<div><br>Air pollution forecasting can be used to alert about dangerous health effects caused by airborne pollutants and, in consequence, to take  actions to reduce pollutant concentrations (i.e reducing traffic, control industrial activities, etc..). Therefore, the development of reliable  air quality forecast systems is a of great interest.<br><br>The system consist of two main branchs. A statistical method based on  Neural Networks is used to forecast (10 days) several dayily air quality <br>index at the sites were historical data is available (i.e. pollution  measurement stations). A dynamical method based on WRF-CHEM to forecast hourly (48h) values of a large variety of species in a high resolution  domain (2km). Both subsystems use GFS and ECMWF forecasts as driving  conditions. The  dynamical subsystem incorporates 4DVAR data assimilation  of meteorological data (first 12 hours of forecast), and dynamical  emissions. The dynamical  emissions consist in changing the emissions of  large factories and trafficc. The emissions data are obtained by machine  learning methods based on historical series and meteorological conditions (mainly big energy factories). The WRF-CHEM configuration consist of several domains one way nested. The mother domain covers the entire Saharian desert in order to incorporante the dust transport contribution to particulate matter concentration. In addition, the base emission data is continuously updated.    The system also incorporates a module for automatic verification by comparing forecast with observed data, and analysis runs (in order to minimize meteorological forecast uncertainty). This verification process permit us to construct a MOS (Model Output statistics) in order to correct <br>possible model bias.</div>

<p>Due to anthropogenic global warming since the pre-industrial era, sea level has been rising along with global temperature. This sea-level rise is due to thermal expansion of the ocean and melting of mountain glaciers and continental ice sheets, mainly Greenland (GrIS) and Antarctica (AIS). The latter are the potential largest contributors as they store a total amount of 63 meters of sea-level rise in the form of ice. Modelling studies agree that these ice sheets will melt more in the future, however results differ due to associated uncertainty in representing several physical processes, as well as in assessing warming projections. Past warm scenarios can help to elucidate this uncertainty as we can obtain information, such as the sea-level standings, the ice extension from continental ice sheets and infer global temperatures from proxy records. The mid-Pliocene warm period (3.3-3.0 million years ago) offers an ideal benchmark, as it is the most recent period with CO2 levels comparable to the present-day (PD; 350-450 ppmv), although showing global mean temperatures 2.5-4.0 degrees higher. The inferred sea-level reconstructions from that period estimate a sea level standing of 15-20 meters higher than PD. Whereas the modern GrIS was starting to form, the AIS was restricted to its eastern region due to warm oceanic temperatures. The Pliocene Model Intercomparison Project, Phase 2 (PlioMIP2) has brought together various climate outputs from different general circulation models to elucidate the pliocene climate conditions. Here we force a higher-order ice sheet model with these climatic outputs at a high spatial resolution. Our aim is to investigate how polar continental ice sheets respond to these different climatic fields and to infer tipping values that can lead these ice sheets to drastically change their topographic shape.</p>