Ανακοίνωση

Collapse
No announcement yet.

Συνέντευξη με Dr Peter Bauer, αναπληρωτή διευθυντή έρευνας του ECMWF?

Collapse
X
  • Φίλτρα
  • Ώρα
  • Show
Clear All
new posts

  • Συνέντευξη με Dr Peter Bauer, αναπληρωτή διευθυντή έρευνας του ECMWF?

    Tον ερχόμενο Μάιο θα έχω την ευκαιρεία να συναντήσω τον Peter Bauer, αναπληρωτή διευθυντή έρευνας του ECMWF, ο οποίος αποδέχτηκε πρόσκληση σε ένα επιστημονικό συνέδριο που εμπλέκομαι. Οπότε θα πρότεινα να ετοιμάσουμε κάποιες ερωτήσεις (ως snowreport forum!) και να του τις μεταφέρω (ηλεκτρονικά) και να δούμε αν θα δεχτεί να της απαντήσει. Εγώ σκοπεύω να ετοιμάσω κάποιες ερωτήσεις ούτως ή άλλως (περί μοντέλων, απόδοσης, παρατηρήσεις, κλπ.) αλλά αν υπάρχει ενδιαφέρον θα μπορούσαμε να τις μαζέψουμε εδώ και να της μεταφέρω. Περιμένω αρχικά αντιδράσεις και αν υπάρχει ενδιαφέρον μαζεύουμε εδώ και τις ερωτήσεις.
    Facebook group: https://www.facebook.com/groups/pefteixioni/
    Llive camera Δροσιάς στα ~400μ με θέα τους λόφους της Ρέας στους πρόποδες της Πεντέλης Αττικής: http://www.pefteixioni.com/fcamera/fcamera.html
    Live camera Δροσιάς μέ θέα Πάρνηθα (τελεφερίκ-Καζίνο και Κρυονέρι) http://www.livecameras.gr/up/drosia.php

  • #2
    Ωραιος Φωταρα μακαρι να σου απαντησει στις ερωτησεις που σιγουρα θα ειναι πολλες.Εχω μια ερωτηση η οποια βεβαια δεν ξερω αν ειναι σωστη εσυ θα ξερεις καλυτερα.Αν ειναι να του τη θεσεις.Η ερωτηση λοιπον ειναι εφοσον το ECMWF χρηματοδοτειται απο τοσες Ευρωπαικες χωρες μεταξυ αυτων και η Ελλαδα γιατι δεν ειναι full accessible σε εμας τους απλους θνητους οπως το GFS και θελει συνδρομη;
    Kaiser vs χιονοζομπι

    Σχολιασμός


    • #3
      μια καλη ερωτηση με δυο σκελη θα ηταν...σε πιο σημειο της ευρωπης θεωρει ο ιδιος οτι κενο το ευρωπαικο και ειναι το αδυνατο σημειο του...και στο καπακι σαν συνεχεια σε πιο ειδος κυκλοφοριας χωλενει πιο πολυ.μεσημβρινη ζωνικη με γροιλανδικο με σβηρικο κτλ κτλ κτλ...
      ΕΡΧΕΤΑΙ Ο ΓΙΑΓΚΟΣ ΔΡΑΚΟΣ...ΣΥΝΤΟΜΑ ΣΤΟΥΣ ΔΡΟΜΟΥΣ ΜΑΣ

      Σχολιασμός


      • #4
        Αρχική Δημοσίευση από notios_polos Εμφάνιση μηνυμάτων
        μια καλη ερωτηση με δυο σκελη θα ηταν...σε πιο σημειο της ευρωπης θεωρει ο ιδιος οτι κενο το ευρωπαικο και ειναι το αδυνατο σημειο του...και στο καπακι σαν συνεχεια σε πιο ειδος κυκλοφοριας χωλενει πιο πολυ.μεσημβρινη ζωνικη με γροιλανδικο με σβηρικο κτλ κτλ κτλ...
        Επισης αν πιστεύει ότι περισσοτερο χιόνι βάζει το Φίλυρο η το Ωραιόκαστρο .

        Συνελθε.

        Σχολιασμός


      • #5
        Μπράβο ρε Φώτη...πολύ καλό αυτό...πολλά συγχαρητήρια για την προσπάθειά σου....τέτοιες ενέργειες αξίζουν στην μετεοκοινότητά μας...

        Πιθανή ερώτηση θα ήταν και το πότε θα γίνει αναβάθμιση στο ευρωπαικό μοντέλο και τι θα σήμαινε αυτό για την πρόγνωση......(είχε ακουστεί καποια στιγμή ότι θα κάνουν αναβάθμιση αν θυμάμαι καλά αλλά δεν θυμάμαι το πότε..)
        Μετεωρολογικός Σταθμός Μακροχωρίου:
        http://penteli.meteo.gr/stations/veroia/
        Κάμερες Μακροχωρίου:
        http://penteli.meteo.gr/stations/veroia/makrohori.jpg
        http://www.stravon.gr/makrochori/

        Σχολιασμός


        • #6
          Αρχική Δημοσίευση από vasilis78 Εμφάνιση μηνυμάτων
          Ωραιος Φωταρα μακαρι να σου απαντησει στις ερωτησεις που σιγουρα θα ειναι πολλες.Εχω μια ερωτηση η οποια βεβαια δεν ξερω αν ειναι σωστη εσυ θα ξερεις καλυτερα.Αν ειναι να του τη θεσεις.Η ερωτηση λοιπον ειναι εφοσον το ecmwf χρηματοδοτειται απο τοσες Ευρωπαικες χωρες μεταξυ αυτων και η Ελλαδα γιατι δεν ειναι full accessible σε εμας τους απλους θνητους οπως το gfs και θελει συνδρομη;

          Συμφωνώ με το παραπάνω. Για ποιο λόγο δεν υπάρχει ανοιχτή πρόσβαση από όλους στα δεδομένα του ecmwf όπως τα δίνει το gfs και άλλα μοντέλα;

          Σχολιασμός


          • #7
            Στο weather.us ανοιχτα ειναι τα δεδομενα του ευρωπαικου
            ΧΙΟΝΙΑ Ε?????

            Σχολιασμός


            • #8
              Ιδού η πρώτη προσπάθεια για τα σχόλιά σας (άργησα να το ετοιμάσω και θα τον ρωτήσω αύριο από κοντά). ΖΗτώ συγγνώμη για τα Αγγλικά αλλά δε γίνεται να το μεταφράσω όλο (μπορείτε να χρησιμοποιήσετε μεταφραστή):
              Interview with Dr. Peter Bauer, Deputy Director of Research, ECMWF
              14.5.2018
              [1]
              Feel free to comment on/correct the questions in case there are inaccuracies:
              1. Business model/Data Policy: One major difference between the ECMWF/IFS and the American NOOA/NCEP/GFS model is that the data (and also the software) in the latter are freely accessible, while the European ones are closed (can be obtained for a fee). As a consequence, the vast majority of weather apps uses the latter (American model), which based on medium range performance verification statistics is inferior compared to ECMWF. What is the reasoning behind this? Given the new scientific movement towards open science/open data/open to the world, also clearly supported by Europe (as Commissioner Moedas has stated) are new business models being considered? If not, why?
              2. Global models’ performance:
              i. Verification statistics: First of all, congratulations, as ECMWF has been steadily leading the race with other global models in terms of medium range verification statistics (also based on the ones coming from the GFS model). What is the secret for that? Is it the model itself and related algorithms-equations (e.g. non-hydrostatic vs. hydrostatic), more accurate initial conditions all over the world, finer spatial resolution, other?
              ii. ECMWF IFS uses 50 ensembles while GFS uses 20. Does this offer a benefit and has it proven so? Or it may be true that 20 ensembles may be good enough and the added-value is not that big? Are there so many different combinations in changing the initial conditions that it is worth it?
              iii. The UK Met Office is running in lower (worse) spatial resolutions than other models and has been renowned for having good performance/accuracy at the 500Hpa (second place behind ECMWF, although it hasn’t been a very good year for them). Is it true that if the spatial resolution is too high, it may be worse in terms of accurate prediction? Or is it simply a myth?
              iv. Following this up, what is your assessment of regional models, e.g. WRF, which are based on other global ones? It appears that if the global data for the region is not accurate, then the regional model will also fail. In case the global data are good enough, the regional ones will simulate better some parameters, including the precipitation, based on finer resolution and topography. On the other hand in several cases the regional models overestimate the precipitation. What is your overall assessment?
              v. There was a story that when the GFS used the initial conditions from ECMWF it was almost as good (in terms of verification statistics) as ECMWF? Is this true? Is there collaboration with the Americans for such or other activities?
              vi. From monitoring the numerical weather models for several years, it appears that ECMWF models the North Atlantic (super)low-pressure systems much better than GFS does (despite the fact that ECMWF is a hydrostatic model and is supposed to perform better in more complex topography, which is not the case in the North Atlantic sea). Is it true and if so, is there a reason for that? On the other hand, with some very long-wave troughs from Russia-Siberia all the way to Europe (back-door), GFS appears to be doing better. Is it true and is there a reason for that? Are there types of weather circulations that one model is doing better than the other? E.g. that ECMWF is doing better with zonal circulation (because it may model the North Atlantic lows better)?
              vii. We know that weather is global and it is difficult to split the map and monitor the performance verification in smaller regions or even countries. Usually it is something like Northern Hemisphere (between this and this latitude etc.) It is already difficult to find statistics for Europe. WMO has Europe as a region but only a few models appear there and comparisons are not easy. Are there any efforts to be able to go into country verification statistics in the future or this will never happen? It appears that Greece is quite a difficult country for forecasts (with a lot of mountains in the main land), especially for predicting long wave troughs. Is this normal given its difficult terrain and also sea around.
              3. Infrastructure:
              a. It was decided to build the new ECMWF center by 2019 in Bologna. What can we expect from the new data center? Better energy efficiency, higher resolutions, and ultimately better performance statistics? Are there any pilot efforts (e.g. towards exascale) going into production soon? We have been hearing about redesigning the code? Given the fact that weather modelling is tightly-coupled (high inter-process communications), is it possible to do major changes?
              b. Upgrade: When is the next upgrade of the ECMWF model going to take place?
              4. Long term forecasts:
              a. What do you think about long-term forecasting? For next month or even next season? Practically after 4-6 days, the forecasts, especially during winter, are not that good. There are several entities providing trends in the long term (Accuweather, Italian sites, and I think ECMWF has also some research efforts). And it appears that CFS seems to be doing relatively well for next month trends. Do several indices (QBO, ENSO, NAO, AO, etc.) help the research observations in this regard? What is your overall assessment? Is there a chance that in the future long-term forecasts will be more reliable?
              5. Citizen scientists-amateur meteorologists: is there a program for using the knowledge-experience of citizen scientists-amateur meteorologists inside ECMWF?


              [1] Some of the questions have been initiated in the Snowreport Forum Greece

              Facebook group: https://www.facebook.com/groups/pefteixioni/
              Llive camera Δροσιάς στα ~400μ με θέα τους λόφους της Ρέας στους πρόποδες της Πεντέλης Αττικής: http://www.pefteixioni.com/fcamera/fcamera.html
              Live camera Δροσιάς μέ θέα Πάρνηθα (τελεφερίκ-Καζίνο και Κρυονέρι) http://www.livecameras.gr/up/drosia.php

              Σχολιασμός


              • #9
                Όλα καλά, μου απάντησε στα παραπάνω και τα είπαμε και από κοντά και τώρα προσπαθώ να πάρω άδεια να δημοσιεύσω αν γίνεται ολόκληρη την συνέντευξη.
                Facebook group: https://www.facebook.com/groups/pefteixioni/
                Llive camera Δροσιάς στα ~400μ με θέα τους λόφους της Ρέας στους πρόποδες της Πεντέλης Αττικής: http://www.pefteixioni.com/fcamera/fcamera.html
                Live camera Δροσιάς μέ θέα Πάρνηθα (τελεφερίκ-Καζίνο και Κρυονέρι) http://www.livecameras.gr/up/drosia.php

                Σχολιασμός


                • #10
                  Λοιπόν επισυνάπτω την συνέντευξη εδώ. Πρώτα είχε δώσει παρουσίαση: http://e-irg.eu/documents/10920/4217...IRG_052018.pdf
                  Να πω ότι στην πρώτη ερώτηση προφορικά πρόσθεσε ότι λογικά στα επόμενα 5-10 χρόνια τα κράτη-μέλη θα ανοίξουν τα δεδομένα.

                  [1]
                  1. Business model/Data Policy: One major difference between the ECMWF/IFS and the American NOOA/NCEP/GFS model is that the data (and also the software) in the latter are freely accessible, while the European ones are closed (can be obtained for a fee). As a consequence, the vast majority of weather apps uses the latter (American model), which based on medium range performance verification statistics is inferior compared to ECMWF. What is the reasoning behind this? Given the new scientific movement towards open science/open data/open to the world, also clearly supported by Europe (as Commissioner Moedas has stated) are new business models being considered? If not, why?
                  Answer: ECMWF’s data policy is the result of the data governance model agreed by its Member States. This policy is continually being reviewed and amended. Beyond medium-range forecasts, many ECMWF products are actually free, for example, the seasonal forecasts through the Copernicus Climate Change Service (C3S), reanalyses, and ECMWF also provides data free of charge for research experiments (e.g. Year of Polar Prediction), field campaigns and in case of national weather related emergencies outside Europe (e.g. past floodings in Bangladesh). Many apps include options for adding ECMWF data for a comparably small fee.
                  2. Global models’ performance:
                  a. Verification statistics: First of all, congratulations, as ECMWF has been steadily leading the race with other global models in terms of medium range verification statistics (also based on the ones coming from the GFS model). What is the secret for that? Is it the model itself and related algorithms-equations (e.g. non-hydrostatic vs. hydrostatic), more accurate initial conditions all over the world, finer spatial resolution, other?
                  Answer: The ECMWF forecast performance is the results of a combination of facts, not of a single cause. It is the combination of an excellent analysis system that exploits the information from about 40 million observations per day – mostly from satellites – and an effective optimization algorithm (4D-Var) that also needs an accurate forecast model to perform well. ECMWF has been able to develop these systems over time into a well-adjusted overall system. ECMWF also benefits from a tight collaboration between research and operations that enables a fast transition of new science into operational performance. This also benefits from significant HPC resources being made available to research (about 50% of total).
                  b. ECMWF IFS uses 50 ensembles while GFS uses 20. Does this offer a benefit and has it proven so? Or it may be true that 20 ensembles may be good enough and the added-value is not that big? Are there so many different combinations in changing the initial conditions that it is worth it?
                  Answer: The value of more ensemble members has been shown in the past and publications about this topic can be made available. Ensemble generation requires the representation of initial condition and model errors in the perturbations that generate spread. The fact that the Earth-system is a very non-linear system with billions of degrees of freedom indicates that even 50 members are likely insufficient. This is particularly true for the prediction of extremes (one of the headline missions of ECMWF), which requires a good sampling of the tails of probability distributions.
                  c. The UK Met Office is running in lower (worse) spatial resolutions than other models and has been renowned for having good performance/accuracy at the 500Hpa (second place behind ECMWF, although it hasn’t been a very good year for them). Is it true that if the spatial resolution is too high, it may be worse in terms of accurate prediction? Or is it simply a myth?
                  Answer: Spatial resolution always helps because it allows to represent key physical processes with more accuracy. This becomes visible in areas of strong land-sea, or sea ice–ocean contrasts, large topographical variability, in the presence of strong horizontal gradients (like fronts, tropical cyclones) and in the presence of small-scale weather events (like summer storms, orographic precipitation). Worse performance can appear due to the double-penalty effect (i.e. RMSE of a parameter that varies spatially because mislocation counts twice) or if model resolution upgrades reveal error compensation in model parameterizations. The latter requires readjustments.
                  d. Following this up, what is your assessment of regional models, e.g. WRF, which are based on other global ones? It appears that if the global data for the region is not accurate, then the regional model will also fail. In case the global data are good enough, the regional ones will simulate better some parameters, including the precipitation, based on finer resolution and topography. On the other hand, in several cases the regional models overestimate the precipitation. What is your overall assessment?
                  Answer: Regional model performance depends on the boundary conditions provided by the global model, the regional initial conditions (if applied) and regional model quality. As in any operational prediction system, performing well 24/7 with a regional model requires a well-adjusted system and a good regional analysis for providing initial conditions that do not produce too much imbalance with the external forcing.
                  e. There was a story that when the GFS used the initial conditions from ECMWF it was almost as good (in terms of verification statistics) as ECMWF? Is this true? Is there collaboration with the Americans for such or other activities?
                  Answer: These experiments have been performed in the past but not necessarily with sufficient statistics to produce general conclusions. Evaluations like this need to look at the entire range of verification measures that are used at operational centres, for example prior to the implementation of the next-generation model version. Only then conclusions on the relative contribution of analysis vs forecast to overall skill can be drawn.
                  Generally, the meteorological community is very well organized with respect to scientific and operational collaboration, data exchange and strategic planning in support of future systems. Beyond its tight interaction with its Member States, ECMWF has bilateral agreements with, among others, the US National Weather Service.
                  f. From monitoring the numerical weather models for several years, it appears that ECMWF models the North Atlantic (super)low-pressure systems much better than GFS does (despite the fact that ECMWF is a hydrostatic model and is supposed to perform better in more complex topography, which is not the case in the North Atlantic sea). Is it true and if so, is there a reason for that? On the other hand, with some very long-wave troughs from Russia-Siberia all the way to Europe (back-door), GFS appears to be doing better. Is it true and is there a reason for that? Are there types of weather circulations that one model is doing better than the other? E.g. that ECMWF is doing better with zonal circulation (because it may model the North Atlantic lows better)?
                  Answer: The performance of a forecast model over the North Atlantic may very well be affected by orography in regimes with stronger zonal flow. This is because location and intensity errors in representing troughs occurring leeward of the Rocky Mountains propagate downstream and amplify. The large-scale interaction between land and sea, orography, troposphere and stratosphere in certain situations, but also the propagation of waves with tropical/sub-tropical origin into the mid latitudes is complex, and the performance of forecasting systems depends on how well these components and their interactions are represented in the model and the initial conditions. The role of non-hydrostatic processes is likely to be small in this context.
                  g. We know that weather is global, and it is difficult to split the map and monitor the performance verification in smaller regions or even countries. Usually it is something like Northern Hemisphere (between this and this latitude etc.) It is already difficult to find statistics for Europe. WMO has Europe as a region but only a few models appear there, and comparisons are not easy. Are there any efforts to be able to go into country verification statistics in the future or this will never happen? It appears that Greece is quite a difficult country for forecasts (with a lot of mountains in the main land), especially for predicting long wave troughs. Is this normal given its difficult terrain and also sea around.
                  Answer: There are more general and large-scale (agreed by WMO Member States in support of forecasting system evaluation standard) than regional verification measures, but many weather services produce their own country-based assessments as well. ECMWF receives some of this from its Member States and it is a very welcome source of verification information of specific forecast performance in selected areas. Particularly areas with difficult local conditions (orography, land-sea, small scale weather events and extremes, strong seasonal variability etc.) provide helpful input to ECMWF’s model development. For this purpose, ECMWF is hosting dedicated workshops with its Member States on a regular basis.

                  [1]
                  Some of the questions have been initiated in the Snowreport Forum Greece
                  Facebook group: https://www.facebook.com/groups/pefteixioni/
                  Llive camera Δροσιάς στα ~400μ με θέα τους λόφους της Ρέας στους πρόποδες της Πεντέλης Αττικής: http://www.pefteixioni.com/fcamera/fcamera.html
                  Live camera Δροσιάς μέ θέα Πάρνηθα (τελεφερίκ-Καζίνο και Κρυονέρι) http://www.livecameras.gr/up/drosia.php

                  Σχολιασμός


                  • #11
                    Και το υπόλοιπο
                    3. Infrastructure
                    :
                    a.
                    It was decided to build the new ECMWF center by 2019 in Bologna. What can we expect from the new data center? Better energy efficiency, higher resolutions, and ultimately better performance statistics? Are there any pilot efforts (e.g. towards exascale) going into production soon? We have been hearing about redesigning the code? Given the fact that weather modelling is tightly-coupled (high inter-process communications), is it possible to do major changes?
                    Answer: The reasons for ECMWF’s next datacenter being located in Bologna are that the current center will not provide sufficient space and power supply for future HPC facilities. This growth in HPC requirements will allow ECMWF to invest in spatial resolution, model complexity, ensembles, better analysis systems ingesting more observational data and serving more diverse demands from its Member States as well as the Copernicus services.
                    Towards exascale computing, ECMWF has invested in the so-called Scalability Programme (
                    https://www.ecmwf.int/en/about/what-we-do/scalability
                    ) that aims to fundamentally revise the entire forecasting system and to cope with the substantial increase in computing and data handling required for fulfilling its strategy. This programme is strongly supported by ECMWF’s Member States. It is also supported by several EC-funded projects like ESCAPE, ESCAPE-2, ESiWACE, NextGenIO, EuroEXA, MAESTRO, EPiGRAM-HS, which ECMWF is coordinating or in partnership with. ECMWF is also coordinating a proposal for a preparatory action for a future Flagship initiative (under H2020 FETFLAG) called ExtremeEarth (
                    www.extremeearth.eu
                    ). We believe that true exascale performance needs a substantial effort that goes well beyond ECMWF and that needs to involve the entire weather and climate prediction community.
                    b.
                    Upgrade: When is the next upgrade of the ECMWF model going to take place?
                    Answer: ECMWF performs forecasting system upgrades about once per year and the next upgrade will be implemented later this year.
                    4. Long term forecasts:
                    a.
                    What do you think about long-term forecasting? For next month or even next season? Practically after 4-6 days, the forecasts, especially during winter, are not that good. There are several entities providing trends in the long term (Accuweather, Italian sites, and ECMWF has also some research efforts). And it appears that CFS seems to be doing relatively well for next month trends. Do several indices (QBO, ENSO, NAO, AO, etc.) help the research observations in this regard? What is your overall assessment? Is there a chance that in the future long-term forecasts will be more reliable?
                    Answer: ECMWF is issuing forecasts up to day 46 twice a week and seasonal forecasts up to month 7 once a month. These are extended to the 12-month range every three months. In general terms, seasonal forecasts exhibit their best performance in lower latitudes but also show skill at higher latitudes depending on the prevailing weather regimes. Due to the complex interaction of many processes at global scale and over such long time scales seasonal forecasts aim to predict anomalies from mean seasonal variability rather than individual weather events. A good example is ENSO that, among others, exhibits sea-surface temperature anomalies in the tropical oceans (Pacific) that affect weather regimes worldwide. As for the medium range, seasonal forecasts are run as ensembles to provide forecast uncertainty estimates, and improvements of seasonal predictive skill and reliability will improve with spatial resolution, Earth-system complexity being better represented in models, better initial conditions (land surface, ocean, stratosphere), and ensemble formulation.
                    5. Citizen scientists-amateur meteorologists: is there a program for using the knowledge-experience of citizen scientists-amateur meteorologists inside ECMWF?
                    Answer: ECMWF makes available a comprehensive workshop, seminar and training programme for scientists, but there is no specific effort to engage with amateur scientist to date. Through its Member States and commercial users ECMWF receives regular feedback on forecast performance. However, citizen based observations from commodity devices such as mobile phones will become important in the future by supplying data with unprecedented sampling. This may provide a forum for more exchange between weather prediction centres and the public.
                    6. Climate change: What is your opinion on this? We certainly observe huge ice areas gradually disappear, but it may be true that over the centuries, such changes did happen in the past. And is the human intervention in the planet causing these changes?
                    Answer: My personal opinion follows the conclusions from the vast majority of climate scientists, and this is documented by the IPCC assessments. The IPCC assessment states that climate change (warming of the troposphere, sea-level rise, glacier melting etc.) is extremely well documented based on observations, that the main causes of climate change are understood despite uncertainties around selected mechanisms, and that it is extremely likely that human influence has produced the climate change effects that we have been observing over the recent past.
                    Facebook group: https://www.facebook.com/groups/pefteixioni/
                    Llive camera Δροσιάς στα ~400μ με θέα τους λόφους της Ρέας στους πρόποδες της Πεντέλης Αττικής: http://www.pefteixioni.com/fcamera/fcamera.html
                    Live camera Δροσιάς μέ θέα Πάρνηθα (τελεφερίκ-Καζίνο και Κρυονέρι) http://www.livecameras.gr/up/drosia.php

                    Σχολιασμός


                    • #12
                      Facebook group: https://www.facebook.com/groups/pefteixioni/
                      Llive camera Δροσιάς στα ~400μ με θέα τους λόφους της Ρέας στους πρόποδες της Πεντέλης Αττικής: http://www.pefteixioni.com/fcamera/fcamera.html
                      Live camera Δροσιάς μέ θέα Πάρνηθα (τελεφερίκ-Καζίνο και Κρυονέρι) http://www.livecameras.gr/up/drosia.php

                      Σχολιασμός

                      Working...
                      X