FloodEst| A Flood of Ideas
A Flood of Ideas
Flood Forecasting and damage minimizing using Machine Learning
Summary
Our project is to estimate the damage caused to infrastructures by floods through monitoring satellite data before, during and after floods. Thus, by developing this idea the main objective is to predict and lessen the damage and create solutions to reduce the factors that cause the flooding.Few main aspects that can cause flooding are drainage basins in urban areas, heavy rainfall or the long periods of rain, snowmelt, steep slopes, compacted or dry soil, water streams and rivers overflowing. By analyzing these factors that highly affect flooding, the process of estimating the damage to the infrastructure can be easily developed.
How We Addressed This Challenge
What we developed
Developed a flood hazard assessment and damage estimating methodology.
Why it is important
Flooding causes a great impact as a natural hazard to human infrastructure around the world. Critical infrastructures include health and educational facilities, transportation, communication, water systems, etc. which are damaged heavily due to flooding. The geographical structure has to be considered when determining the disaster damage and it is important to understand how infrastructure systems are affected due to large-scale flooding so that the disaster risk can be reduced substantially.
What it does
The solution introduced addresses the challenge of estimating the flood impact to critical infrastructure using combined past collected satellite data.
How it works
The working of the introduced solution can be summarized by the diagram shown below.
Basic diagram of the project
Figure 01 : The introduced solution model as a basic block diagram
What we hope to achieve
Discover new ways that Earth observations can contribute to the monitoring and reporting of critical infrastructure impacts from flood events across the world
How We Developed This Project
What inspired our team to choose this challenge
Floods are considered as one of the most deadly natural disasters which causes losses in lives, loss of economic status, and assets of people. Therefore, appropriate preparedness to floods can save lives as well as damages caused to infrastructure. So that the thought of giving a solution to this highly concerned problem inspired us to choose this challenge and we hope that by our solution a great impact can be made on the disaster risk of floods world wide.
Our approach to developing this project
Flooding areas are divided into small cells and dynamic and static parameters in each cell can be observed in different manner (by satellite data/mapping, real time updated data bases).
Figure 02 : How the map is divided to cells
Dynamic parameters
o Rainfall
o Water streams, rivers and their directions
o Temperature
o Soil data(Moisture/infiltration)
o Population distribution
Static parameters
o Infrastructure type and location
o Height from sea level
Machine Learning Code 1
The model 1 is developed using historical data of flood scales and selected parameters. Then new data are input to the model to calculate the scale of flood and flood probability.
The above mentioned parameter data are taken as input to the system and overall transfer function(how that area responds for flooding) for the relevant flooding area can be derived. This is called as the state space(Plant). Such that, the model(observer system) can also be derived in the above manner .Then using previous flooding hazard databases and other resources, the model can be modified and trained using Machine Learning (ML) techniques that are given Figure 03.
Figure 03: Machine Learning methods used for flood prediction in the literature
· ANFIS - Adaptive Neuro-Fuzzy Inference System
· MLP - Multilayer Perceptron
· WNNs - Wavelet Neural Networks
· EPS - Ensemble Prediction Systems
· DT & RF & CART - Decision Tree & classification and regression tree & random forests
· ANNs – Artificial Neural Networks
· SVM & SVR – Support Vector Machine & support vector regression
For speeding up Machine Learning Algorithm PCA is used, as PCA's main idea is dimensionality reduction (the maximum varying directions/components can be found by neglecting other small varying components). Therefore, the training time and testing time of the machine learning algorithm is sped up since having a large number of features can make the ML algorithm too slow.
Since flood hazard is a quick scenario to minimize the damages(for infrastructure), it is important to estimate the flood reaction of the area (Plant) before some considerable time (about 1-2 days). But the introduced system is a real-time execution system. Therefore, we have to execute our Plant and estimated observer until some critical point before the flooding occurs (last point that hazard can be minimized). This can be done by using Kalman Filtering technique. That is the observer model is estimated with minimizing the error between real time Plant and estimated observer model until it reaches the critical point that is mentioned above. Therefore, resultant process covariance is more biased towards the estimated observer model. Then we have to neglect the real time plant and work with the estimated observer model using previously collected data resources and static parameters that are mentioned above.
Therefore, by the Model 1 the scale of flood and flood probability for each cell of the area concerned can be calculated.
Machine Learning code 2
Machine Learning code 2 is trained using historical damage data, flood probability, scale of flood, type of infrastructure and cell number and when new data is given the damage that can happen to the relevant area can be calculated.
Infrastructure damage risks can be categorized in two ways.
· Direct risk: People getting affected directly due to a failure of an asset which they heavily depend on. For example: If a power station was damaged due to floods, electricity customers are directly affected.
. Indirect risk: People getting affected due to a failure of an asset which indirectly causes damages to a service or an asset they depend on. For example, The passengers of an airport get indirectly affected by a failure of a power plant.
Considering these infrastructure damage risks, a value for risk probability can be calculated for each cell depending on the priority of infrastructures and their damage risk type. Therefore, necessary actions can be taken in order to prevent large scale risks by alarming and taking safety precaution actions according to the risk estimation map.
Problems and achievements
We were able to learn about the challenge referring to the resources given.
We could get a handful of experience referring to the given resources and the satellite observation data.
The team members could learn more about how to handle satellite data and apply them more practically for the solution.
It was difficult to manage the limited time in the competition to read all the resources available and allocate time to find the solution.
How We Used Space Agency Data in This Project
The below listed data from the space agency are used as inputs to the developed methodology to achieve the required solution to the challenge.
- The global rainfall data that was acquired from JAXA Global Rainfall Watch was used to input the system in estimating the desired results. https://sharaku.eorc.jaxa.jp/GSMaP/index.htm
- Homeland Infrastructure Foundation Level data https://hifldgeoplatform.opendata.arcgis.com/searchgroupIds=6db5b0248ea2468aa966e79d8508eeb5
- Population distribution data https://www.worldpop.org/geodata/listing?id=78
- Real time land information https://weather.msfc.nasa.gov/sport/modeling/lis.html
Project Code
Data & Resources
- Pant, R., Thacker, S., Hall, J.W., Alderson, D. and Barr, S., 2018. Critical infrastructure impact assessment due to flood exposure. Journal of Flood Risk Management, 11(1), pp.22-33.
- Unterberger, C. How Flood Damages to Public Infrastructure Affect Municipal Budget IPant, R., Thacker, S., Hall, J.W., Alderson, D. and Barr, S., 2018. Critical infrastructure impact assessment due to flood exposure. Journal of Flood Risk Management, 11(1), pp.22-33.
- Jetten, V.G., Alkema, D., van Westen, C.J. and Brussel, M.J.G., 2014, November. Development of the Caribbean handbook on disaster risk information management. In International Conference on Analysis and Management of Changing Risks for Natural Hazards 2014.
- Mosavi, A., Ozturk, P. and Chau, K.W., 2018. Flood prediction using machine learning models: Literature review. Water, 10(11), p.1536.
- Real time land information https://weather.msfc.nasa.gov/sport/modeling/lis.html
- Rainfall https://sharaku.eorc.jaxa.jp/GSMaP/index.htm
- Population distribution data https://www.worldpop.org/geodata/listing?id=78
- Homeland Infrastructure Foundation Level data https://hifld
- geoplatform.opendata.arcgis.com/search?groupIds=6db5b0248ea2468aa966e79d8508eeb5