 Great. Yeah. So we have a great number of participants. So we're excited to have you all here for our second agroinformatics tech talk. So thank you for joining. These are a series of virtual but also hopefully soon physical events where we're just inviting colleagues, experts from academia, from other private sector, other other agencies to share their research and their work with us. And the idea is to really create discussion, questions, share our work as well, and create a community and really strengthen, I think, the communication with everyone who's working in the agroinformatics field and anyone who's just working in tech and agriculture. And today we're happy to have with us Dr. Alec DeWitt, senior scientist from the Wageningen University in Research. And Jean-Jing Chen will open the session and say a few words about, I think, agroinformatics and the tech talks in general, what we're trying to build. Over to you, Jean-Jing. Thank you. Dear colleagues, who doesn't know. As Luna said, the agroinformatics tech talk is a forum for, say, we invite experts from all around the world, including inside AFU, to share their expertise and experiences using information technology in agrofood system digital transformation for better production, better nutrition, and better environment, and better life. Today we are honored to have Dr. Alec DeWitt to give us a speech on open source tool and data for model applications in agriculture. Dr. DeWitt is a senior scientist with Wageningen University in Research. For more than 20 years, he has been working in the domain of agrometrological, mid-metrology, remodelling, and crop modeling for agricultural modeling and yield forecasting. He has coordinated many projects and international projects focused on developing of crop model system, including in the European Union, China, Russia, and Morocco, as well as capacity building in this area. With the framework of Dutch G4AW program, he adapted this technology for supporting smallholder farmers with crop advisories in Ethiopia, Bangladesh, and Myanmar. He is one of the major developers of Walford's proper satellite model. Walford is one of the early crop growth models in the world, and now is one of the major crop growth models while they use all around the world. And he is also the developer of PSE modeling framework, which is used by a lot of researchers around the world. So thank you again, Dr. DeWitt, for all your skills. Okay, thank you very much, Zhongjin. Today I will start first here on my screen. Okay, yes. Thank you very much, Zhongjin, for introducing me. Today I will give a small short talk about the open-source tools and data for model applications in agriculture, and that really boils down to things that we developed here in Wageningen over the last five to 10 years, which I think are useful for others as well. So I'm not going very deep in many of these tools or data, but I just want to try to give you an overview and point you to the resources which are available to explore it yourself. A very short introduction on Wageningen University and Wageningen Research. This is an overview of the Wageningen campus, as we have in the Netherlands, with a couple of the buildings of the various institutes here. And what many people often don't know is that Wageningen University research is actually two organizations in one. On the one side, we have Wageningen University, with about 13,000 students, 2,300 PhD candidates, and 3,700 staff. And on the other side, we have Wageningen Research, which is the contract research organization. So Wageningen University is more fundamental research education, and Wageningen Research is more application-oriented research. And both organizations are quite similar in size, and they work together very closely. They're in the same buildings on the same campus. And what you should understand is that these two organizations are largely symmetric. So for example, for Wageningen University, we have, and for example, the Department on Agro-Technology and Food Sciences, while on Wageningen Research, we have Wageningen Food and Biobased Research that applies to livestock, to environment, to plant, and also to social sciences. So often people wonder, who am I dealing with? Someone from the university or someone from research? Well, it depends. I actually am from Wageningen Research, from the Wageningen Environmental Research group here. But I actually do cooperate a lot with people from Wageningen Plant Research and also University on Plant Sciences. Okay. Well, about open source tools and data products, I want to give you an introduction today on some modeling tools that we developed. First of all, this PCA modeling framework, the models which are in there, and some other resources which are helpful in, if you want to start working with that framework. An example, actually, of what we used to build, which includes this framework where we built some of the broader framework for crop advisory generation, which was also applied operation. And some of the open data sets, actually, that we have been developing over the last few years, the AgRF-5 Global Meteorological Data Set. The global recently has been put on the Copernicus Climate Data Store, so called the Global Crop Productivity Indicators, and finally something called the AgroStack Initiative. A little bit about crop models developed in Wageningen. These models have been mainly developed for what we call quantitative analysis of development growth and production of annual field crops, and by now also perennials. For example, there are models now for cocoa, palm oil, for banana. They simulate for different agro-ecological production levels, and what we mean with these production levels is that our models usually target a so-called potential production level, where we say that this is the production level that is reachable when everything else is perfect, and in that such a production level, it's only, your yield level is only determined by the amount of radiation that you have, the temperature, and the features of the crop itself. If you go one level, production level down, you take into account water limit limitations. If you go one limit level down, you take into account nutrient limitations, and finally, you reach basically the lowest production level where your yield is actually reduced by all kinds of factors like pest disease or pollutants, which is typically what you find on the farmer field. But people often get confused about these models because they don't understand that that concept of agro-ecological production levels. Now the characteristics of these models further is that they are mechanistic, mechanistic, they're dynamic, and there are different approaches ranging really from relatively simple summary models to complex models that take into account biochemistry and physics. And most of these models, they are public, so there are open source implementations available. Now this is always a nice picture that I like to show is because there's really a pedigree of what they call the School of DeWitt models. Well, actually this DeWitt, that's not me, it's the same name, it's not even a relative, but this is Professor C.T. DeWitt that you see in this picture there over there, and he started, he wrote a very influential paper in 1965 called the Photosynthesis of Leaf Cannabis, and from that paper actually a whole family of models actually developed. And some of these models are still developed today, such as Lytto, G-Cross, Wofors, and also Orisa, the rice crop model from the, nearly in the Philippines is actually a branch of C.Cross and Wofors. Well, this PSJA modeling framework, PSJA stands for Python Crop Simulation Environment. This is really what I would call a modern implementation for crop modeling Python, under an open source license, so it's licensed under the European Union Public License. It is both a modeling framework, so you can use it to build models, and it is a re-imlimitation of many bargaining models that have existed for many years already. And it promotes good model design modularity, and it also provides easy ways to communicate between models. And what it also has, it has many tools available for reading all kinds of data, so it can read what we call legacy cable files, old weather files, but you can also connect, for example, to the NASA Power API for getting weather information. Now, it also has some opportunities, possibilities for building testing of modules and models, and it integrates very well with the scientific software stack. So, for example, things like pandas, machine learning, optimization algorithms, sensitivity analysis, you can implement that quite seamlessly through this modeling framework. Now, of course, there are also some limitations. This is a focus strongly on the agro domain. So, it's not a generic ODE solver, I would say. So, it's not a generic solver, which you can feed a couple of differential equations, and it will basically solve these different differential equations through time. It's also much slower than similar models in Fortum, for example, or C++ or FST, and it is limited to what we call rectangular Euler integration with a fixed daily time step. It also doesn't have a graphical user interface, because in many cases there is no really need for having a graphical user interface for modellers, and often what I see is that there are third parties which are very good in developing graphical user interfaces, which do that kind of work. Now, there's a list of models available that we have in PSJA. We have the Wofors 7.2 potential production PP model available, also the water limited production, then we have Wofors 8, but that's a bad I release. I hope by the end of this year we can actually release 8.1, which will have a much better implementation of nutrient limited growth. Now, there is the so-called Lingra model, which stands for lintel grassland, and that is a model for the simulation of grassland productivity, which has some very detailed physiology of growth of grasslands, which is quite different from annual crops like Wofors simulates. Another lintel model, which is lintel 3, it's also there, and there's the implementation of the FAO, water requirement satisfaction index, slightly different because you can compute it in different ways, but it basically does the same thing. And finally, there's also a separate implementation of the phenological module only for Wofors. We have quite a lot of documentation available now on PCZ. If you go to this site, PCZ.readthedox.io, that gives you the full documentation available for the framework, and partially also the models. So there is the user guide, which basically tells you how to start, how to install it, these kind of things. There's the reference guide, which gives you more insight in the different components which are inside the framework and how they work. And finally, there's the documentation of the code itself. And you can really jump around. I can show you an example. For example, if you go here, again, for example, here's the reference guide, where you have an overview of the different components. And for example, here's something about the agile manager and how you define agile management in the model. And for example, if you go to the code documentation, then you can actually go to all the different modules that are available. And you can easily actually connect to the code here. It brings you directly to the code itself and also go back to the documentation over here. Let's get back to where it was. Another, I think, very useful resource are the so-called Jupyter notebooks with examples that I created for many different types of situations that you can run into. These are also on my GitHub page. An interesting thing about this is that if you scroll down a little bit down, then there is this button over here, which allows you to launch actually a basically a docker container where the system runs, which you can then interactively run, for example. We go here, here are all the examples. For example, there are examples of getting started, running a model, running in batch mode, advanced agile management, there's advanced topics on data assimilation, parameter optimization, sensitivity analysis. There's also an example here on the linker model with crossland productivity, but if you go here, here's the launcher button, you click on this one, and it will start basically a virtual machine or docker container which pulls in the repository, and you are actually capable of interactively running those notebooks. Now, this may take a little bit of a long time, so I'm not going to wait for it, but you are able to play around with those notebooks interactively within your browser. So that's, I think, that's a very interesting way of being able to run the Python Crop Simulation environment. Now, models in the environment itself, they need a lot of parameters, so one of the things we make available is the model parameter library, which is also on my GitHub page, and that provides you the parameter sets for the Wolfos Crop Simulation model. There's another library for the Lingera model, and it gives some insight in how these parameter files are structured and how you can also add your own, basically your own varieties or cultivars to the parameter file easily. And finally, we have, of course, some specific model documentation. For example, if you go to the Wolfos website on the Wageningen University research site, you will find here under the documentation, you will find two links. One is called a gentle introduction to Wolfos. So if you're interested in understanding and running the model, then I suggest you start there, because that really gives you a, well, a gentle introduction and insight into how the model works and how to interpret the outputs and all these kind of things. And there's also the Wolfos 7.2 reference manual, which gives you information, detailed information on how the model works from a mathematical point of view. So if you weren't interested, but that's a pretty tough read. Okay. Now, one thing we developed over the last few years with this, basically, a framework is a framework for crop advisory generation. So that was really targeted at providing farmers through the cropping season with advisory on how to handle, how to cultivate their crops. And it was developed first in Myanmar together with a seed provider, lull tier seeds, and it's the name of the company. And they provide basically tropical vegetables, so-called everyone hybrids, which need a different management than the farmers usually did. They wanted to have a system for supporting their farmers. And basically, what we did was based on the idea that the day-to-day management of these crops is driven by, well, the crop phenological stage, depending on which crop stage you are, you have to do different things. And it also depends on past, current, and forecasted weather. So what we did was we developed quite a simple model in PSSA to simulate and predict crop phenological stages based on the BBC age scale for phenology. And then those advisories were developed by lull tier seeds and also by the Bangladesh Agriculture University. And they were connected to this BBC age scale and also to certain weather events. And then farmers were registered into the system based on their location, rough location, municipality, their crop type and their sowing date. And based on that information, we had sufficient data to run a model specifically for a particular farmer with a particular crop and a particular sowing date and provide them with the advisories that were relevant for that particular location. And these were then broadcasted through SMS. Well, for example, this is an example of the simulated phenological development. Well, actually, this is made in Ethiopia for an area with quite stable conditions. So you see that the orange line, which is a development stage is almost straight line. And you start at the lower left with the BBC age zero, which is sowing. And then it starts BBC age 10, which is emergence, BBC age 20, first leaf, 30, third leaf. I don't know from the top of my head. Up to the BBC age 99, which is harvest or full maturity. And in this case, like I said, it's a flat line. But for example, this is another example in Myanmar for sugarcane, where you see that those lines start to deviate based on different years. So depending on the climate conditions, you get difference in phenology. And you get differences in when actually those messages are being sent. Now, this is an example. For example, seed sowing was, I think, is BBC age 01. And that was connected to TSM zero. But for example, seedling emergence, BBC age 10 is at TSM 80. And then for each of these stages, we have message IDEs related to, in this case, seed quality, fertilizer, pesticides. And they have an so-called offset day, which means that seven days before that stage is reached, these message will actually be sent to that particular farmer. And on the right side, so see a couple of weather events where there are certain thresholds which are related to, like I said, weather events. And if that event will trigger, then the farmer would also receive an SMS message. And by connecting, let's say, the model output from the predicted phonological stages with the advisories, which connected to the phenological stages, they were actually loaded in an SMS delivery platform, where you could, for example, see this kind of information where you see the Bittercourt variety, which is called TF1. And where there's an alert type management at a certain BBC age states. And then there's a recommendation for that particular crop and that particular stage, which can then be sent from the platform. In this case, at first it was still done manually, but this was later automated when the farmers actually was growing. And this was actually developed for quite a number of crops and seasons. For example, we had maize, tomato, potato, pumpkins, okras, quite a number of, and also for different cropping seasons. In Bangladesh, they have three seasons. And so for certain seasons, the advisories could be slightly different than for other seasons. So that was also taken into account. And it was actually some of the, let's say, promotion material that was developed by Laltier seeds to actually develop their, promote their advisory system. Now, the latest information I have, because this is already four or five years ago, is that currently this particular system is not operational anymore, because what you see often in the end still, although from a modeling perspective, it's actually quite simple. It's still too complicated to maintain for third parties, and we didn't have a maintenance contract on this one. But I actually found out last month that the Bangladesh government is working on an Agumet portal called BAMIS, and where exactly those ideas, because the same partners have also been working on this BAMIS portal, are now integrated into this Bangladesh government portal, which I thought was a very nice result, actually, seeing that some of the knowledge that you transferred in this project finally ends up in the government portal for Agumetology. Well, some insight or some, let's say, some news about some of the data products that we've developed over the last few years. I think one of the most important one is AgRF5, which is the agriculturally modified R5 product of, I would, I should say, agricultural tailored R5 product, and this is a global product derived from the ESAWF R5 reanalysis. There are several steps. Well, one of the things is, for example, that it's bias corrected towards the operational ESAWF forecasts. So you're going to actually combine AgRF5 with the ESAWF forecast, and they should be relatively seamless. The bias between them should be removed for most part. It has a resolution of about 10 kilometers. We have daily variables from 1969 until real time with a delay of one week. The original R5 reanalysis actually gives you hourly values, but for many people in agriculture that's basically too much. They want daily values, so daily minimum maximum temperature, et cetera. So what we did in these AgRF5 is that we converted R5 from hourly or three-hourly to daily, taking into account the fact that the definition of them of day differs across the globe, of course. Now, what it provides is 22 variables which are relevant for agricultural applications. So temperature, precipitation, precipitation type, global radiation, daily average vapor pressure, wind speed, and also the relative humidity at specific times of the day. So I think eight times during the day it gives you the relative humidity which is particularly relevant for modeling plant disease. Now, you can find it on the Copernicus Climate Data Store. This is the, if you go to the, for example, the UI that you see on the right bottom side of the screen, you will find the, you will get here. And by download data you have either a form where you can click what you want or there's also the CDS API that you can use to download this data. There are also some limitations. I wouldn't say that bias collection is perfect. So biases still exist, particularly for precipitation. For example, the chart that you see here compares the chirps. I think it's the green one with AgRF5 in orange. And you see that AgRF5 tends to overestimate precipitation quite dramatically if you compare it to chirps for this location. Although the product is near real time, we still have an eight-day delay. So about the, I think today, which is the 29th, you get the AgRF5 data for the 22nd. It's based on numerical weather prediction reanalysis, which may fail to resolve local climates. And also there's no easy API yet, unlike, for example, NASA Power. If you go to NASA Power and you click a particular point, you can get the whole time series for that particular point, for the variables that you want. It doesn't work yet for the climate data store. You first have to download net CDF files. That's not so nice. Another open data products that I would like to show you something about is the so-called crop productivity indicators that is actually available since the end of August this year on the climate data store. And that is also an operational product which provides insight into the productivity of four major crops, soybean, rice, maize, winter and spring wheat. It's a global product at decadal time steps with a 0.1 degree resolution similar to AgR5. It is available starting in 2000 until current with a delay of about 10 days on real time. But it's only available for what I call dominant cropping areas, and I'll demonstrate you why that is. There is actually an app available on the climate data store, but I'm not sure whether it still works, and it also was very slow. It has to do with the way the climate data store actually treats this kind of data in net CDF files. And what we did actually is some validation of this product against the Faustat yields at regional level for US, China and India, but the manuscript is still pending for review. I can't provide you with DOI yet. What the product does is actually it takes, it is a hybrid product between a crop model and a satellite input. What you see here is basically for, we have a crop mask for let's say the major maize growing areas in the world. And what we derive based on the crop mask is the so-called time series of FAAPR, so that's the fraction of absorbed PAR. And if you look at the time series like this, you very clearly see these crop cycles into that data. And for example, here we have another one for soybean in North China and Heilongjiang. And also here you very clearly see these crop cycles into that product. And what we now do is we combine that with a crop model where we say, okay, we take our ag error five weather inputs. We take the FAPAR inputs from the satellite time series. We know more or less where the cropping cycle starts based, for example, on the sage cropping calendars or the FAO GIAZ, the global archeological donation. And we combine it with a light use efficiency crop model. And that light use efficiency crop model basically estimates the amount of intercepted light from the FAAPAR inputs and converts that into an estimate of growth. And that gives you, for example, these curves that you see here on the right hand side where you have soybean for a particular area in terms of total biomass, which is the upper one and the yield, which is the lower one. This is the principle that's been used to generate that product for basically the whole globe. Now this is an example of, for example, wheat total biomass 20th of June 2022, so quite recent. And you see that, for example, here it runs from about zero to 13 tons per hectare. The cropping season is not entirely over. You see, for example, here in Western Europe in France that there's a lot of wheat which is turning yellow, almost 13 tons are also here in the North China Plain. For example, there's nothing yet here in Australia and also not Argentina because the cropping cycle has not started there yet. And you can find similar outputs for maize, soybean, rice. The only fact is that this only works for dominant, let's say, areas with dominant cropping patterns. So you'll find that most of Africa is dropping out of this because the FAPAR inputs that we're using are at a resolution of one kilometer, which are insufficient to resolve the mostly small order cropping systems that we have in Africa. And that is really a big limitation of the system. Now some validations that we carried out, for example, this is what we see for validation for soybean in the U.S. at the county level. And what you shoot is more or less, if it is yellow here, we have now almost no correlation between our integrators and the reported yields. And for blue, we get good correlations. And if it is actually red or orange, it is actually getting worse, more or less. And you see that, for example, the right figure here, we have predominant, let's say, yellow to blue, which is the total crop biomass. Well, the left figure we have more red and orange, which is the yield. And that's actually often what we see is that the total crop biomass that we're simulating is a better predictor of yields than the simulated yield itself. Now, we have similar, you see, for example, that there are some interesting patterns here. You see, for example, in Missouri here, we have mostly very dark blue counties here, which means that we have very good correlations between our output and the reported yields. And certainly across the border here with Illinois, and most of the correlation is gone, or here with Iowa. These are patterns that we still have to look into to see whether this comes from. Could, for example, be a change in the cropping calendar as described by Sage, which is not working out well, something like that. Another example that we have, for example, here is, this is a time series for wheat in Spain, where we compare the FAO statistical yield here in gray with the output, the annual final at the end of season output from the system, which is here in blue. In blue, this is basically the product which is now on the climate data store. And here in orange, we have a slightly improved product. And you see that both products are able actually to follow the inter-annual variability in yield actually quite well. And improved products actually does even a little bit better than the current product. Okay. The last one I want to discuss with you is the Agostec initiative. That was an initiative started to collect and harmonize key agronomy observations of crop type phenology, biomass yield, and leaf area. It's also basically published open data sets. They are screened and they are harmonized. At this moment, the available data sets which are in the system are still limited, but it's growing. And the nice thing is also that we have an API and the development that would allow you to actually query the database in a programmatic way. This is the Agostec portal. And we have also a viewer, including there where you actually can query the data, which is in the Agostec portal, and visualize what is in there. Okay. I think this is the last slide. Well, some conclusions. I would say that the Python Prop simulation environment provides an open mature modeling framework and implements also several Wagner club models. We've quite a wealth of examples and documentation available. I must say that support is limited, also limited by funding. So if we really want to work on something together, we have to write joint proposals. Nevertheless, I answer emails nearly every day by now. So there are really a lot of people, I try to support in using the system. And I think we have some, well, useful open data products available now, such as agri5, devolves, corporate parameter data sets, Agostec, and the corporate activity indicators. And I think that finalizes the presentation. So be happy to take any questions. Feel free to also write comments or questions in the chat. I know we have a lot of work that is obviously related to what you're doing, very similar. There's a lot of synergies. I know we have one project specifically that's also working in Rural Advisory Services. Yeah. For agri5-logical data, as you know, I don't know if Zhang Jing, you want to say a few words or if anyone has comments, questions, please feel free to please your hands. Go ahead, Zhang Jing. Yeah. Thank you a lot for your very, very informative presentation. Actually, I found we have a lot for our life between our works with working in Washington. Yeah. But first of all, as Luna said, I found for the work you work in Bangladesh, for the for the advisory information dissemination, you use SMS. Actually, in a few, we have a tool called Digital Service Portfolio. We can use that to disseminate the information to the farmer. Okay. Later we can talk about how it's possible to use it. Actually, I have another question. I'm quite interested in the last one you talk about the agar stack. So who is leading this initiative? So it's possible, I feel, to try and work together with you? It's led by us, by Thito, but it's also connected to the, I think, the Duke-Lam activities and where there is this activity also for the data collection on the Duke-Lam. What's the name? There's also an activity. Yeah, I think that's one, GCAM. Yeah. I think it's also, there are some connections with GCAM as well. You know, if you hand in a geospatial platform, you have a data channel to host the we call the IGO culture essential data science. So the idea is quite similar. That you can get. Yeah, yeah, yeah. Science is something to join efforts on. Okay, I found a question from Peng Yu. Peng Yu, maybe you can ask a question directly to Dr. Dewey. Yes, thank you for your presentation. So my question is about the last part of your presentation. It's about the use of the land use efficiency model for the crop simulations. So in that data, I see you use the far part as input for the model and to estimate the barrel mines in the entire green segment. So maybe what I care about is the first is about how do you get the crop type information at a global level and what is the special resolution? Well, the crop map, the crop mask is based on, I can send you the links, is based on a product from, I think, the United States crop global crop dominance mask or something. And that's what we use to derive what we call pseudo crop specific masks, because at the one kilometer level, they are never crop specific, of course, but they are pointing at areas with very dominant crop systems for what it's worth. So a special resolution of that data is still very close, such as one kilometer or even lower? Okay. So maybe the model is far part, but also at one kilometer. Yes, it's not modest, it's from the Copernicus side, so it's spot vegetation and Sentinel. Okay. So from your model, I think it is potential to estimation of the crop calendar, the actual crop calendar from your model, because you'll have a biomass. Yeah, you could, we didn't have time to do it, but you could indeed try to basically fine tune the crop calendar based on the FAPA curves itself. In certain cases, that would work. There are also cases where it's much more difficult, because the curve FAPA curves are more difficult to interpret. So for the moment, we use the sage crop calendars or the crop calendars from the Faro GIZ product, but yeah, there are certainly options to refine. Let's put it that way. Yes. So if we have the immerse data from the mass-sensing data as another input, and there is potential to calculate some pro-calendar from the model output. Yeah, at the moment the crop calendar is an input. So you would have to basically develop an algorithm to estimate the crop calendar from the FAPA time series and then use that as an input. Sure. And finally, is that to have any plan to downscale the data side to higher resolution? Not at the moment. And that has to do with the fact that let's say for global analysis, it is more important to have a long consistent time series than it is to have high spatial resolution. So that's one thing. And the other thing is that the current indicator product, the crop model has no water balance attached to it. That's because we assume that the water stress is visible in the FAPA, but we did run a test actually with an improved model where we did attach water balance and we see that it improves. So my next step would be improve, make a second version two product, which includes water balance. Yes. Okay. No other questions from my side. Thank you. So we have a couple more questions in the chat. Mahmood, I don't know, and Henry, if you want to ask to read them or if Mahmood, you want to start and ask your question. Otherwise, Dr. Dewey, you can easily just read them. I can read them from Mahmood. Models allow to include different stress test scenarios. For example, whether disasters, flood, conflicts, and show how these can impact production and crops. No, these models are not designed to simulate the effect of, let's say, weather disasters, food conflicts, because the models basically they simulate for one hectare of crop lands and simulated biophysical products, processes that happen at that level. Of course, you could, let's say, connect them to a geographical information system and see what happens in time when there comes a flood, which has, for example, destroys certain areas or these kind of things. But the model itself does not have the option to include these kinds of effect directly. Thank you. There's no follow-up. We'll go over to the next question. Henry, Do you want to say a few words? We have your question, but if you want to deduce yourself and maybe describe the question a bit more, because of the context. Sure. No, I mean, it's basically following up on what Dongxing was saying about the digital search portfolio and about the advisory messages that you explained in Bangladesh. So even if they are a few years old, do you plan or can you make these messages available through, for example, APIs or publish them as open content? So if the conditions are met, you mentioned seven days before a certain condition is met, then these messages get out now. If there are any plans to make them available as open content, so it would be easier, let's say, to connect to them. And I gave the example of an initiative from Mercy Corps, where they have this proud open content platform, where they publish some open content. And of course, the digital search portfolio from Fowl, where we now also try to make the link and publish them on on on on Seacan as open content in the hope, and this is more hope than reality at the moment, to be able to link them also to the data on the Hamden-Hangio space platform. So wondering how you look at that. We could try to do that, but I do have to say that these these advisories, they were basically developed by a lot of your sheets that this company together with Bangladesh Agricultural University. So they are the owners of those advisories, and they would have to allow us to open to publish these as open content. So I could request them whether they can do that. The other thing I must say is that often these advisories, they are quite connected to socioeconomic context, and that you have to take into account that they are not necessarily applicable for a similar club in South Africa or something like that. Yeah, but we could try. I can ask whether that is whether the advisories are available, whether they are available somewhere, but whether they can be published as open content. Great. Any more questions or comments? I think this is a great discussion. We'd like to have more of these more interactive. So anything you guys want to add, comments, anything you want to share about your projects as well, anything you're working on, that's open space for just discussion. Sounds like a no. I think so. Yeah. Well, I think we had some great questions. Thank you. I was wondering, well, we are a user of the Gear 5 data. So thank you so much for I work in the Vapor, on the Vapor program that monitors water productivity in agriculture and publishes data on the FAO portal, including the hand in hand. And yeah, one of the question we have is from our users is if there is any possibility you think in the future to reduce the latency of the Gear 5 data, because sometimes I've heard that there are commercial products that are available on a shorter latency. And I couldn't figure out exactly how that works. The point is that the latency is defined by the error 5 reanalysis itself. And that is seven days. So as long as error 5 itself doesn't improve the latency, there's no way of improving the latency of ag error 5. The only way you could do it is by moving from the error 5 product to the ECMWF aberrational forecast. So basically you take the forecast for you take the forecast from ECMWF for tomorrow and store that as the metrological data for that particular day, until it gets overwritten by ag error 5 for that day. And that's where you can bridge the gap for those seven days between ag error 5 and real time. But the problem is that the ECMWF forecast is not an open product. So you have to buy it, you have to pay for it. Yeah, just to make it... Understand whether there is any interest and pressure. And if you think that there is room for advancing in that direction... No, I would say push for it at the Copernicus data store of the European Commission. I know, for example, that for the MARS system, from the Joint Research Center for the European Commission, that that is actually done. So they bridge the gap with the ECMWF operation forecast. So it's possible that the algorithms are there, but it's just a matter of now open data and not... Thank you. I have another question and look over to you. Thanks a lot, Luna. Thanks, Dr. Haller. Very, very interesting presentation indeed. You mentioned a few times of the global agroecological zoning. So this is for not a question, really, but rather a comment. IMB working, as my background suggests, on the agroecological zoning. And we are developing something similar to what you presented with the same need and idea to open the system and to move it in Python. So I perfectly understand the limitations. So we're also moving from Fortran to Python with advantages and disadvantages, of course, in terms of performance. So this is just for your information and also the participants. We are soon launching an update version of the package at the end of November. So we are very pleased to invite you to this presentation and also all the other participants. I think what it is very interesting in this domain and the ability to try to see how the different crop models perform. So what are the limitations? So what are the data needed? Because also those are many often constrained on developing these crop models in the field. So thanks a lot for this and I'm sure I will continue to be in touch. Okay, thanks. And just a note on that, of course, we'll share all the presentations and recording of the session and all the contacts details so we can really grow the community and the ideas to take this offline after the tech talks and see how we can work together and learn from each other and potentially continue to evolve the discussion and spark new discussions. So everything will be shared. We have another question. It's not really a question. No, it's a question for comments. Yes. Great. So anything else we want to discuss today? Anything you want to add Dr. DeWitt to close? I think we're running out of time as well as questions. They've got a very interesting discussion and I think it's clear that there's I think a lot more work we could do together and we can definitely take this discussion. I think offline to several different teams and groups here that probably have questions for you that they're going to reach out individually. Yeah. So Jean-Jean, do you want to say a few words maybe to close? Just a one word. I agree towards. Thank you. Thank you a lot. Thank you. It's really informative. Do you know we have an MEO between FU and WR? So actually we can use that to develop some product. We can work together. So there was a question you have found that we have a lot of the in common. So we really can do something together. Just a start. Okay. It would be very nice. Absolutely. Okay. And also thank you very much to our body to join this very interesting presentation and discussion. So we can, if you have a further question, feel free to contact a lot. He's very open and happy to work together with us. Okay. Thank you all for joining. Then we close this session. Okay. Have a nice day, our body. Yeah. Have a nice night. Bye bye. Luna, I think you received my presentation, isn't it? Yes. So I'll share the presentation, a recording of the session and just more information of it on how they can contact you, tech talks and everything that we can add to make sure that everyone can keep in touch and continue the conversations. So I'll add everyone in this group who joined, but also I think the invitees. There's a few people who were on the list and didn't join. So we'll just spam everyone. And yeah. And then I think people will reach out and we can continue the conversations definitely offline. Thank you so much for your time and for the presentation and thanks everyone for joining. And look forward to working again together soon. If there's any other events or anything, we'll definitely keep in touch and same with you. Any new research or projects that you're working on, let us know. We can open to doing more of these talks and presentations and also sharing things from our side.