AI FOR WATER
RAINSMORE/SWOT WORKSHOP ON ARTIFICIAL INTELLIGENCE
OCTOBER 24th-28th 2022
Online and Presential Conference
FORTALEZA, BRAZIL
This first workshop is finished, but we will come back next year for season 2 !
See some photos and download the presentations from the present workshop here
Realization
Speakers
Rafael
Reis
FUNCEME
Emilie
Caillault
ULCO
Kelly
Grassi
WeatherForce
Guilherme
Barreto
UFC/DETI
Nicolas
Araujo
UFC/GTEL
Manuel Sánchez-Montañés
Universidad Autónama de Madrid
Valdivino Alexandre de Santiago
INPE
Organization
MORNING
Lectures open to the public registered in the event via ZOOM
09h-10h
Introduction to the workshop
- Context and objectives of the workshop (Dr Marielle Gosset IRD ; Dr Geraldo Ferreira EOLLAB ; 15’)
- Why and how studying Extreme rainfall, limits of usual estimation / prediction models (Dr Modeste Kacou UFHB; Dr Romulo Oliveira, IRD/HydroMatters ; 30’)
- Why and how studying Small Water Surface (lakes, reservoirs), satellite data and actual limits (Rafael Reis - FUNCEME ; 15’)
10h-12h
Introduction to AI methods
- IA methods, a general overview (Pr Emilie Caillault Univ Lilles, 45’+questions)
- Time series forecasting using kernel regression methods (Pr Guillerme Barreto, UFC/DETI, 45’+questions)
- Week agenda and provided data set (M Gosset, 5’)
MORNING
Lectures open to the public registered in the event via ZOOM
09h-11h
Time series prediction, from time alignment to basic algorithms
(Dr Kelly Grassi, Weather Force; 45')
Times series completion algorithms used for prediction - DTWB
(Pr Emilie Caillault Univ Lilles ; 45')
11h15-12h15
Times series prediction by LSTM. Theory and example from Ceará
(Pr Nícolas Araujo ; 45’)
Example of problem to be solved - Rainfall time series analysis from Commercial Microwave links or sound sensors
(Dr. Marielle Gosset, IRD 15')
MORNING
Lectures open to the public registered in the event via ZOOM
09h-11h
Detection of patterns by Deep NN, transfert learning
(Pr Nicolas Araujo, UFC/GTEL ; 45')
Deep learning and object detection - Examples from medical applications
(Pr Manuel Sánchez- Montañés, Universidade Autónoma de Madrid; 45')
11h15-12h15
Training Rare Object Detection in Satellite Imagery by GAN and autoencoders
(Pr Nicolas Araujo, UFC/GTEL ; 45')
MORNING
Lectures open to the public registered in the event via ZOOM
09h-10h30
Clustering approach : Density, Shape, Hierarchical Kernel - approaches
(Pr Emilie Caillault Univ Lilles ; 45')
MSC a multi-level spectral clustering approach to detect extreme events, unknown patterns
(Dr Kelly Grassi, Weather Force; 30')
11h-12h
How to combine clustering and classification results, relevant object class
(Pr Emilie Caillault Univ Lilles ; 45')
Examples of application of unsupervised learning
(Dr Kelly Grassi, Weather Force; 30')
MORNING
Lectures open to the public registered in the event via ZOOM
09h-10h
Semantic Segmentation of Satellite Images
Dr Valdivino Alexandre de Santiago, INPE; 45’)
10h-11h
Results from participants
Discussion
11h-12h
Panel discussion / next workshop
AFTERNOON
Limited attendance: Class at the LABOMAR’s Computer Laboratory
14h-17h
Hands on - EOLLAB - Computer Lab (limited attendance)
- Python installation/setup (installation, creation of virtual environments, required packages)
- Regression with data visualization.
- Data treatment/splitting: training, testing, and validation datasets.
- Usual metrics.
AFTERNOON
Limited attendance: Class at the LABOMAR’s Computer Laboratory
14h-17h
Hands on - EOLLAB - Computer Lab (limited attendance)
a- Notebook explained / Guided tutorial by Emilie Caillault 1h. Alignment. completion. prediction (kalman, DTWBI, LSTM, Random Forest)
b- Application to study cases with support (Emilie CAillault, Kelly Grassi) 2h min. Using Dataset #1
Obs: Apply algorithms to vector data (not images at this point) which will meet the nature of the data presented in the morning sessions
AFTERNOON
Limited attendance: Class at the LABOMAR’s Computer Laboratory
14h-17h
Hands on - EOLLAB - Computer Lab (limited attendance)
- Supervised learning methods
- Introduction to NN: MLP, Transfert Learning (deep NN)
- Mention of more advanced techniques: GANs, Federated Learning, etc., but not hands-on (due to time/computing constraints).
- Which dataset will be used? (small resolution!)
a- Notebook explained / Guided tutorial 1-2h (data preparation and classification)
b- Application to study cases with support 2h min.
AFTERNOON
Limited attendance: Class at the LABOMAR’s Computer Laboratory
14h-17h
Hands on - EOLLAB - Computer Lab (limited attendance)
- Detect on Dataset #1 the raining period (start, end)
- Clustering vs explicit segmentation based on rupture/breaks
a- Notebook explained / Guided tutorial 1h (clustering)
b- Application to study cases with support
2h min.
14h-17h
Hands on - EOLLAB - Computer Lab (limited attendance)
- EOLLAB (computer lab)
Q&A on the exercises & practice of the week?
Tutorial
Schedule
October 24th
Displayed in Brasilia time (GMT-3)
Introduction to rainfall/water monitoring needs and Artificial Intelligence (AI) methods
October 26th
Displayed in Brasilia time (GMT-3)
Images classification and pattern detection using AI
1- Supervised approach
October 25th
Displayed in Brasilia time (GMT-3)
Time series analysis and prediction using AI. Application to rainfall estimation from MW links ; radar and satellite
October 27th
Displayed in Brasilia time (GMT-3)
Cmages classification and pattern detection using AI
2- UNSupervised approach
AFTERNOON
Limited attendance: Class at the LABOMAR’s Computer Laboratory
October 28th
Displayed in Brasilia time (GMT-3)
Other methods and applications: Results from hands on
04
11
203
DAYS
SPEAKERS
PARTICIPANTS
and counting...
About the event
The Rainsmore project
RAINSMORE, for ‘Raincell et Autres INnovations, Satellites et Mesure Opportunistes issues des Reseaux de telecommunication, pour Estimation et spatialisation des precipitations’ is an International Research Network (IRN) financed by the french Institut de Recherche pour le Dévelopement (IRD).
RAINSMORE objective is to gather experts in data science and hydrometeorological sciences, to innovate together for better rainfall measurement and analysis.
RAINSMORE gathers scientists from Africa, South America and Europe.
The SWOT Mission
The SWOT (Surface Water Ocean Topography) mission, which is a fruit of cooperation between France, the United States, the United Kingdom, and Canada. SWOT is a satellite dedicated to studies on ocean topography and terrestrial water bodies dynamics and will open new perspectives for fields such as oceanography, hydrology, among others.
SWOT NordEste is a project that gathers various reserach and operation agencies in Brasil and France, with the objective to validate and valorise the data from the upcoming SWOT mission in Nord Este Brasil, for different applications.