How to investigate Europe’s biomass industry with open-source data
A guide for journalists on how to conduct analysis on Europe’s biomass operators and tree cover loss
There is a wealth of lessons in using open-source intelligence techniques to investigate Europe’s biomass industry. The consequences of forest cover loss for the benefit of biomass power generation is at the centre of a recent cross-border investigation and a number of news and research pieces recently discussed forest loss and the link to biomass harvesting (such as this one for National Geographic).
To hold companies accountable, this little guide wants to shed light on how you can produce some valuable insight on the matter with the help of open-source intelligence (OSINT) tools and practices. We will discuss the merit and limits of some satellite data, too. We will gauge how forest cover loss data can best serve to aid monitoring biomass-related business practices and how some of the standard #OSINT techniques apply.
Journalists and academic researchers argued for greater oversight and tighter scrutiny of biomass companies cutting forest in Europe and elsewhere. I hope this little write up encourages other investigative journalists and researchers to start where we left off and keep tabs on roundwood hungry corporations.
A bit of background first
The premise is to understand how some of the leading European wood pellet manufacturers (one based in Estonian) use EU protected Natura 2000 forests to produce biomass for the export market. One company did produce explanations why it cuts forest in what appears to be EU protected areas. The company in question and its sister companies, own vast swaths of land on these zones.
Quote by company: Estonia allows cutting trees in buffer zones, adjacent to the Natura 2000 protected sites, where the economic activities are allowed but limited in order to not compromise the established conservation values. The Estonian Ministry of Environment said that those limited management areas do not have habitat’s directive habitats and the cutting has been necessary to either boost vitality or maintain the conservation value of each specific site, often semi natural. Management of such [Estonian] forests is thus important to preserve their health”.
The truth is, within Estonia, the company doesn't necessarily act against the law. It’s a job for policymakers to see the peril. As things stand and under current local regulation, the government opened up new avenues, not closed them, to allow cutting in EU protected areas in the past four years.
There are a range of critical voices, including a number of academic experts who say it can be devastating for the balance of the forest and its biodiversity when forest is removed in sensitive zones. This post is not about the policy and will only speak of the techniques to produce the necessary intelligence.
Google Earth Pro and QGIS
Google Earth images and QGIS are useful tools to spot changes in forests and to investigate boundaries (another of my posts addresses the importance of boundary files for clearly).
Google Earth temporal coverage for satellite images for Estonia is fairly good. Via Google Earth Pro’s timeline features you can scroll back and forth in time, — though, sometimes we are missing large parts, at time year-long gaps (which we will later supplement with Sentinel 2 images).
The first essential file we will consult is the updated boundary files for Europe’s Natura 2000 areas.
These areas comprise an ecological network composed of sites designated under the Birds Directive (Special Protection Areas or SPAs) and the Habitats Directive (Sites of Community Importance or SCIs, and Special Areas of Conservation or SACs). It is up to each EU member state to implement policy to protect these areas. Download the database ZIP file of 58.98 MB. It should include data in shapefile format which we can upload into GQIS — note it’s the file called ‘Natura2000_end2019_epsg3035.shp’.
To analyse only Natura areas that concern Estonia, we will need to filter the MS column by ‘EE’ for Estonia.
We end up with a boundary vector file that shows only Estonian Natura 2000 areas. Next on the agenda is to export it as a KML file, a format Google Earth Pro can read.
Exported now as a KML file, we can drop it into Google Earth Engine (download here). You should get the following image, a sea of red-lined polygons covering the baltic country.
Now that we know where sensitive areas are we can test for changes. A word of warning, there may be a number of reasons why forest areas vanished, including wildfires and a number of other reasons — but undoubtedly, with commercial forest cover removal to be the big elephant in the room.
Also note that the areas appears large enough to allow making a rough calculation of the change in forest cover loss. Instead of doing this for all areas, for this example we only pick one EU Natura protected area (out of three large nature parks: Otepää Nature Park, Soomaa national park and Haanja Nature Park, all three areas are surrounded by Natura 2000 zones). We’ll take an in-depth look into Haanja Nature Park.
Studying protected areas
First, let’s check Google Timelaps (type in the coordinates: 57.7265785, 27.0819402). You can see some drastic changes over the years. Some areas of forest cover disappear completely and the image alters notably in the past decade.
Let’s use open source GQIS to single out the Haanja Nature Park Natura 2000 polygon. Export it as a single feature as a KLM file (click on it with the feature selection tool and then right-click on it and save it as ‘selected feature’).
Let’s quickly check the areas in Global Forest Watch, a portal that relies on algorithmic calculations on annual forest cover loss data from open-source satellite images (for all those who scream caution, we’ll mention some caveats shortly).
The data is based on Hansen, UMD, Google, USGS and NASA satellite observations. Go to map view and drop the boundary file for analysis. You can do this in a more algorithmic fashion via R or Python using the data published but for simplicity reasons, we will stick to the simple method and the online portal (note, the page annoyingly breaks sometimes).
You can test one area with boundaries presented here. We immediately receive a number of statistics, such as tree cover as of 2010, tree cover gain pre-2012 and tree cover loss, per year up to 2019.
Global Forest Watch forest cover loss data
We mentioned caveats. Global Forest Watch data does have some shortcomings but from an OSINT point of view, this doesn't mean it useless. Some of it is described here (for instance, there is an error rate of 20 per cent or more).
The reason I feel compelled to mention some limitations here is that some, mostly lobbyist groups, pointed out that a recent Nature study published in 2020 relied heavily on the data in question. The criticism expressed by trade associations raised few eyebrows but more credible organisations did question it, too.
The Natural Resource Institute of Finland, a forestry harvesters, argues that comparison of change over time may have led to unreliable results. Industry group, the Confederation of European paper industries thinks that satellite imaging may be a misleading proxy for wood demand. The European Forest Institute, probably the most credible source here and which is now preparing their own study, says that determining large scale forest changes with the help of satellite images is invaluable. But it did caution that the results the researcher drew for this study across Europe aren't really ‘plausible’ and needed more ‘thorough checking’.
Dr. Žiga Malek, assistant professor at the Institute for Environmental Studies at the VU Amsterdam says that despite (some) criticism of the approach from various stakeholders, there was so far no major criticism from the scientific community. The GFW data, while not perfect, presents a suitable first step for identifying forest disturbances and subsequent in-depth analysis (in this case higher resolution satellite imagery).
Katharina Schulze, a PhD candidate at IVM Amsterdam says that there are some limitations to the tree cover loss data from GFW, but she wouldn’t say the data isn’t wrong. Limitations starts with the definition. Tree cover and forest cover are used interchangeably, as it is often the case in the literature. Forest cover would therefore also include large orchards or other non-timber plantations, such as palm oil, but we can assume that this confusion isn’t of high concern in Estonia. Additionally, as the [Nature] paper says correctly, forest cover loss can also occur through windfall, wildfire or pest. If these occur on large areas, it is more than often directly caused by human influences and indicates a disturbed forest system.
Another limitation, pointed out by researchers from SLU concerns the improvement of satellite images and detection methods over the years. Comparing absolute numbers between single years can therefore be tricky. Nevertheless, the tree cover loss data can give a good indication on where forest dynamics are changing and human activities impact the ecosystem.
Checking forest cover loss data and its validity for Estonia
We can check whether the forest cover loss data is largely correct for the Haanja Nature Park. One way is to build an overlay in Google Earth Engine. We take a screenshot from GFW and use the overlay feature (a tutorial here) with the outlines for the Natura area. They should perfectly line up if you used the same KML files.
The next thing is checking each patch. We choose one area and use Google Earth Pro’s time-slider to go all the way back to 2015. Then compare it with 2019 and 2020 images in Photoshop. Changes line up fairly well with GFW estimates from satellite images but we can also see the limits low-resolution images introduces.
We can also double check on another area, much larger this time. Logging occurred on property owned by a pellet manufacturer.
Activities have another story to tell. If large swathes of trees are felled near human settlements, risk increases that sawdust can pollute the air for nearby residents. Sawdust can be carcinogenic. Occupational wood dust exposure can cause asthma and dermatitis. The British HSE has its own page to warn the public and companies.
The felling shown below only took a few months to execute and cut between April and July of 2019. It’s less than 2km from one thousand residents-strong town of Vaida and a mere stone throw away from residential housing (see illustration).
Experts also worry about birds and their offspring. In an area that is an eldorado for bird species, such as Haanja Nature Park, felling is particulate bad for biodiversity when companies clear areas during bird breeding (nesting) season — which for Estonia starts as early as mid-March and lasts until mid-July. For the example above and other instances found, companies disobeyed the rule to cut forest during these sensitive months. It isn't illegal but questionable from an ecological standpoint, biologists say.
Where does wood pellet corporations own land?
You’d want to understand where a specific wood pellet company owns properties. A dataset from the Estonian Land Registry contains property codes for areas owned by one of the largest Estonian wood pellet manufacturer, mostly via a number of subsidiary holdings the company calls ‘sister companies’. Each property has a ‘katastritunnus’ number, a specific property identified that can be found also in Estonia’s national database of land ownership.
Each of such Katastritunnus link to a specific property. We also have the size of each property expressed in hectares. Now to map them out we can use the government’s boundary file available for anyone to download online.
When downloaded, it’s available as a shapefile, machine-readable by QGIS. Each polygon feature has this katastritunnus id, mentioned before. Now, all we need is to tell QGIS we want to filter polygons according to the company’s katastritunnus that we have in the datasheet.
We can add the EU Natura 2000 boundaries and we get a pretty solid set-up to investigate forest-cover-change that took place on these properties.
If you are only interested in the intersection between the company’s properties and EU Natura protected areas, then we can run a script that filters out only (red) areas that sit within green protected zones. If we put it all together, we can use the data visualisation platform DataWrapper to visualise some of the results (below).
You can now export the intersecting property areas as a KML file and take it into Google Earth Pro to inspect changes over time, one by one. To investigate and to avoid errors (see the sections about wildfires, windfall cases etc), that’s a suitable way forward for OSINT journalists.
In short, it can lead you to specific case studies and provided a way to locate the area that we see below: property on EU Natura protected land in Haanja Nature Park and owned by a company that makes wood pellets for the export market.
To further investigate each property with our reference IDs, we can check the Land Registry database, a Flash website (Yuk, what an awful interface). The tool allows searching with our property number (for the example above: 18101:001:0680). We get more information and can cross-check our findings.
Now, in order to pin down exactly the days when we lost forest cover, we need more time-sensitive images that Google Earth can’t provide. It encouraging that the cleared areas are large enough to be spotted from Sentinel 2 images (with a spatial resolution of 10 metres for its four visible and near-infrared bands, 20 metres for its six red edge and shortwave infrared bands and 60 metres for its three atmospheric correction bands). Sentinel Hub EU Browser online tool offers you less precise images but more images covering the time period in question (see below).
Though possible for commercial companies to cut forest in EU protected areas, there are restrictions that apply, such as cutting no more than a few hectares each time. These granular Sentinel 2 (or commercial Planet inc data) satellite data can show you more reliably if and when companies may overstep their allowance.
The impact of deforestation on tourism
Lastly, I want to share my thoughts on forest loss and it’s economic impact. Economic studies are yet to determine whether and how much harm logging causes to the country’s tourism sector known for its ‘eco-tourism’— there are a number of confounding factors including Covid-19. But one sign of concern is a statement by Peeter Raudsepp, the head of Enterprise Estonia (EAS), an Estonian tourism industry group. He warmed that felling could harm tourism sites that are “advertised as a wildlife destination with beautiful nature pictures”. He said that those who spent the New Year in Otepää notice the felling in Otepää Nature Park. People have to move between felling sites while walking along the Kekkos trail and that interests of the forest industry don’t override those of tourism, Raudsepp says.
Last words
The tools and techniques mentioned here can be useful elsewhere, whether it’s in Romania, Hungary, or the Amazon rain forest.
On investigations of the EU biomass industry, I word of warning. Most of what you’ll claim will be challenged by corporate and lobbying groups. The powerful biomass lobby in Europe proved to be very sensitive when we published and opted for criticism right away. Any journalist should prepare from that. We received a number of emails by lobbying groups and biomass operators. One energy giant claimed that I didn't send an email of enquiry (for an interview/comment) when I send it to three media contact at the press department. Others who challenged the biomass industry with findings made ostensibly similar experiences.
A Chatham House expert told me that when he published a report on biomass some time ago he perceived explicit pressure from a large British biomass company. In Estonia, a biologist who found that much of the ‘forest management’ was breaching rules, was threatened with legal actions by a local wood pellet manufacturer when speaking out on foreign television.
