 Our first keynote this morning is by Attila Lazar, who's a research fellow at the University of Southampton, and his title is as you see up here. Attila, yeah, how to run this, but did you hear us? Okay, so you got it? Yeah, I got it, thank you. Well, good morning, everyone. During the last two days, we have heard many interesting presentations about complex system modeling and complex issues. Well, today I would like you to introduce to another kind of complex system modeling, which is gonna be quite different from the previous presentations in many aspects. One aspect is that we are working with multidisciplinary research teams and try to integrate the results together. And also that this project is ongoing currently, so we are just halfway through, so I can't provide you final results. I can only give you the vision, the tools that we are using, and what is gonna be the output, why we are doing all this hustle around this project. So that's the outline of my presentation. So I'm very briefly talking about the project and the integration names, showing some results, talking a little bit about uncertainty within the integrative tool and model testing and just wrap up very quickly. So this is a four year long project financed by the UKA, the UK Environmental Research Council, and the Economic and Social Research Council. The aim is to provide the novel agent tools to the Bangladesh decision makers to make more informed decisions. So this is a study is in the present. We're trying to look into the near future, so next 50 years, and try to do some policy scenarios within the work. We have a large consortium to do this ambitious job. We have 20 some partners from three different countries, and on a regular Bangladeshi consortium meeting, we have 60, 70 participants normally, so it's a large group to work with. So as I said, we are located in coastal Bangladesh, the tidal influence part of the Ganges-Brahma-Putscha-Magnar river data plane, and as a result, we have a focus on the coastal zone, but to be able to tell anything useful about the coastal zone, we have to really consider what is outside of the study area. So we are considering exogenous drivers where the Bangladeshi decision makers have no influence on, or at least very little influence on, like upstream floater version in China and in India, or climate change, some macroeconomic changes. We also consider the upstream river basin to provide boundary conditions to the study area, so we are doing detailed hydrological and sediment transport modeling. Also colleagues working on the Bay of Bengal modeling, what's going on with the tidal ranges, sea level rise, fishery productivity. When we do, yeah, sorry, I forgot to mention the Andagina's governance, which is the national scale of the project, that's where the Bangladeshi decision makers can make a difference too. So what policies they create, how subsidies are formulated, what they support, what is the flood protection initiatives, infrastructure development, so these aspects are considered to some extent in this project. And when we zoom down to the local scale to the study area, really we are interested in morphodynamic changes, land use, land cover changes, and also the changes of the productivity of the system, so the relationship of the environment, how that changes over time and space. But we don't stop at the biophysical environment, we are really would like to couple what is happening in the environment together with what is happening in the social system, how demographic changes are likely to take place, what are gonna be the likely market changes, and how security and livelihood and well-being are changing at the rural population, at household level. This slide is really just to point out two things, that governance is in working together with every project element, and from day one we have a stakeholder engagement with national level decision makers, they are working together on creating issues, what they are interested in, developing scenarios that we are test with the models, and when this integrative tool is developed, then we are gonna have an iterative learning loop together with the stakeholders, we try to learn together about the system, how it works. So just very briefly about the project elements because just to show you how multidisciplinary this project is, the first pillar of the project is governance research. Our lawyers from Dundee University assessing the laws and policies in Bangladesh, what are the gaps, implementation efficiencies, conflicts in between legislations, also working together with the stakeholders or during the entire project period. And then the second pillar is the social science, how I call it demographics, household economics bit. We collect primary data at household level, qualitatively and quantitatively about how they live. We also create population projections and create a statistical associative model that I will introduce a bit more detail later. And all these feed into a dynamic and quantitative understanding of what is happening on ground in terms of people and households. And the third pillar is the biophysical modeling, practically the environment and entire system, starting from climate models, projections, hydrological sediment transport models, modeling the Bay of Bengal, modeling the data plane with F3 command F3D. Those feed into the morphodynamic analysis and the land cover model and all these together affect the productivity models that we are applying in this project. But we don't stop here, obviously, we would like to integrate this knowledge together. So it is a very ambitious aim to bring these disciplines together, working at different scales, different timescales, spatial scales and create something useful, useful information for national level decision makers. So I just would like to highlight that we have a special focus. We have a hypothesis within this project that rural population in Bangladesh is highly depending on the ecosystem services. So the quality of the environment, they actively modify it and the quality is affecting their livelihood. So that's our hypothesis that we would like to test with this integrative tool and predict what is gonna be the impact of any change in the system. The change can be governance decision or climate change or environmental change or a change in the behavior of the people. As I said, it's a large ambitious project. We have multiple working groups working on different elements of the system and they are working together. So they pass information to each other, mostly as a unilateral flow of information. But what we would like to create with this integrative model is a more rapid assessment tool that encapsulates all this research that is going on within this project and create a dynamic framework which allows a forward-stepping feedback in the system and which allows a more rapid and more numerous testing of different scenarios. Yeah, so this work is mainly building on the in-depth research of our colleagues and trying to encapsulate and make the calculation faster. So how we do this, we are gonna use simple process-based models if they are simple enough to include in this integrative framework. If they are too complicated, like the DAF 3D model results, then we are aiming to reduce the complexity to Bayesian emulators. So just capture if the inputs change, how the output is gonna change and that is gonna be used in the integrative tool. Just keep in mind, we are trying to say something about the cost of changes. So some of these models, in the integrative tool, we don't need the detail, why is happening in that, what causes the change? We just need to know in the integrative tool if something change happens, how the output of that model is changing. So let me just briefly introduce this DeltaDM model framework, which stands for Delta Dynamic Integrative Emulator Model. It's designed to be a holistic tool and encapsulating both the biophysical environment, social behavior, livelihoods, and some governance issues. It's a meta-model which practically just harmonizes the run of the different model elements together in an efficient way and a harmonized way. And the model elements are gonna work at different spatial and temporal scale and have different complexities. Some of them just statistical relationships, some of them probabilistic emulators and some of them, especially for the household level decision model, is probably gonna be an agent-based type model. So actually this DeltaDM model is gonna have two versions. One is what I call a hybrid version. When we estimate the environmental changes with process-based models, some demographic changes considered governance actions and try to estimate how land-used land cover would change and using statistical associative model directly estimate livelihood and poverty changes. So this is based on, I will talk about this later. And then the second approach is a more process-based version. When we don't stop here, but we apply productivity models which feed into a household level livelihood model and that will give us indication about the well-being of the whole population. Hopefully these two approaches will support each other, but if not, again, we can learn something from the differences. So we will see in a year's time what is happening. So this hybrid model, as I said, is using this statistical associative relationship. And this statistical model is trained actually based on land use, land cover information, shown here, environmental quality information like soil serenization and poverty information which is coming from the census data that is available for three years over the last 30 years. And this is a map created based on the census data. The red means that it is high poverty, green means that it is relatively better off. It doesn't mean that they are rich, just better off not necessarily at heart or poverty. So what our colleagues do in Southampton actually applying a number of statistical techniques to create relationships in between those two, actually those three, we also consider actually, I forgot to put it on, the infrastructure and network, road networks. So here are two examples. The first one is really self-explanatory. So it is the fact the relationship between irrigated land and poverty. It is not surprising that there is a linear relationship. The more you irrigate, the higher your productivity and therefore you are getting better off. The second is a bit more thought provoking. If you have high larger of mangrove area around you or if you live in cities, you tend to be poorer. If you are in between, you tend to be better off again. We don't know why this is just a pure statistical relationship in between different input and output variables. But these are just two examples. There are other on salinity and other aspects that we would like to combine in a statistical model that will be applied in this data VM framework. The more process-based model is simplified here. So we will have again, the hydrology sediment transport models. We will emulate the coastal inundation and salinization. That will feed into a land use land cover model together with the demographic changes. We run productivity models, household level assumptions and we will estimate human well-being. This is a simplified version. This is a bit more in depth what is gonna go in going in this model. But there's not much time for that. So rather, I will show you some expected output. What kind of information we are gonna get out when the model is fully operational. So first is the demographic changes. We will be able to track in space and time how people move in between districts, how population changes in terms of sex and age group, distribution of the population. So this calculation is actually done at district level, we have nine districts, but the results are gonna be downscaled to union level, which is the smallest planning unit in Bangladesh, having an average surface area of about 26 square kilometers. And we have 655 unions within our study area. You will see some maps about that in a minute. Like here, so these are the unions and the output of the model is gonna be at union level. We not necessarily present the results to stakeholders at this level, but the calculations are done monthly and at union level in the more process-based approach. So what you can see here, actually it is the changes of crop productivity under just one scenario. If there is no irrigation and the soil salinization increases continuously over time, then you can see how the staple foods would change, the productivity of the staple food would change in space and time. We have about 36 different crops in our crop library, so we can do complicated cropping patterns with this model. But as I said, everything is preliminary. It has to be updated when the researchers, colleagues have updated, finalized their results. We can do household level indicators and indication what is going on with the people. So for example, we can do profit margin calculation that what is the fraction of the revenue that remains in your pocket when all the cost is paid. You can see that preliminary results shows a quite distinct difference in between regions of coastal Bangladesh. We can check this at agent types. So these are my farmer agents, large landowners, small landowners or shalcopers and landless laborers. Each of them have different characteristics and behavior how they operate. What you have to know here is one doesn't mean that they are rich, that means that they are okay. They don't have too much cost in relation to their income. When they close to zero, that means that they are in big trouble. So they need either loans, so loan schemes are gonna be important in this research and or they have to do multiple jobs to survive. And finally, some likely indicators that we are planning to include in this integrated framework output. These are just some of the indicators when on which my social scientist colleagues are working on. And these are the ones I have selected that are likely to be included in this framework as an output. So for example, income expenditure ratio, we are gonna have a routine to estimate the likely diet of a household based on wealth and background. And also based on we can estimate the calorie and protein intake and food insecurity and hunger periods over time because we are considering seasonal changes in the system. But there are also other indicators that we are thinking of. Some of them are just an empirical relationship developed based on primary data that we collect on the field. So uncertainty, well, as you can see, it's a complex model. It causes me a big headache how we are gonna do this formal uncertainty analysis of the model. But I have a simple plan so far that we are doing formal uncertainty analysis on each of the process based models, creating these tornado charts if they are not exist already. And we are gonna have an automated built in Monte Carlo sequence in the integrative framework. So when the integrative model run once, actually this crop model, crop productivity model is gonna be run 16 times varying those most important parameters and estimating the uncertainty around the mean output of the model. When we have a more complicated model like the F3D, as I said, we will simplify it to an emulator. And when we create the emulator, we keep track of the error around the mean output of that emulator. And again, that mean and the mean value, the error and the mean value is passed on in the model chain. And we're counting all the uncertainties all along the model chain. So when you have stakeholder agent types, you always have an envelope of response and an envelope of livelihood indicators. Validation is another key issue. We have known for a long time that we never be able to validate environmental models and no mere askation. Not if you are familiar with this paper. It's a theoretical paper, but it's worth reading if you're not aware of it. She argues that in life sciences, you will never be able to fully test and fully validate and fully verify your models. And this is certainly true for our case, which is not just pure biophysical processes governed by physics, but we also have people inside in. But still, we have to do some kind of validation and testing of this model. So the first and most obvious step is checking the code and then checking each model element separately. So productivity tools like the agriculture crop productivity tool, checking it against published lens suitability maps, published farmers yield data in statistical yearbooks and check if your results are resembles the observed pattern. If this is done, or for all the elements, we are moving on to the entire model, testing it in a full extent and checking more global variables like land cover changes, poverty levels, inundation as a result of a historical floodiment. So these testings are on the historical period. And the final step that we would like to do is really compare the hybrid and the process-based results because the hybrid statistical model is 100% representative for the study area. The census data is 100% representative. If they support each other, that's good. If not, we are trying to learn what makes the difference. And these results actually are discussed with the stakeholders. So every time we produce a result, we take back to the stakeholders for discussion and in the policy cycle, we will allow them to test ideas how they would tackle with some of the results that we have in the modeling. So we try to learn together with the stakeholders how the system works. But why do we do this? So this last slide that I was planning to show you, just give an indication what are the questions that we would like to answer and assess with this integrative tool and with all the in-depth research about what the colleagues are doing in our project. And to remind you with this integrative tool, we are not aiming for 100% accurate to three precision results. We are aiming just to give trends, likelihoods and robustness of different changes and different governance decisions on the different elements of the system. So these are just some of the questions that we had in mind when we started the project. What will be the extent of inland flooding on a hypothetical events, storm surge or river flooding? Where will be the isoline for threshold salinity lie for different crops in the future if we assume different scenarios, what will happen and different management interventions like more polders, less flow diversion or more flow diversion in India and so forth, and so forth, how productivity would change over time and space under different scenarios and how poverty levels would change in relation to productivity changes and environmental changes? What is gonna be the effect of further a more significant flow diversions and reduce sediment transport? Also, what does drive migration and where would you expect higher levels of migration in the study area? What would subsidies and remittances, whether do they make any difference to the poorest of the poor in coastal Bangladesh? And whether global commodity prices like diesel, rice, shrimp, do they have an effect on the lives and the environmental change in coastal Bangladesh? So these are just some of the questions. We have actually a four-page long document just listing different questions and grouping them into sections whether they are depending on just one discipline, so output of one research group or whether it requires a full integrative model to answer those questions. And finally, just a summary that we are aiming for a holistic generic tool which the methodology can be applied as well. Obviously, the details have to be updated. We are aiming to link the environmental change with livelihood change of the rural population in coastal Bangladesh. The in-depth research is still ongoing, so both the biophysical models are still being calibrated and tested. Hopefully we will have the final outputs by summer this year. The quantitative social research is just started and having the first results in the next month or so. So everything is ongoing and the integration is ongoing, so hopefully by the end of this year we will be able to start doing the simulations and running the full model. So the hybrid model actually is expected to be operational by November this year and the full process-based model early next year and we are hoping to start the policy cycle in March, April, next year in 2015. So that was it, a brief overview and the vision of our project and the elements of the project and I'm happy to answer any questions if you have. Thank you. Thanks, Attila. I have a couple of questions. First you said that my first question is about the uncertainty and you are doing these uncertainty and sensitivity analyses. Wondered how you intend to communicate that to both the Bangladeshi government as well as the stakeholders and if there's any sort of plan in place for how you might handle any pushback on the uncertainty that you produce in the models and the second question is about the stakeholders and once you go into the field and present your model results to them with or without the uncertainty, will you use their feedback as inputs or will you modify boundary conditions based on the stakeholders' comments? Yeah, difficult questions. So communicating uncertainty is always a difficult issue with stakeholders but we can't avoid. So your model result is as good as your input data and your assumptions. So how you can reduce it, you can partially reduce uncertainty if you work together developing your assumptions based on the stakeholders. We just had a stakeholder meeting three weeks ago in Bangladesh where we had about 80 participants from different ministries and national level organizations and research institutes and we are going through with all the assumptions that we make for business and usual in the future and we try to implement those assumptions like flow diversion rates or changing in diet habits of the people or every aspect. We have actually 104 items on the issues that the Bangladesh national level people identified that they are interested in. We can't model all of them. We try to include as much as we can in the model framework and using their knowledge on those issues, we are setting up the model. So that's how we work. So, but yeah, we have to communicate uncertainty because otherwise it can be misleading. Question is similar to the other one but I'm interested in the level of information particularly on that you're gaining and monitoring and particularly on local climate. And I note that the kind of precision that you're talking about for your interests is overlaps very much with mesoscale models as opposed to global models and getting information back and forth to the locals is very helpful for locals but also hugely helpful for those of us who are interested in understanding the climate and being able to get information back. So I'm curious what your method is of communicating data in that case. Communicating data to actually monitoring a long-term modern thing of any of the components that you have but also particularly of climate variables. Well, in terms of climate variables, we have a difficult situation because we were not able to get any climate data from Bangladesh. But we have a partner, the UK Met Office, and they provide us climate projections for both the historical period and for the future to 2100 actually, so it's much longer that's what we need. And we are using these actually to drive the models. So these are done-scaled regional model outputs of climate models. So we have selected three ensemble members of a 17-member ensemble and tailored to our needs. So those are the future flag changes and would highlight uncertainties. I'm not sure if I answered your question. We can talk about it later, but it has to do with monitoring, local monitoring of information, particularly taking advantage of cell phones and other ways of transferring information different scales. As I said, we don't use local monitoring data. We have a grid cell about 50 square kilometer and we have climate data at those cells. So that's what we use.