High-Level Project Summary
The Amazon is subject to changes in its landscape as a result of natural or anthropogenic actions. These events happen mostly unannounced. As a solution, we developed the idea of an application to identify risk areas for landslides due to rain erosion, which would work remotely, using data from Sentinel-2, with data from Amazonian rivers. Volunteers registered in the app would share information such as location, photos and detailed description of possible landslides. Satellite images will be used in the training of CNN, in a classification learning with images of the location where the landslide occurred, in a predictive way, with images prior to the occurrence of landslide.
Link to Project "Demo"
Link to Final Project
Detailed Project Description
Object
The Amazon floodplain region is affected by several changes as a result of natural forces, such as torrential rains or changes in the course and forces of rivers, or anthropogenic actions, changes in land use, such as the opening of forests for agriculture and livestock. All of this, in addition to contributing to climate change, enhances the occurrence of disasters such as stronger floods or droughts, flooding and landslides in some regions, leaving community residents in areas at risk at the mercy of luck in the absence of competent bodies. For this, a monitoring tool for risk assessment that would issue an alert to the responsible institutions.
Use data provided by the Sentinel 2 Satellite to identify areas at potential risk of landslides due to fluvial erosion, in the Amazonas-Solimões River channel and prevent any socioeconomic and environmental damage caused by these landslides, protecting the population. The action takes place between the application that will collect and store information, Fallen Soils Analysis, and the competent bodies, such as the Civil Defense and Fire Department, which will be able to use this information to relocate people to a safe place with prior notice.
How to use the App?
The app is an identifier of landslide risks caused by river erosion that allows users registered in the App to share information such as the location of possible landslides in the area they reside or are visiting, the climatology and environment of the place, in addition to other details that they note. Users can also update the information entered according to the database provided in the App of past events.
The Brazilian Amazon Fallen-soils Identification APP (FSIBA) will not need a password to start and will work offline if the smartphone tries to built-in GPS. The dynamics of the application will have automatic pre-made information to speed up the insertion of information and with text boxes with details that the user himself describes something that is not included in the pre-made answers, in addition to using the "Drag and Drop" tool (drag and drop). The function of calling the competent bodies is also available, if necessary. For example, if volunteers notice that a landslide is likely to occur in a certain area immediately, it can be indicated by the App, the competent Civil Defense agency.
The use of the Application is extremely important, so that possible landslides in some region of the middle Solimões are less catastrophic in relation to damage caused to residents in the area of occurrence, mediated by prior notice. Also considering that in the Brazilian Amazon region this resource is not available, in which users are residents or visitors in risk areas and contributors to the data capture, it is necessary to empower the community itself to manage the data and protect itself same.
The application will also provide data for the web App (FSA: Fallen-soils Analysis) based on the R language that will make estimates by satellite images classifying in risk areas or not.
Data analysis
To remotely identify and classify risk areas, satellite data (Sentinel-2 [1]) will be used to identify areas in the Brazilian Amazon (Rio Solimões – Amazonas) with a high risk of landslides. Satellite images will be used to train the Convolutional Neural network (CNN), a deep neural network, in a supervised learning classification with a set of images of the location where the landslide occurred, predictively, with images prior to the occurrence of the landslide, testing within one year. You will also have images where there is low risk, following the same procedure. These two classes, high risk and low risk, for training, will be identified by historical observation of the banks of the Rio Solimões – Amazonas. The initial identification of the locations will be by searching news about the landslides on news websites on the internet, on social networks and by the NASA Landslide Viewer application. The CNNs by the shape and reflectance (of exposed soil, climax forests, capoeira forest, pastures) of the riverbanks can find patterns that identify areas of high risk of landslides.
The R software [2] will be used in the image processing, specifically the TensorFlow [3] and Keras [4] libraries. 80% of the data will be used for training and 20% for testing. Activation function will be ReLU (Rectifier), learning rate will be 0.002, dropout rate will be 0.2, density will be 1024, batch size will be 50 and 300 epochs. Pre-trained models (xception) from ImageNet will be used [5]. The accuracy of the model will be tested on the basis of training and testing. Images with low incidence of clouds (less than 10%) will be selected. The Sentinel -2 RGB (GeoTIFF) images will be fragmented by 5% vertically and horizontally.
The trained networks will be used to create an APP published on shinyapps.io built in RStudio [6]. The application on the website can be used to estimate whether the new satellite image could be of the slip risk class or not. R script attached (FSA: Fallen-soils Analysis).
To improve the location of areas where the landslide occurred, a mobile APP will be created for end users residing in the Brazilian Amazon. The Brazilian Amazon Fallen-soils Identification APP (FSIBA) allows the landslide location record that can be used to improve CNN training, with more satellite images in the subsequent data processing. It is emphasized that the FSI can be used by community members in the region who suffer daily from landslides caused by the flood pulse of rivers.
After using the FSIBA APP, with the generation of a lot of data, the heat maps generated by the Kernel analysis using the spatstat package [7] of R will be included on the site. This allows a visualization of landslide concentration in the region.
Space Agency Data
1. Sentinel. https://sentinel.esa.int/web/sentinel/sentinel-data-access
2. R Core Team (2021). A: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/
3. Abadi, M et al. 2015. TensorFlow: Large-scale machine learning on heterogeneous systems, 2015. Software available from tensorflow.org.
4. Kalinowski et al. 2021. R Interface to Keras: keras (rstudio.com). https://keras.rstudio.com/
7. Baddeley A, Rubak E, Turner R (2015). Spatial Point Patterns: Methodology and Applications with R. Chapman and Hall/CRC Press, London. https://www.routledge.com/Spatial-Point-Patterns-Methodology-and-Applications-with-R/Baddeley-Rubak-Turner/9781482210200/.
Hackathon Journey
The Space Apps Challenge 2021 in Tefé was a stimulus to the development of science and technology in the interior of the state of Amazonas. The construction of an innovative idea provided a unique experience for the “Bertholletia excelsa” team, which is necessarily tireless in solving the challenge. We chose the challenge “Risk identification with science + communities” to propose a tool with the potential to solve recurrent problems in the Middle Solimões region, where the city of Tefé is located. We use the citizen science approach, in which data collection encourages scientists to make a difference, contributing information about the displacement of land that takes place around the community. The team spirit was essential to deal with the lack of internet, which was the main challenge for the development teams in the municipality of Tefé.
References
Barros, D. F.; Albernaz, A.L.M. 2014. Possible impacts of climate change on wetlands and its biota in the Brazilian Amazon. Brazilian Journal Biology, 74 (4).
Fearnside, P.M. 2008. Mudanças climáticas globais e a floresta amazônica. pp. 131-150. In: Biologia e Mudanças Climáticas Globais no Brasil. Marcos S. Buckeridge (ed.), RiMa Editora, São Paulo, Brasil. 295 pp.
PORTO. Kátia de Souza. Impactos socioambientais do processo de ocupação da orla do município de Tefé/ Amazonas- O Bairro de Juruá. São Paulo: USP, 2011. Dissertação de Mestrado – Universidade de São Paulo, 2011.
PORTO. Kátia de Souza. Impactos socioambientais do processo de ocupação da orla do município de Tefé/ Amazonas- O Bairro de Juruá. São Paulo: USP, 2011. Dissertação de Mestrado – Universidade de São Paulo, 2011.
Ramalho, E. E.; Macedo, J. (2009) Ciclo hidrológico nos ambientes de várzea da reserva de desenvolvimento sustentável Mamirauá – médio rio Solimões, período de 1990 a 2008. UAKARI, v.5, n.1, p. 61-87, jun. 2009.
Silva, Amanda Caroline Cabral da: As cheias excepcionais e os impactos socioambientais na cidade de Tefé-AM.
Tags
#Amazon, #fallen soils, #water, #river, #ribeirinhos
Global Judging
This project has been submitted for consideration during the Judging process.