- 04/10/2023
The project: Virtual Forests is co-funded by the Erasmus+ Programme of the European Union. This project was coordinated by a member of the IEFC and the University of Valladolid and aims to develop educational resources for university students, professionals and the general public based on the virtual tools available to foresters. The diversity of resources included educational data sets (LIDAR, field, etc.), MOOCs, test plots and virtual forests allowing 3D immersion in a wide variety of forestry contexts in the temperate and tropical zone.
Catégorie
📚
ÉducationTranscription
00:00:00 So, good afternoon. I am Christophe Radiot, the head of IEUC, the European Institute for Planted Forest. I'm glad to welcome you for this webinar.
00:00:13 This webinar, let's talk about planted forest, our regular webinar, organized by the Institute to give an opportunity to share experiences and knowledge on planted forest.
00:00:25 These events are part of the EFI research network on planted forest and also part of the IEFRO task force on planted forest.
00:00:35 They are regular events that we organize almost every two months and they are very open events, which means that at any time, feel free to make questions, to contribute, to ask anything to our speakers.
00:00:57 I also take advantage of this event to advertise two very important events. One is the International Congress of Planted Forest, ICPF 2023, that will be in Nairobi in November.
00:01:12 So, it's still time to register. And of course, IEFRO asked me to advertise the World Congress of Stockholm in 2024.
00:01:20 So, today we have three speakers that will talk about an Erasmus project called Virtual Forest.
00:01:30 Virtual Forest has been a special project that was developed during the COVID time, trying to facilitate online and distance learning on the forest sector.
00:01:49 Also, to try to promote the rapid recovery of the socio-economic after the COVID, to facilitate knowledge exchange in that context.
00:02:01 And so, we have three speakers. We have José Arturo Peque-Kilkenham. I hope I did not pronounce it too bad.
00:02:12 Then we will have Guillaume Ovega and Pilar Valbuna. We will be talking together with Edouard Bergomet-Sanchez.
00:02:23 So, this will not give you a full view of what has been done in Virtual Forest. There is other intellectual output, other results coming out from the project.
00:02:34 But this will give you a flavor of what has been achieved and how we can serve academics and professionals in the forest sector as much as possible.
00:02:47 So, the coordinator was the University of Valladolid. This is why we have so many colleagues from the University of Valladolid.
00:02:54 I'm glad we can advertise this project led by one of our members. I'm glad we can share knowledge on this. It is the main aim of the Institute.
00:03:08 I will give the floor to José. José, you will have a bit more than 10 minutes for your talk. Then we will have questions and answers.
00:03:20 Once all the questions will be done, we will continue with Guillaume. José, the floor is yours.
00:03:26 Can you share your presentation?
00:03:28 Yes, I have it. Sorry. Wait a minute. I want to show you. Okay, sorry. I have it here. Are you seeing my screen?
00:03:48 Yes, it works.
00:03:50 It works. Okay. Fine. Well, thank you very much. Thank you for the invitation. It's a pleasure to meet you virtually again.
00:03:59 We have met so many times, but I have never been with you in person.
00:04:03 And as introduced, I will introduce one of the first didactic innovations we have developed in the project.
00:04:15 It's focused generally on forests.
00:04:18 Your screen displays only the folder. We don't see the PowerPoint on the screen.
00:04:25 Sorry.
00:04:28 We see the folder with the OneDrive.
00:04:32 Okay. Why? I don't know why. Wait a minute. And now? Are you seeing the presentation?
00:04:41 No.
00:04:42 No. Sorry. I'm doing something wrong.
00:04:45 You must have a double screen, but you did not select the good screen. When you click on the share the screen button at the bottom, you have to share the good screen.
00:04:54 Share?
00:04:56 Yeah, in Zoom, you are not sharing the good screen.
00:04:59 No, I'm now with one screen only.
00:05:03 Okay.
00:05:05 Now?
00:05:06 Now it's okay.
00:05:07 Okay.
00:05:08 Well, I don't know what was happening. Sorry.
00:05:10 Well, I will introduce some innovative didactic technologies, protocols we have together developed with the partners of the project just presented.
00:05:21 And mine will be about virtual forest virtualizations, a tool that can be applied for plantation forests or for any kind of forest.
00:05:32 My name here. And well, in my previous presentation of this project, we presented a protocol that well you can access to the documentation in the documents of the project.
00:05:47 But I want to share it directly with some examples, and I want to open some examples I have already prepared.
00:05:57 Are you still seeing my screen?
00:06:00 I want to fly. Yes, it works.
00:06:03 I want to fly with all you have prepared it to France, and I prepared here a virtual tour I want to show you what's a complete virtual tour and the protocol we have developed.
00:06:15 Well, let me fly to you.
00:06:20 To close to Charlotte.
00:06:25 And here I show you a virtual tour I created several years ago.
00:06:30 And when I first want to show what a virtual tour is itself.
00:06:34 Here I have an image of a Castaneda sativa mainly copies forest here at the right you can see I'm close to Charlotte.
00:06:45 And what you can see the structure I can scroll up and down at my own pace.
00:06:51 And I can put over this panoramic image.
00:06:54 Some hotspots in example here in this image.
00:06:57 Maybe I'm interested in showing the machine and I will show here the harvester you can see the Valmet 862 a small one, a forwarder.
00:07:06 I can show you here the timber classification and in that tactics imagine now we are in a class, I could start thinking, talking about well I need a machine for the harvester but you can see that's it based on clear cuts for us classic timber classification you can put here.
00:07:26 Several images, and I can, but here you see the timber classification I could scroll in scroll out. I can go from one stand to the other.
00:07:35 And I can create a story about the silviculture of Castaneda sativa in this part of France.
00:07:45 I can go from one stand to the other and always the idea is that I am based on a panoramic image that allows me to go up and down.
00:07:55 I can, in example here I want to highlight this tree that is a seed origins tree it's a Quercus robur and I could start talking about high quality selection thinnings, whatever I would like, I could move and highlight here, the Castaneda sativa the resprouting the
00:08:13 vegetative resprouting. I can here in example put a highlight to see that tree here.
00:08:19 I can highlight here.
00:08:21 This old Polar standard that shows a different background than the one of the Coppice forest.
00:08:31 And well this protocol is very nice can be very proud to be adapted.
00:08:39 But we found that it is very time consuming. It's complicated to make you need special software, and in the break. We wanted to make an easy protocol of this same idea.
00:08:54 And that's the idea I want to synthesize you here.
00:08:59 What we did is a protocol, based on free software and very simple tools were instead of making a complex thing and very nice thing of course, like the one I showed you.
00:09:13 We created something easier, based on websites and blocks block pages what here you have in the presentation you have the link to the virtual tour I showed you, but just to comment you.
00:09:24 Each of these pictures, I used the panoramic picture is made of around 48 to 56 pictures that are then stitched together.
00:09:36 And well, each picture is you see here 1.42 gigas. It's very heavy. It gives a very good quality.
00:09:45 But it's very, let's say heavy in the US for each of these hotspots you have to take here you see in the center, the picture of the big polar old tree showing back to the cultural backgrounds the ground or the corporate robot I showed you before.
00:10:02 Then you stitch it together with a special software.
00:10:06 And then you can also now use a special cameras that I will comment later on. And then with a special virtual tool software that are commercial that you need a license.
00:10:17 You can build up the tool normally you upload in the virtual tool the panoramic view. And then you place over the panoramic image on a flat screen you place a different hotspots.
00:10:28 And then you can use it to create a panoramic view of the kitchen. And then you can use it to create a panoramic view of the kitchen.
00:10:35 And then you can use it to create a panoramic view of the kitchen. And then you can use it to create a panoramic view of the kitchen.
00:10:43 And then you can use it to create a panoramic view of the kitchen.
00:10:59 And then you can use it to create a panoramic view of the kitchen.
00:11:14 And then you can use it to create a panoramic view of the kitchen.
00:11:29 And then you can use it to create a panoramic view of the kitchen.
00:11:49 And then you can use it to create a panoramic view of the kitchen.
00:12:14 And then you can use it to create a panoramic view of the kitchen.
00:12:21 And then you can use it to create a panoramic view of the kitchen.
00:12:33 And then you can use it to create a panoramic view of the kitchen.
00:13:02 And then you can use it to create a panoramic view of the kitchen.
00:13:09 And then you can use it to create a panoramic view of the kitchen.
00:13:21 And then you can use it to create a panoramic view of the kitchen.
00:13:48 And then you can use it to create a panoramic view of the kitchen.
00:13:55 And then you can use it to create a panoramic view of the kitchen.
00:14:03 And then you can use it to create a panoramic view of the kitchen.
00:14:16 And then you can use it to create a panoramic view of the kitchen.
00:14:23 And then you can use it to create a panoramic view of the kitchen.
00:14:24 And then you can use it to create a panoramic view of the kitchen.
00:14:25 And then you can use it to create a panoramic view of the kitchen.
00:14:26 And then you can use it to create a panoramic view of the kitchen.
00:14:27 And then you can use it to create a panoramic view of the kitchen.
00:14:28 And then you can use it to create a panoramic view of the kitchen.
00:14:29 And then you can use it to create a panoramic view of the kitchen.
00:14:30 And then you can use it to create a panoramic view of the kitchen.
00:14:31 And then you can use it to create a panoramic view of the kitchen.
00:14:32 And then you can use it to create a panoramic view of the kitchen.
00:14:33 And then you can use it to create a panoramic view of the kitchen.
00:14:56 And then you can use it to create a panoramic view of the kitchen.
00:15:20 And then you can use it to create a panoramic view of the kitchen.
00:15:49 for the project, I can create here a block, I create a simple block, block a free block and
00:15:55 here with the data of the Martelloscopes, now I'm going to the Martelloscopes back
00:16:00 I have here my image uploaded, this Martelloscope is in, I could open it in
00:16:05 the IRL satellite image and from here I can move from one part to the other but I cannot include hotspots or
00:16:15 additional information over the image. What we did is, once you have it here in the block spot or in the web page
00:16:21 we created, you can afterwards
00:16:23 include them one after the other, the different
00:16:27 hotspots or information you want. In example, suppose I want to use this forest as a mixed Pinus sylvestris and Quercus
00:16:35 Copis forest, I want to use it for whatever class. I can include here a map,
00:16:41 I'm showing it on the map, I can use a more detailed map or I can use
00:16:46 complementary information about this forest. And the simple idea is
00:16:52 that if you write on the telephone or on the computer Martelloscope Juva
00:16:57 it will fly to this place or to the Fove de Chambril I just created especially for this project.
00:17:06 Here I have summarized the basis of the simple protocol we have described. Just
00:17:12 take the picture with an easy tool, not shooting 46 images
00:17:17 overlapping one with the other, that will be extremely heavy and then you won't be able to move it and then
00:17:24 once you have it you can embed these images in whatever didactic platform you consider.
00:17:31 This gives you a very easy
00:17:35 way to use it. I can, if I want now to add here another information example,
00:17:41 I have a suddenly I have a video of a machine working here,
00:17:44 then I just put it here in my block or I want to put a poet has been writing there
00:17:50 about something and then put there the poem. In the virtual tool, in the heavy virtual tool
00:17:56 it's nearby impossible because you have to repeat the whole complete process of uploading and creating the virtual tool.
00:18:02 This
00:18:05 is the result of our project, the idea of making simple something that is
00:18:12 much more precise and much
00:18:15 more fancy and nice to show but easier to create.
00:18:20 Yes, I'm finishing and we created this in the web page of the project and you can download there all the information.
00:18:27 With that I finish. Thank you very much.
00:18:30 Thank you very much, Jose. I have to make an official announcement. This event is not sponsored by Google.
00:18:36 This is just the result of the
00:18:40 pragmatical approach from the
00:18:44 researcher of the group.
00:18:46 So basically you make a blog and you include these Google tools and also additional information.
00:18:52 So now it's time for questions. So I don't know if
00:18:57 there is any question. Perhaps I can have a look. It's better if you speak. You can also post
00:19:01 questions in the blog if you are a bit shy.
00:19:04 The speakers are here to explain what they have done and why they have done it. So please feel free to
00:19:14 to question or share your doubts or
00:19:18 whatever you would like to discuss about related to this. So basically it's kind of very easy to handle
00:19:27 didactic tool. Quite straightforward. Quite simple to implement.
00:19:33 And so you can find the example on the Virtual Forest website. We will put the link of Virtual Forest at the end of the talk.
00:19:41 Okay, if there is no burning question now,
00:19:45 we will move to the second speaker and we can come back to Jose if you have some question popping up in your mind.
00:19:54 So Guillermo,
00:19:56 we'll talk about the Linked Open Data Modeling and Simulation Platforms.
00:20:01 Is Guillermo here? Yes. Yes.
00:20:05 Perfect. Okay. Guillermo, the floor is yours.
00:20:09 Okay, thank you.
00:20:12 Good afternoon, everybody. I'm going to share my screen, my presentation.
00:20:16 Okay, so let's see.
00:20:24 So
00:20:26 Okay, I think that everybody is
00:20:35 seeing my screen.
00:20:39 Yes, it's fine. Okay, good.
00:20:41 So I can proceed with this presentation. So I'm going to share what we have done in this
00:20:47 Virtual Forest project about this
00:20:52 Linked Open Data Modeling and Simulation Platforms.
00:20:56 Since Linked Open Data is not so known to foresters, just some
00:21:06 indication about what it is about. So it is a methodology for publishing data on the web
00:21:13 with the aim of making data more shareable, reusable and also to allow the integration with different sources.
00:21:22 So there's a
00:21:23 long body of research in this topic. There are many standards, many
00:21:27 papers, technologies.
00:21:31 So
00:21:33 the essential of this Linked Open Data, there are these four components.
00:21:38 The first is ontology. So this is basically data models that are shared
00:21:44 among a community to define the terminology, the terms of a domain. So for example, how to define what is a
00:21:51 what is a tree, what is a measure, what is a plot, everything that we want to
00:21:56 to share, we want to define the terminology for this.
00:22:00 Once we have these models, these ontologies, then we can annotate the data.
00:22:04 So then we can talk about the Linked Open Data. This data is then expressed with these
00:22:10 terms from the ontologies and also can be interlinked with other data sources that
00:22:17 maybe from different people. And this is a very good idea to have
00:22:20 other data that is commented by other people. So then the data is more interesting.
00:22:27 Part of the infrastructure is the use of IE
00:22:31 offerings. So this is something
00:22:33 just for making the terms
00:22:36 and the entities
00:22:39 identified with universal identifiers. So it is in a way similar to URLs.
00:22:46 So in this way we can use
00:22:49 IRIs for
00:22:52 precisely
00:22:53 identifying a resource.
00:22:55 And finally, we have the
00:22:58 software management tools, the Drupal stores. So they are specialized databases and they are
00:23:06 purpose for handling Linked Open Data, especially using this query language that is Sparky. So these are the main components and
00:23:16 what we have done in the project here is
00:23:20 in this
00:23:22 intellectual output 4
00:23:24 we have worked with two
00:23:26 different datasets.
00:23:28 One is the cross-forest dataset. This is a very complex dataset because
00:23:34 it is composed from inventory data
00:23:37 land cover maps from Spain and Portugal
00:23:40 and also the reinforced
00:23:44 gradient that I will talk later.
00:23:46 Then we have adapted a web application that is Forest Explorer for
00:23:53 accessing Linked Open Data
00:23:57 from Forestit, so the first resource. And finally we have adapted a forest simulator, Simant4, for using
00:24:04 Linked Open Data.
00:24:06 So very quickly,
00:24:08 just a grasp about this cross-forest dataset. It is, as I said, very complex. It is
00:24:15 the main
00:24:18 purpose in our cross-forest European project. So at the end of this process we have
00:24:24 more than 90,000 plots, more than 1 million trees and patches from Spain
00:24:31 and Portugal. All put together in a single
00:24:37 dataset and we have
00:24:39 many resources, especially interesting the ones in
00:24:43 the github, Blaze, the ontologies and the Sparkle endpoint for Query.
00:24:50 Well, then for the reinforced dataset the case was a bit different because the previous one was a consolidation of official
00:25:02 datasets from the Spanish Ministry and also from the Portuguese Ministry of
00:25:07 Climate Change and so on. So in this case with reinforced it is a consortium. It is also a complex resource.
00:25:16 Essentially it is a gradient of
00:25:18 arboreta from Scotland to South Portugal
00:25:22 with different trees, different species and in different plots that can be seen in this map.
00:25:30 The initial input was in CSV
00:25:32 format.
00:25:34 So we wanted to
00:25:36 release this as link-open data and
00:25:40 for doing so we had to produce a set of ontologies.
00:25:44 So it was an interesting
00:25:48 exercise. We had to define concepts from reinforced like arboretum, provenance, survival rate, slope and so on.
00:25:56 We also had to
00:26:00 define things like hybrid variants of species that are interesting of course from a
00:26:06 forest point of view.
00:26:09 Then we also reuse many ontologies from God's Forest,
00:26:14 especially the species ontologies and also the measures in order to describe things like the
00:26:20 growth of
00:26:23 trees or to specify climate
00:26:28 measurements like degrees Celsius and so on.
00:26:30 I have now some slides that I'm going to skip because I don't have so much time, but this is just to make
00:26:37 you know, this is part of the
00:26:41 development of the different ontologies. So you can see here the taxonomy of a species defining the hybrid
00:26:48 class there or
00:26:52 Okaiftus gumdal and things like that. It was a lot of work doing this.
00:26:56 So we needed to specify what is a hybrid, what is
00:27:00 each of the species and subspecies and hybrids employed in this reinforced arboretum.
00:27:06 Then we arranged the different concepts in this. This is a visual representation of it, of this
00:27:16 ontology with the trees, provenances and so on.
00:27:21 And then we also converted the data. So using these terms we were able to, for instance,
00:27:26 describe information about an arboretum in North Scotland, in Wool.
00:27:31 But I can't go into the details.
00:27:34 So this is part one, part two, more data about the, you know, the climatic data from this place.
00:27:41 That is quite
00:27:43 cold.
00:27:46 And more data about the provenances and so on. So essentially we prepared this
00:27:51 nice dataset with the CSV, which was in this link open data
00:27:56 format. And we deployed this in a special database. It is
00:28:05 called the PLOSO product.
00:28:08 It is available for querying using this SPARQL
00:28:13 language. And this is mainly purpose for knowledge engineers. Knowledge engineers are
00:28:19 people that know these semantic web
00:28:23 standards and can use easily SPARQL queries. But for foresters, this is
00:28:29 probably not very convenient. So we prepared a web API and also we
00:28:35 developed a preliminary script in R for accessing this data.
00:28:41 I'm going now to simplify some
00:28:45 different questions that can be done. I'm going to use the SPARQL
00:28:51 part of it. I'm going to
00:28:55 express the
00:28:58 information needs that was
00:29:01 collected from foresters. So for instance, one competency question is this one. We want to get which are different arboretum
00:29:10 reinforced with a mean annual temperature that is less than 12
00:29:15 Celsius degrees. So it is very specific. And this is the
00:29:21 query that we prepare in order to
00:29:24 accommodate this
00:29:28 information need. I can copy this and paste it in the
00:29:32 query editor of our
00:29:37 database. And the answer is going to be this one. So you see this in the bottom part of the slide.
00:29:43 So there are three arboretums with less than 12 degrees. The first one is Muldepeat, the one in the
00:29:49 top north part of the
00:29:52 gradient.
00:29:55 Another example,
00:29:57 we can ask for a specific provenance. So it is a variety of Castanea sativa in
00:30:03 Mediterranean France.
00:30:06 We can also ask of all the different
00:30:09 arboretums in the reinforced
00:30:13 gradient, we can ask which ones have a survival rate of more than
00:30:19 95%. And this is the query for this. In this case, we have many
00:30:24 arboretums.
00:30:28 Here, okay, and we can see
00:30:31 different provenances and
00:30:34 different arboretums in different places, Seboviem or Aldernez, for example, and the specific survival rate that was
00:30:42 registered. So this is from the data and can be
00:30:45 formulated using this query language and using the terms in the
00:30:50 defining the different entomologies.
00:30:52 Okay, so this was one part of it, of how to convert into linked-open data this
00:30:58 reinforced, how to exploit it with
00:31:01 using Sparkle query.
00:31:03 The other thing that we did in the project was to adapt
00:31:08 the forest exploration tool application for
00:31:15 browsing the contents, in this case not the reinforced dataset, but the
00:31:22 cross-forest dataset. So this is a web application and you can see there the URL and
00:31:30 yeah, so this
00:31:33 a web tool that aims to simplify access to linked-open data. So essentially it is
00:31:41 for free users from using Sparkle or other
00:31:47 languages like RPF that are
00:31:52 in a way
00:31:54 complex and not expected for
00:31:56 foresters. So this is
00:32:00 one example of
00:32:02 forest explorer. So here I will go in to show some snapshots
00:32:06 of
00:32:09 Iberian Peninsula. In this case, you can see the
00:32:12 left part, you can see
00:32:15 inventory data from the different regions in Portugal and in Spain and we are
00:32:22 specifying where we are filtering by two species, Pinus silvestris and Pinus vinaster, and using
00:32:29 indigo and brown colors. So this is very aggregated data, in this case from
00:32:34 Sardinia, one province in Spain, and in the right we can see this is the part of the map
00:32:39 and you can see
00:32:42 crops in Orleans, I mean
00:32:46 farms, but also forest in brown color, the Pinus vinaster, and in indigo color, the Pinus silvestris
00:32:54 patches.
00:32:57 You can do this with the tool easily. If you zoom in, you can go to
00:33:02 a very specific and very narrow area like this one in
00:33:08 Sierra de la Demanda, in Soria, in Spain, and you can see an interesting place with pure
00:33:15 Pinus vinaster plots, pure Pinus silvestris plots, and mixed plots. You can see easily with this representation that is showing
00:33:24 the land cover map at the bottom and on top of that the plots, and in this case focusing on one specific
00:33:31 plot. And if you zoom in more, then you go to the plot and
00:33:38 you can see a nice plot here with Pinus vinaster and Pinus silvestris, and you can see data from the plot.
00:33:46 It is the tooltip on the left.
00:33:48 If you
00:33:50 go to the tooltip, you can also
00:33:52 get the measures of
00:33:54 every tree in the plot.
00:34:00 Indeed, the dataset is quite large, but you can go to the area of interest and it is
00:34:07 a nice example of how to consume link-open data, so all the effort with the ontologies and converting data and so on that can be
00:34:16 navigated with
00:34:18 tools like that. So this is just an example. You can do something similar with Rainforest or other
00:34:23 forested datasets in this format.
00:34:27 Finally,
00:34:30 this part is more a work in progress, but we are currently
00:34:36 working on this, so how to consume link-open data not only for
00:34:43 visualizing this kind of data, that is the example of Forest Explorer, but also for doing things with the data that make sense, like
00:34:51 forest management using simulators. So this is something that
00:34:55 is very useful in this domain, and
00:35:00 particularly using inventory data.
00:35:04 Using inventory data is
00:35:07 not so easy, even for
00:35:10 forestry people that know the variables and so on, and even the formats that prepare the data for
00:35:16 running a simulation, it is quite
00:35:21 time-consuming. So the thing is that if we have all the data in this format,
00:35:27 we would like to automatically consume the data from our resource, our cross-forest dataset, and then
00:35:35 run simulators with
00:35:38 this
00:35:39 CMAN4 tool that we have in Valencia.
00:35:42 So what we have done in Virtual Forest is to prepare an API that
00:35:48 obtains
00:35:52 inventory data
00:35:53 according to a
00:35:55 selection criteria for obtaining the plots of interest.
00:36:00 And now we are working
00:36:03 on our web interface for
00:36:06 selecting the input easily with menus, interactive maps, and so on. So in this case we can
00:36:13 basically
00:36:16 specify which are the plots of interest, and then we can run our
00:36:21 simulation with the parameters that we want. This is the current
00:36:27 way of doing things in CMAN4 and also other
00:36:30 simulators. And we have this
00:36:35 API published and we can provide the guide. We hope in the following months to produce a new version of this
00:36:42 simulator of CMAN4 that makes
00:36:45 easy this consumption of official inventory data for running simulations.
00:36:52 And I think that's all. Yeah, sorry for the long talk.
00:36:57 Yeah, I'm available for questions.
00:37:01 Thank you very much. So it was a good opportunity to see how to apply open forest data,
00:37:06 open data in the context of forest.
00:37:10 So I'm not sure all our guests are familiar with the concept you shared. Is there any question for Guillermo?
00:37:18 Me, I have a question.
00:37:25 What is the benefit of semantic
00:37:29 approach compared with other queries? Because it looks like it's more complex to implement,
00:37:35 but then you must have an advantage at the end when you have sorted out the data in that sense.
00:37:40 Well, I mean you need modeling
00:37:43 at some point if you want to access the data. So you need to model the...
00:37:50 We are modeling every time. So even if you use...
00:37:55 If you're using open forest, you have a CSV file and you were
00:38:00 specifying the number of columns, so you are using a model. It could be more explicit or more implicit,
00:38:06 but you are modeling. The good thing about using anthologies is that
00:38:10 the semantics are made explicit. So basically your model, you are
00:38:15 exposing your model to others. And this is good because then
00:38:19 and then it is easier to
00:38:23 manage the data. And not only this, but you can also
00:38:26 interconnect the different
00:38:29 different datasets. So this is especially
00:38:32 useful. The case of
00:38:35 cross forest is quite interesting because even the Spanish
00:38:41 forest inventory dataset, it is not just one dataset. It is
00:38:49 100 datasets because there are two different databases per province and there's like even some small differences
00:38:56 between
00:38:59 each file. So the format is not even coherent. It is not the same for every province.
00:39:05 So we were able to put everything together in a single place and also to interconnect with
00:39:13 data that was prepared from other people in Portugal
00:39:17 about the inventory data and also maps. So this is, I mean, it is more useful
00:39:23 if you are able to interconnect more data.
00:39:29 For example, it means that if you know that in a database there is something called tree and you find it in another
00:39:37 database, then you can get all the information for all the objects called trees. That's what you are explaining.
00:39:43 And you don't need to know the colon and how it is stored because the semantic will make you
00:39:49 clear that you can access to this kind of object. That could be a very interesting case that you have a
00:39:56 database of trees and it is like local one. You have like a local identifier.
00:40:01 That's the reason that we need these global identifiers because we want to be very
00:40:06 precise. I mean, without ambiguity, we want to identify that tree and then we can interconnect things.
00:40:13 This is a bit tricky. I mean, sometimes it is
00:40:16 challenging. I mean, sometimes it is easy if the things are
00:40:26 the same. We are talking about the same things because they can be very polysemic in a way.
00:40:34 It's not the same talking about the tree, but
00:40:37 if we are
00:40:42 working with inventories, we are talking about the inverse. So this could be a bit...
00:40:47 Sometimes it could be difficult to integrate the different datasets if the context is not the same. So this is
00:40:54 one of the reasons that the technology can be a bit tricky and the solution for this, because the problem in this
00:41:02 integration problem is quite complex. It's complex, but
00:41:09 this is the best solution that we have
00:41:12 till now for dealing with it. So you have seen how it works. So essentially we need to
00:41:20 put some effort in developing this model. We have to agree with users.
00:41:27 This ontology has to represent, you know, to be something that
00:41:34 people in the domain agree with. So this is something that we need to
00:41:39 share the different opinions and then prepare a model that can be
00:41:45 good for
00:41:49 that field.
00:41:54 Once we have it, then we can make the conversion of data and then, you know, if the model is good and the data is good,
00:42:00 then
00:42:02 things can be... So this is something that we saw in Forest Explorer. The original data, the source data,
00:42:08 is quite good. I mean, it is a good method for
00:42:13 doing the forest inventory. The same for the maps. We can disconnect because there are, you know,
00:42:20 the rules are known, the method is known, and the original data is good. It's just
00:42:27 showing it, making it accessible to everybody. So I think this is a nice example.
00:42:33 But then in specific cases, we have to work case by case and,
00:42:38 you know,
00:42:41 always some
00:42:44 things can go bad, but if we do
00:42:49 everything well, I think that the outcome can be quite...
00:42:55 So the key message is that we can easily exchange data. You have a topology in progress,
00:43:02 and the virtual forest offers a good opportunity to make all these things improve.
00:43:07 I mean, I think this is... The thing is that now we have
00:43:13 produced some relevant link-up data in the field in forestry. I think we are
00:43:19 already in the process of continuing with it.
00:43:22 We are finding that the case of SIGMA4, of this
00:43:27 simulator, it is a very nice tool. It is useful in the field, but it is very
00:43:35 cumbersome to produce this input data. So in a way,
00:43:40 if data is more accessible, it's more easy accessible,
00:43:46 we can do more things, and we are in the process of releasing more data,
00:43:51 probably making things more interesting for domain experts
00:43:55 like you. So, I mean, it is a long road in a way, but we are
00:44:01 working it.
00:44:03 Okay, good.
00:44:05 Is there any other question? You can raise hand, type them into...
00:44:10 Yes.
00:44:13 There is a question. This technology
00:44:20 with its inventory. Okay, it's more comments than a question.
00:44:24 Do you want to elaborate on that?
00:44:29 Only a bit. I think for inventory data, it is definitely
00:44:34 a case that is interesting for you. We are working...
00:44:38 Everybody in the world is working with inventory data.
00:44:42 It would be... Just imagine that if we're able to have a common inventory ontology
00:44:48 that could be used across countries. This is the case.
00:44:52 In a way, we
00:44:55 were able to do something like this for Portugal and Spain, but it would be
00:44:59 nice to have other inventories from France, for instance, or Germany.
00:45:05 The interesting point is that there is a willingness to have a kind of European inventory,
00:45:12 you know, making all the national inventories able to communicate to them.
00:45:17 This is something that is discussed at
00:45:19 European level in Forest Europe.
00:45:22 Yeah.
00:45:23 And so probably it would be important to get aware of this work.
00:45:28 Yes. One, in this, the architecture that we followed for the ontologies regarding inventories was to
00:45:36 make the
00:45:39 national ontologies for inventories, like to have like the singularities of its model.
00:45:44 So it is not the same, the plots in Spain as in Portugal. So the methodology is different.
00:45:50 So we can... And even the data that is collected at this sample is different.
00:45:55 So we have like these local ontologies, but then we have this upper level ontology.
00:46:00 So at the international level, we can
00:46:03 ask about the different things. So there are common concepts like, you know, plots, trees, heights,
00:46:10 diameters, and scale, at least in a minimum set of terms.
00:46:17 And that's, yeah, we want to work with this. I mean, it takes
00:46:21 effort to produce, to make these ontologies, to make these conversions, but now we have it in place and we can go
00:46:27 to something more useful. And just imagine that you can make these simulations, for instance, using
00:46:33 automatically gathering data from
00:46:35 inventories in different sources. So this is something we'd like to do.
00:46:41 Now, I think it's time to give the floor. So thank you, Guillermo.
00:46:45 I think that now it's time to give the floor to Pilar.
00:46:50 So we will talk about another virtual tool.
00:46:53 This time we will talk about conversational agents.
00:46:58 So in the context of virtual forests, you applied them to forests.
00:47:04 So Pilar, you will explain us what has been done. Thank you.
00:47:09 Sorry, let me intervene one second before the presentation.
00:47:12 In five minutes, I will have to leave the meeting. I say goodbye and thank you very much for the
00:47:19 invitation, for being here. As announced, I will have to leave in a couple of minutes. Thank you very much.
00:47:25 Thank you, Jose. Thank you. If there is questions, we'll forward them to you.
00:47:29 Bye-bye. So Pilar, the floor is yours.
00:47:33 Thank you very much. No, it's been very interesting
00:47:37 presentations before me. Please let me know if you can see my presentation in full screen. Is this okay?
00:47:44 Yes, it's okay. Okay, great. Thank you very much.
00:47:47 So Eduardo is here in the meeting too. So please, Eduardo, feel
00:47:52 absolutely free to intervene if I say something that is not okay, or you want to complement anything.
00:47:59 So I'm gonna present this
00:48:04 how we can enhance forest education through conversational agents with chatbots. We have seen with
00:48:11 TATTPD all the, maybe the risk and all those things related to chatbots, but
00:48:17 we think we can use them for improving some aspects of the forestry education.
00:48:23 So
00:48:26 as you may know, one of the aims of the Virtual Forest Project is to enhance
00:48:32 online distance and blended learning in the forestry sector,
00:48:35 and more to international students
00:48:38 to improve their access to open educational resources in this sector.
00:48:46 So in university virtualized courses in the forestry domain,
00:48:53 these chatbots, like the one we are presenting, and then a former one that was already
00:49:00 developed in the context of the project, can also play a crucial role for providing not only
00:49:08 knowledge or definitions like the ones I prepared, but support
00:49:11 for students for administrative inquiries, like was the previous chatbot
00:49:17 realized,
00:49:19 and the case of this chatbot is a domain-specific
00:49:22 concerns about
00:49:25 forestry concepts and definitions.
00:49:29 So
00:49:31 this chatbot has been
00:49:33 built in Dialogflow, and I will show you in
00:49:38 the next slides, too, we are using TATTPD API
00:49:44 to add more, better interaction to the chatbot for the
00:49:49 students of the Master's Degree on Mediterranean Forestry and Natural Resources Management, MEFOR,
00:49:56 and for the
00:50:00 course Introduction to Forestry and Natural Resources. So it's one of the
00:50:05 subjects that the studies have to take if they are not foresters and they're going to make this study this Master.
00:50:13 So this Master is a two-year world-class international program on the field of forestry and sustainability, and here in the map
00:50:22 you can see that the students can
00:50:26 come from many different countries, continents, and
00:50:30 it's not in the map, but very many different backgrounds.
00:50:34 So the context of the
00:50:37 information that this chatbot provides is
00:50:43 linked to this
00:50:46 Master, and it's Mediterranean Forestry Systems, Services on Mediterranean
00:50:53 Trees, Species, Forest Landowners and Other Stakeholders, Department of Social Forestry, Ecosystem Services,
00:51:00 Forest Diagnosis, Plant and Diseases,
00:51:03 Measurements and Forestry-related Data, Common Forestry Practices, Forestry and Natural Resources,
00:51:10 Economic and Social Perspectives,
00:51:13 Forest Disturbances and Risk, and Careers and Forestry Organizations.
00:51:17 So,
00:51:20 the first thing that we have done is we have made this
00:51:24 extensive search on all the concepts related to these subjects,
00:51:30 from
00:51:34 the
00:51:35 dictionary from the Spanish Society of Forest Science, for example, that they have
00:51:39 15,000 concepts included there, but we have used databases and dictionaries from
00:51:46 FAO, from universities,
00:51:50 from
00:51:51 specific from some
00:51:53 professors or
00:51:56 universities that they have
00:51:58 this compendium of
00:52:01 concepts, or from organizations like
00:52:05 like
00:52:08 scientific societies. So, first of all, we have
00:52:11 put all this information together, made a selection what was more important for each of the subjects,
00:52:18 and we have it in Spanish and English. So, this was like the first part that is more
00:52:24 knowledge management than the other part of the work, but
00:52:31 it's very important to have better results afterwards.
00:52:36 So, the chatbots are software tools that use natural language processing to interact with humans,
00:52:43 and they have been used in educational contexts
00:52:48 in many, many cases, and for things like tutoring, question answering, and language learning practice.
00:52:54 And
00:52:56 why it's very interesting?
00:52:58 One important point that is very interesting is because they are always available to answer questions.
00:53:04 So, it's something that you fill with information and it's always available for the student
00:53:10 in an easy way.
00:53:14 So, first we use Dialogflow, that is a tool,
00:53:18 a Google tool,
00:53:22 and you have to build all this.
00:53:25 You have this agent that is your chatbot and you have to put all these
00:53:30 intense training phrases and
00:53:33 you,
00:53:35 we see it here.
00:53:37 So, you train the chatbot with the possible questions that the student can ask
00:53:43 and you give the right answer. I don't know if it's included here, the right answer that the chatbot has to give.
00:53:50 And
00:53:54 then we have some other,
00:53:56 this is called entities, that is some
00:54:00 acronyms or some specific things that we tell the system to identify,
00:54:06 like PSA for
00:54:10 Payment for Environmental Services. So, sorry, this is in Spanish, but we have it the same in
00:54:16 English.
00:54:19 So,
00:54:20 once we have included all that information one by one with all the questions and if you see here, we have
00:54:27 seven different
00:54:31 questions, like
00:54:36 does to have sense a contract for environmental service?
00:54:41 What is,
00:54:44 can environmental services be paid?
00:54:47 What are payment for environmental services? So, we formulate different questions for each of these concepts
00:54:54 that in fact we have
00:54:57 more than 1000
00:55:01 in English and as we have included the ones in Spanish for the
00:55:06 dictionary of the Spanish Society of Forest Science, but none of them, we are around
00:55:14 8000. So, it's quite a lot of information and it has to be done one by one with all these questions.
00:55:20 And then we have it, it's integrated with Telegram.
00:55:27 So, you go to this chatbot in Telegram and you can make these questions you see in the video there.
00:55:34 This is an example
00:55:35 of
00:55:37 these specific
00:55:39 terms and the chatbot answers with the correct definition.
00:55:45 So,
00:55:48 we use it for concepts related to forest activities.
00:55:55 First, they get the first answer and then
00:55:59 we have included in some cases to follow up questions, like more information about some of the
00:56:07 concepts.
00:56:09 And then it can handle like some small talk to interact a little bit more.
00:56:15 The feeling is like it's a little bit robot. So, and when we were
00:56:19 building this, the TGPT
00:56:22 arrived and
00:56:25 the people is used now to interact with other chatbots that they are much more
00:56:31 small talkers, we could say.
00:56:33 But I'm going to explain later that they can have some problems related to specific concepts.
00:56:40 So, they're in Dialogflow.
00:56:43 Every time someone interacts with this, we can follow up and see all the interactions
00:56:48 to redefine and acknowledge.
00:56:53 It's going to be sent to the students of Medford at the beginning of this year and we will send them
00:56:58 to
00:57:01 fill a form about their experience
00:57:03 working
00:57:05 using the chatbot.
00:57:07 And then we have
00:57:10 TGPT, but if you go and ask directly
00:57:13 some forestry concepts to TGPT, the question
00:57:19 maybe is not the right one. So, if you give it the context, it can give you better answers.
00:57:26 But if we want
00:57:28 to
00:57:29 the chatbot to give us
00:57:31 the definition we have made, like a specific this definition,
00:57:36 you have to give it like a dictionary. So, there are different ways of working with the TGPT API. You can
00:57:48 train it with your data if you want it to
00:57:52 not to give it just the exact answer,
00:57:55 but something that is very close to that, but it can play a little bit with the
00:58:01 definition and with the language. It can do that. And another option is like you give it this dictionary,
00:58:08 it feeds from there and then all the
00:58:12 all the interactions are more fluent. And if you ask for some concepts or some information
00:58:18 that is not in the dictionary, it can go and look inside of the system or on
00:58:24 Google with the next or internet with the next
00:58:28 actualization that they're going to implement, I think, in the next month.
00:58:33 But
00:58:35 in that case, when it doesn't get,
00:58:39 it's not something that we have given it, the dictionary, we have told it like
00:58:44 that it has,
00:58:46 we have given it a context, we can say, that you are an expert in forestry, you have to
00:58:52 give me a definition related to forestry concepts and
00:58:57 and all that. So, you made the dictionary, it's interacting with with TGPT, this is made with
00:59:07 now we have made it with Google Collaboratory
00:59:10 and it works while you're working on the Google Collaboratory.
00:59:14 Now we have to implement it in the virtual machine. I'm working on that and
00:59:19 and
00:59:21 I had some problems with that, but
00:59:23 we will solve it very soon. And then we will implement it in Telegram,
00:59:28 like the one. And in the, and in both of them,
00:59:34 in addition to Telegram, they are implemented in the website for the
00:59:37 of the project and for the
00:59:40 students of the Med4Master.
00:59:43 So,
00:59:46 why is interesting this? So, we can, we have
00:59:51 someone or a bot that can answer
00:59:54 24/7 questions to students about forestry related
00:59:59 definitions and concepts and
01:00:03 we
01:00:05 we
01:00:07 we make sure that they get the right definition and right information.
01:00:11 Because maybe if they go to Google and they're not good with
01:00:15 where they get the information on some of the concepts, they can
01:00:19 they
01:00:22 they can
01:00:23 take to misunderstandings. In Spanish, we have some examples and they cannot maybe they don't find the right definition.
01:00:30 We give them like a like a dictionary where they can interact in a very like human way
01:00:36 human way.
01:00:40 So, that is the
01:00:41 the
01:00:42 QR code to access the
01:00:45 English bot made with Dialogflow on Telegram.
01:00:49 If you go and interact
01:00:54 please let us know or let me know if you have any questions or if you want to
01:01:00 fill in the form about the experience with the chatbot.
01:01:05 And
01:01:07 I thank you very much to Eduardo, Felipe and Irene who are helping me with this and I don't know if
01:01:14 Eduardo, you want to add something? I leave you the floor.
01:01:21 Well, not at the moment, I think you presented it very nicely. So, if in the questions
01:01:27 I have to add something, I will.
01:01:31 Okay, so it's time for questions.
01:01:34 So,
01:01:37 did you expect this kind of chatbot to be developed?
01:01:40 Are you using these kind of things? Do you have questions for the people who made them?
01:01:45 We have a question, how many students already are using it?
01:01:51 We haven't yet used it.
01:01:53 We are starting to use it this term. So, I don't know how many students are there in the course.
01:02:00 And so, when we use it with a Telegram, it means that we just type the address like if it was a contact in Telegram
01:02:08 but in fact it arrives to the server.
01:02:10 Yeah, that's it. It's a
01:02:12 it's a chatbot.
01:02:16 So, it's like you're writing to someone who is called Forestry
01:02:19 Concepts and then it connects
01:02:24 with the Dialogflow
01:02:27 and
01:02:30 it's like with the API, like
01:02:33 it connects.
01:02:36 So, the Telegram
01:02:37 is a chatbot.
01:02:39 So, it's a chatbot.
01:02:43 It connects.
01:02:45 So, the Telegram calls the information you have on Dialogflow.
01:02:48 Okay.
01:02:50 It's very important that Dialogflow collects these interactions.
01:02:54 So, you can through the interaction with the chatbot, we can see how to improve
01:03:00 at some questions or maybe different issues that are missing that the people is asking about them. So, we can improve on the goal.
01:03:09 So, as this project is funded by European funds and this is again a Google tool that is an American tool.
01:03:16 Can you tell us why did you select this tool? Because we could expect that European funds have to develop European tools.
01:03:25 Yeah, well, this is something that we did
01:03:29 actually at the beginning of the project.
01:03:34 We searched for available technologies to develop bots
01:03:39 and
01:03:41 one of the things that
01:03:43 we need to consider is that
01:03:45 if we only want to develop a bot that is not
01:03:51 based only on detecting keywords
01:03:53 but also has some natural languages abilities,
01:03:57 there are not so many alternatives which are not owned by any of the big
01:04:06 IT players like Google or Amazon. Amazon has one or Microsoft has another one
01:04:11 and so on. Facebook has another one.
01:04:14 So, we decided to use Google Dialogflow because
01:04:18 it can
01:04:22 allow to exploit the benefits of
01:04:28 built-in natural language processing in some way.
01:04:32 So that if we formulate some training sentences as Pilar explained
01:04:36 for an intent,
01:04:39 we don't have to formulate every one of them. Because of the natural language processing capabilities,
01:04:45 if a user formulates this in a different fashion, it will be detected. This is not easy to build
01:04:52 on your own premises or your own so forth.
01:04:55 Also,
01:04:58 in the selection of technology, we consider the compliance to GDPR.
01:05:02 Well, at least on the papers,
01:05:05 you can offer this from a server in Europe so that
01:05:10 it complies with GDPR. But I'm not sure about this, of course. We have to rely on the
01:05:18 published information.
01:05:21 But yes, it is true that
01:05:23 we rely on a technology that is a third party and is also American. But there is no
01:05:31 to our knowledge, at the moment of starting the project, there was no alternative that was
01:05:37 good enough and based on an European provider.
01:05:44 Okay, so it means that you explored it and that is the result of
01:05:50 a comparative analysis. Yeah.
01:05:53 Is there any other question?
01:05:58 In fact, if somebody is interested, I think we summarized this
01:06:02 in the first deliverable of the... well, kind of deliverable, because this type of European project is not
01:06:10 reported through deliverables like other projects. But still, we wrote some reports that are published in the website of the project.
01:06:18 So yes, to answer the last question,
01:06:22 all the webinars are available
01:06:26 online for IEFC members. They can log in and have access to all the webinars.
01:06:31 And the results of Virtual Forest
01:06:34 are also on the Virtual Forest website.
01:06:40 So the presentation will be in the restricted area of IEFC and
01:06:44 the results are available on the Virtual Forest
01:06:48 website.
01:06:51 Okay, if there is any question, this is time to
01:06:54 ask. We have our speakers already here.
01:06:58 So if nobody has no question, I don't see any hand up, it's the last chance.
01:07:05 So I would like to thank you a lot,
01:07:08 the coordinator of the project, Felipe, who is not here, and the Rebel. I would like to thank a lot the speakers,
01:07:16 Pilar, Guillermo, and Vardo José.
01:07:22 The next "Let's talk about planted forests" should be in November.
01:07:28 Hopefully, it will talk about insurance for planted forests.
01:07:32 You know, when you plant, you are interested sometimes in insuring your forest to make sure that
01:07:37 if they burn, you get some money back.
01:07:40 And
01:07:44 so thank you very much. Have a nice afternoon and a nice weekend after.
01:07:49 And do not hesitate to contact us if you want more information on one of these topics. And you can also suggest
01:07:56 new topics for the next "Let's talk about planted forests"
01:07:59 if you have some nice ideas or some nice information to share.
01:08:04 Bye-bye.
01:08:07 Bye. Thank you.
01:08:09 Bye. Thank you very much.
01:08:12 [BLANK_AUDIO]
Recommandations
2:24
|
À suivre
1:24:49
1:59:16
1:31:51
1:45:13
0:54
11:33
2:19
2:19
2:19
1:16:47
2:43
2:19