Differences

This shows you the differences between two versions of the page.

--- working_groups:computational_earth_sciences:ai4es:applications [2020/02/19 14:29]
cgome1
+++ working_groups:computational_earth_sciences:ai4es:applications [2020/08/07 15:44] (current)
cgome1
@@ Line 1: / Line 1: @@
-===== Current BSC-ES projects involving AI =====
+===== BSC-ES projects where the AI4ES research line plays an important or leading role =====
-==== Atmospheric Composition ====
+For a more complete list of ongoing and potential use cases, see the strategy document.
-Inference of surface pollutant concentrations from satellite and global scale meteorological and chemical simulations (AQ-WATCH project). Application to NO2 (TROPOMI sensor) and aerosols.
+  * AI for extreme extreme weather events and climate services (ASTERISKS H2020). Deep Neural Networks from the field of computer Vision can be used for pattern recognition in a supervised way, exploiting reanalysis data and historical event databases (e.g., IBTrACS for tropical cyclones) or other impact data.
+  * Learning to simulate precipitation with Deep Neural Networks ({{https://www.esiwace.eu/events/6th-hpc-workshop/presentations/learning-to-simulate-precipitation-with-deep-neural-networks|slides}}). Talk presented at the 6th ENES HPC workshop. We implemented encoder-decoder networks, such as those used for image translation in Computer Vision, for the task of learning precipitation fields from other predictor variables. The  transfer function aims to capture information from predictor variables to infer precipitation, a variable highly dependent on parameterizations schemes.
-==== Climate and Weather Prediction ====
+  * Simulating precipitation with supervised and generative learning models. Abstract submitted to the ECMWF-ESA Workshop on Machine Learning for Earth System Observation and Prediction. We compare data-driven models based on supervised convolutional neural networks and conditional generative adversarial networks for simulating precipitation fields.
+  * May Deep Learning contribute to improve Downscaling techniques? ({{:library:external:pen_a-izquierdo_downscalingwithdl_isms2020.pdf |poster}}). Presented at the ISMSM 2020 conference. We implemented encoder-decoder networks for the task of downscaling, that is taking climate variables from a low resolution simulation (predictors) for training an deep neural network to produce an observational gridded field at higher spatial resolution.
-A. Kahira in collaboration with L. P. Caron, worked on the application of convolutional neuronal networks to Hurricanes and cyclones observational databases to predict the number of such extreme events for the subsequent years.
+  * Deep neural networks as surrogate models for creating or improving data-driven parameterizations. Collaboration with Leonie Esters (Uppsala University) focusing on sea-air gas exchange and ocean turbulence.
-==== Earth Sciences Services ====
+  * Inference of surface pollutant concentrations from satellite and global scale meteorological and chemical simulations (AQ-WATCH project). Application to NO2 (TROPOMI sensor) and aerosols.
-Statistical correction of air quality forecasts with machine learning algorithms (ML4AQ grant). Application to Mexico City.
+  * Identification of Wastewater CH4 Emission Sources Using Deep Learning Applied to Sentinel-2 Observations. Abstract submitted to the ESA phi-week 2020 conference. Here we train object detection models from the field of computer vision for the task of localizing wastewater treatment plants and inferring the amount of processed water (as a proxy for CH4 emissions).
-==== Computational Earth Sciences ====
-  * Extreme Events (Hurricanes/TCs). Deep Neural Networks from the field of computer Vision can be used for pattern recognition in a supervised way, exploiting reanalysis data and historical event databases (IBTrACS for TC). The idea here is to obtain segmentation masks with the identification of specific events and with the possibility of tracking their future state. The problem addressed here is similar to the one tackled by A. Kahira, but re-framing the goal (which entails using different networks) and using data with a finer temporal sampling.
-  * Deep Neural Networks as surrogate models (learning to replace parameterizations). We explore the use of encoder-decoder networks, such as those used for image translation in Computer Vision, for the task of learning precipitation fields from other predictor variables (geopotential, SST, etc). In a way what we do is to learn a transfer function, capturing information from predictor variables to produce precipitation, a variable highly dependant on parameterizations schemes.

User Tools

Site Tools

Differences

Page Tools