extract it from the SAR images. 

We are trying to look for the TWV change using machine learning algorithms with images with no terrain change.

To achieve that, we must train the ML model under specific criteria datasets, we used earth engine to see the total water vapor column distribution and selected some possible locations to train the model under these criteria:

• Big columns of water vapor (Thailand, Vietnam)
• Medium accumulation of water vapor (Brazil, Venezuela)
• Low water vapor (United States, Iraq)
• Very low (Greenland, Chile)

We also wanted to include terrain criteria, which are:

• Stable sites, with null seismic or volcanic activity
• Not too much vegetation
• Flat surfaces if possible

The results of this selection will train the ML model to ensure the detection of only TWV. After the ML model, we will be able to extract TWV, which we will treat SAR images before making interferometry.

-What benefits does it have?

SAR image cleaning will provide clean data for interferometry without TWV bias. This way, the InSAR analysis will provide reliable data about the terrain change only allowing us to perform reliable analysis on the Earth’s surface

-What do you hope to achieve? 

The aim is to improve the quality of interferometry to eliminate the bias produced by TWV noise, allowing reliable and safe analysis of InSAR images, which can help to address the existing problems with InSAR imagery.

-What tools, coding languages, hardware, or software did you use to develop your project?

By reviewing open source literature, a python script for machine learning will train performing tropospheric water vapor detection using google collab as a notebook platform. The coding language used in the elaboration of this project is Python. 

Provide specific details about what space agency data you used in your project:

Earth Engine Data from Google was used to understand the changes detected in SAR images with Mort Canti's tutorial. Likewise, information was obtained from VERTEX of the ASF.

-How you used it?

Public data is used to obtain SAR imagery from the SENTINEL 1 satellite.

-How it inspired your project?

Advanced Rapid Imaging and Analysis (ARIA) resources were of great inspiration, observing the effort they make to analyze the data after a catastrophe and make it available to everyone. 

Similarly, the VERTEX portal of the ASF with the search by event inspired us by looking at the images of the most recent seismic event in our area of interest, Mexico.

-How would you describe your Space Apps experience?

It was a real challenge because at the beginning we started without really knowing what we needed to deliver for the challenge, but it was not until we reached the team dynamics, where with the help of our machine learning engineer, we were able to better understand the technical context. The integration of the team was challenging at the beginning, since we had a 10-hour time difference with our teammates, however, we had a really beneficial experience while meeting each other. Just before the start of the hackathon, we were fortunate to find a new member who arrived with the best willingness to help. We experienced many emotions, such as amazement, excitement, disappointment, overwhelm, and despair, all together, and even one of the members caught a disease during the event. However, community support led us to continue and propose our solution to the challenge.

-What did you learn? 

We learned a lot about remote sensing, the processes needed to obtain critical data, such as centimeters of displacement, as well as the different techniques for performing analysis using SAR imagery, recent literature explaining current efforts to make use of this data. We learned the value of individual contributions, and without a doubt, we would not have solved without the support of the whole team.

-What inspired your team to choose this challenge?

Undoubtedly, the first step to entering this challenge was researching the mystery of the September 19 earthquakes in Mexico. This inspired the first two members of the team, during the team we started building dynamics with our third member that joined us later, because of his love for data and research. When we asked on the Discord channel if anyone else was looking for a team for the challenge, we found our machine learning engineer who was researching algorithms for image processing. Finally, our last member joined us by looking for us through the Discord channel.

-What was your approach to developing this project?

Initially, we were looking to improve the main problem described in the challenge by analyzing InSAR data, however, after researching and brainstorming about the high level of technical knowledge needed to solve this issue, we decided to go back to basics and look for a root problem. Therefore, we defined the route to follow and the subsequent steps of the project.

-How did your team resolve setbacks and challenges?

We were greatly affected by this, but perseverance and the constant question of "what if" led us to complete the delivery, and with no regrets. 

-Is there anyone you'd like to thank and why?

Every scientist in the world, working tirelessly to provide the world with the best quality data and make it available on a not-for-profit basis. The community that contributes to open source development will always have our heartfelt gratitude.

Oluwasesan A. Falaiye, Oladiran J. Abimbola, Rachel T. Pinker, Daniel Perez-Ramírez, Alexander A. Willoughby. Multi-technique analysis of precipitable water vapor estimates in the sub-Sahel West Africa. Heliyon 4 (2018) e00765 https://doi.org/10.1016/j.heliyon.2018.e00765 (1)

Martin Lainer, et. al., (May 2019) Significant decline of mesospheric water vapor at the NDACC site near Bern in the period 2007 to 2018. Retrieved from: https://www.researchgate.net/publication/333176830_Significant_decline_of_mesospheric_water_vapor_at_the_NDACC_site_near_Bern_in_the_period_2007_to_2018/figures?lo=1 (2)

Fadwa Alshawaf, et. al, (August 2017) Estimating trends in atmospheric water vapor and temperature times series over Germany. Retrieved from: https://amt.copernicus.org/articles/10/3117/2017/amt-10-3117-2017.pdf (3)

Benjamin Fersch. Tropospheric Water Vapor: A comprehensive high-resolution data collection for the transnational Upper Rhine Grabe Region. Retrieved from: https://essd.copernicus.org/preprints/essd-2022-57/essd-2022-57.pdf (4) 

Chen, C.; Dai, K.; Tang, X.; Cheng, J.; Pirasteh, S.; Wu, M.; Shi, X.; Zhou, H.; Li, Z. Removing InSAR Topography-Dependent Atmospheric Effect Based on Deep Learning. Remote Sens. 2022, 14, 4171. Retrieved from: https:// doi.org/10.3390/rs14174171 (5)

Xiofan Qu, et. al., (2020) Change Detection in Synthetic Aperture Radar Images using a dual domain network. IEEE Geoscience and Remote Sensing Letters, vol. 19, pp. 1-5, 2022, Art no. 4013405. Retrieved from: https://github.com/summitgao/SAR_CD_DDNet#readme

Canty, M.J. (2019). Image Analysis, Classification, and Change Detection in Remote Sensing: With Algorithms for Python (4th ed.). CRC Press. https://doi.org/10.1201/9780429464348

#SAR #ML #TMW #SpaceApps