From Historical Archives to Algorithms: Reconstructing Biodiversity Patterns in 19th Century Bavaria

Abstract

Historical archives hold untapped potential for understanding long-term biodiversity change. This study introduces computational approaches to historical ecology, combining archival research, text analysis, and spatial mapping to reconstruct past biodiversity patterns. Using the 1845 Bavarian Animal Observation Dataset (AOD1845), a comprehensive survey of vertebrate species across 119 districts, we transform 5400 prose records into structured ecological data. Our analyses reveal how species distributions, habitat associations, and human–wildlife interactions were shaped by land use and environmental pressures in pre-industrial Bavaria. Beyond documenting ecological baselines, the study captures early perceptions of habitat loss and species decline. We emphasise the critical role of historical expertise in interpreting archival sources and avoiding anachronisms when integrating historical data with modern biodiversity frameworks. By bridging the humanities and environmental sciences, this work shows how digitised archives and computational methods can open new frontiers for conservation science, restoration ecology, and Anthropocene studies. The findings advocate for the systematic mobilisation of historical datasets to better understand biodiversity change over time.

1. Introduction

Understanding historical changes in biodiversity is essential for addressing present and future environmental challenges. Archival records provide a rich, albeit often underutilised, window into past ecosystems, enabling researchers to trace species distributions, population trends, and human–animal interactions long before the advent of modern ecological surveys [1,2,3,4,5,6]. The 1845 Bavarian Animal Observation Dataset (AOD1845) [7] represents one such resource—an extensive and unique survey conducted across the Kingdom of Bavaria in the middle of the 19th century that captured vertebrate presences and absences in remarkable detail.

Compiled from the reports of foresters across 119 districts, the dataset offers rare insights into the pre-industrial landscapes of Bavaria. These accounts, while qualitative in nature, convey the impacts of land use change, deforestation, and habitat degradation. A particularly telling example is provided by senior forester Lorenz Reber, who commented on the local decline in duck population following the drainage of ponds (References to the AOD1845 dataset are given by entry IDs. Translations from the original German sources by the author):

Since the desolation of several ponds, especially the draining of the Pfrentschweiher, the ducks are very few and rarely come to the few small ponds during the migration season.

(E_00968, Vohenstrauss, Upper Palatinate)

Drawing on the AOD1845 dataset, the study combines archival research, Natural Language Processing, and spatial mapping to reconstruct historical biodiversity patterns. By analysing over 5400 original entries, we explore the ecological and cultural dimensions of 19th-century species distribution, the influence of human activity on animal habitats, and the interpretative biases embedded in the archival record. The analysis culminates in case studies of key species—including the Eurasian otter, beaver, and capercaillie—highlighting how historical baselines can inform modern conservation science and policy. Ultimately, the paper advocates for the systematic digitisation and computational analysis of historical biodiversity data as a means to enrich long-term ecological understanding.

2. Materials and Methods

2.1. The 1845 Survey

2.1.1. Historical Background

The Bavarian State Archives (Bayerisches Hauptstaatsarchiv, BayHStA) preserve a set of files associated with the Zoological State Collection (BayHStA, Zoologische Staatssammlung, 208–217). These files mainly consist of the responses to a detailed and systematic survey on 44 selected vertebrate species, officially conducted by the Ministry of Finance in the Kingdom of Bavaria in summer of 1845: “by His Majesty the King’s highest command” to gain “knowledge of the geographical distribution of animal and tree species in Bavaria—from a scientific point of view” [7]. The four-page questionnaire, sent out by the Ministry to its subordinated 119 forestry offices (Forstämter) [8,9], yielded over 5400 statements in short prose by professional foresters concerning animal occurrences and distribution throughout the entire country, covering an area of roughly 77,000 km². An overall of 520 pages in nine files were preserved by the State Collection in Munich and transferred to the Bavarian State Archive in 2013. Only in 2024, the material was fully catalogued, finally digitized and made publicly accessible online.

This survey was scientifically overseen by Johann Andreas Wagner (1797–1861), then deputy head of the Bavarian State Zoological Collection [10,11]. Wagner was a distinguished zoologist [12], full professor at Munich University, with an extensive publication record in zoology and palaeontology [13], though later a prominent repudiator of Darwin’s emerging evolutionary theory. Among Wagner’s scientific contributions are annual reports on developments in global research on mammals and birds [14,15] and his description of the geographical distribution of mammals, published in three parts between 1844 and 1846, in which Wagner depicts the mammalian fauna of Africa, tropical America, Australia, and the Magellanic Province. Wagner’s stated goal was to contribute to understanding the “laws according to which organic beings are distributed over the earth” ([16], p. 4).

In 1846, Wagner outlined both his personal motivation and his methodological approach behind a systematic faunistic survey of Bavaria the year before, as follows:

Honoured by His Royal Highness the Crown Prince of Bavaria with the commission to attempt a representation […] from the Bavarian fauna on a larger map, I could not conceal how much is still missing, even with such a limitation of the task, for its satisfactory solution at present. A description of the species would only be complete if the fauna of a sufficient number of localities in our country were known and recorded. […] In order to fulfil the highest commission as far as possible, I have requested contributions from various sources for its execution, and in particular I have obtained notes from all the royal forestry offices of the kingdom. In particular, I have obtained notes on the occurrence of the most important animals in their districts.

([17], p. 649)

Wagner chose 44 vertebrate species (16 mammals, 27 birds, 1 reptile) for the survey (see the AOD1845 dataset description for the full list) [18]. Why Wagner made this specific selection remains unexplained, though. According to his own words, these were “the most important animals” (Vorkommen der wichtigsten Thiere), but it is unclear how he decided what was important and what not. Some species were probably considered remarkable then as now, as the brown bear or the wolf, and publicly discussed whenever they occurred. Some species on the other hand, such as roe or red deer, wild boar, and some birds, were apparently associated with hunting and poaching. They can be viewed in conjunction with the fundamental reform of hunting laws implemented in the revolutionary period in the middle of the 19th century: hunting as a privilege and a source of income and hence subject to fiscal and governmental purposes. But it was as well a subject of a power struggle: Maximilian, King of Bavaria since 20 March, 1848, was the first monarch in the German Confederation to ban feudal hunting on foreign land, on 4 June of the same year. Wagner’s survey—initiated by Maximilian three years earlier, while he was still crown prince—may also have served as a preparation for the implementation of such stately issues. Hence, apart from scientific interests, the survey of 1845 can also be seen and understood in the context of political turbulence of that revolutionary time [19].

2.1.2. The Responders

The intended recipients of the survey—and likely its actual respondents—were the heads of the 119 forestry districts. By the early 19th century, these were typically held by senior civil servants with formal education: a three-year-long advanced course at the forest academy in Aschaffenburg or the University of Munich, both institutions of considerable international repute [20]. Zoology formed part of this training [21], indicating that foresters possessed at least the basic zoological knowledge necessary to answer the questionnaire competently, although people in office in 1845 might have received their training much earlier in the century. The period also saw a general increase in public knowledge about wildlife species [22].

The first section of the questionnaire provided instructions on how to respond, listing 44 fields—each led by the name of a target species and headed with “Place of residence and other remarks“ (Wohnort und sonstige Bemerkungen). These brief instructions also conveyed Wagner’s expectations for the survey:

List of animal species of whose existence and place of residence information is desired. Note: If the distribution is limited to certain localities only, the nearest village or locality should be included in parenthesis. In the case of birds, only those which breed in the district itself or which spend the winter there, or which are currently the subject of hunting, should be listed.—It should also be stated whether the species is common or rare.

The survey achieved a 100% response rate—likely due to the officers’ obligation to comply with royal administrative duties. However, regarding substance and extent, quality and quantity, the responses varied significantly. Figure 1a shows that the distribution of response lengths ranges from a terse 74 tokens in Berchtesgaden (Salforste) to a notably verbose 657 tokens in Haag (Upper Bavaria). Figure 2 gives a visual impression of this range using the example of the questionnaire sheets from neighbouring forestry offices of Berchtesgaden and Tegernsee, which are at the lower and upper end of this range, respectively. One pragmatic reason for foresters to respond only quite briefly might be the fact that, at least in Lower Bavaria, they were given only 14 days to reply (StA Landshut, Akt der Kammer der Finanzen Niederbayerns, A_357, 7). But one might also speculate about other reasons apart from the personality of the responder [23].

2.1.3. Place and Time

Prior to the unification of Germany in 1871, the Kingdom of Bavaria was an independent state with an area of 76,770 square kilometres—almost twice the size of modern-day Netherlands [21]. However, it is not only the size of the territory that renders the AOD1845 dataset significant. Bavaria’s ecological richness lies in its diverse landscapes and habitats. Wagner himself emphasizes the horizontal and vertical diversity of the country. Including the so-called Salforste [24], whose area of responsibility also includes parts of the Austrian Alps, our study area covers two of the three biogeographical regions of Germany (the Continental and Alpine, but not the Atlantic region [25]) and 21 of the 24 modern landscape types classified by the Federal Agency for Nature Conservation of Germany [26]. Elevation across Bavaria ranges significantly—from as low as 85 m above sear level in the Upper Rhine Graben near Ludwigshafen to 2962 m at the summit of the Zugspitze (Wetterstein Mountains), Germany’s highest mountain. This topographical diversity results in a total elevation of 2877 m.

In the mid-19th century, Bavaria would be commonly characterised as an agrarian and forested state. The population density in 1845 stood at 57.84 persons per km² (see

Table 1. Area, woodland share, and population of the Bavarian governmental districts in 1845 as well as some key figures from the dataset. Note that the Salforste are not governmental districts and area or population data are not available.

District	Total	Woodland			Population		Species
	Area a	Area a	Share b	Private c	Capita	Density	Count d	Average e	Extra f
Upper Bavaria	17,143.4	5990.4	34.9%	46.0%	694,344	40.5	38	21.5	16
Lower Bavaria	10,700.4	3746.6	35.0%	72.9%	535,499	50.0	37	17.0	26
Upper Palatinate	9610.5	3000.7	31.2%	40.0%	463,187	48.2	33	18.8	68
Upper Franconia	7010.5	1691.7	24.1%	29.5%	496,783	70.9	35	17.9	14
Middle Franconia	7581.3	2540.0	33.5%	44.0%	518,478	68.4	34	17.1	9
Lower Franconia	9333.0	3048.6	32.7%	8.6%	587,887	63.0	37	20.8	12
Swabia	9575.8	2420.0	25.3%	36.8%	548,956	57.3	40	20.6	38
Palatinate	5815.4	2122.7	36.5%	5.8%	595,193	102.3	34	17.9	4
Salforste	n/a	n/a	n/a	n/a	n/a	n/a	35	23.7	13

^a In km². One Bavarian mile was 7419.5 m in 1845 ([29], p. 253). ^b Of total area (SD = 4.6). ^c Percentage of woodland area in private property (SD = 21.5). ^d Number of species (out of 44) present within the governmental district. ^e Mean average number of species (out of 44) present within the forestry offices of the governmental district. ^f Number of additional species reported to be present. Source: ([21], pp. 18–19) and own analysis of AOD1845.

)—a figure that had only risen to 90.8 by the eve of the First World War. This remained significantly below the population densities of neighbouring Württemberg and Rhineland, and markedly behind the more industrialised Kingdom of Saxony, in both absolute and relative terms ([27], p. 50). As of 2023, Bavaria’s population density stands at 190.0 [28].

Nonetheless, describing Bavaria as purely agricultural obscures the beginnings of a selective industrialization that gained momentum around the time of the 1845 survey [30]. Significant industrial centres emerged in Augsburg, Nuremberg/Fürth, in Upper Franconia, in Munich, as well as the Maxhütte steelworks in the Upper Palatinate. The first railroad on German soil ran between the Central Franconian cities of Nuremberg and Fürth from 1835. With the Ludwig South-North Railway between Lindau on Lake Constance, the southwestern and the Upper Franconian Hof on the northeastern border of the country, and the Ludwig Canal between Kelheim and Bamberg, which was completed in 1846, crossing the main European watershed, enormous infrastructure projects were implemented. Understanding the ecological implications of these developments is crucial for any historical analysis of biodiversity.

The 1845 survey also coincided with the final decades of the Northern Hemispheric Little Ice Age [31]. Some responses explicitly referenced climatic conditions. While only two cited selective warmth, thirty-six explicitly referred to a cold climate and harsh winters as having influenced animal populations. Deer, primarily, were frequently linked to winter severity in statements such as the following: “The not infrequent harsh winters do not allow for a good deer population, which is why this game is rare here“ (E_02268, Goldkronach, Upper Franconia). Other entries noted the impact of winter conditions on migratory birds, such as wild geese, and on predators such as wolves and lynxes, whose movements during the particularly severe winter of 1844–45 were reported across several districts.

2.1.4. Usages of the Survey

Andreas Wagner shared some outcomes from the 1845 survey in a lecture series to the Bavarian Royal Academy, delivered in 1846. This report was later published in the Academy’s proceedings [17]. Aside from this lecture and a map he commissioned based on the results [32], the survey appears to have received little attention in subsequent literature. To our knowledge, it has rarely been cited, and even more rarely analysed in detail.

Some references to Wagner’s findings appear sporadically in the ethnographic compendium Bavaria. Landes- und Volkskunde des Königreichs Bayern [33], which was published between 1860 and 1867 in eight volumes. However, these mentions are infrequent and unsystematic. The contributors did not engage directly with the primary sources of the survey, nor did they offer detailed discussion of Wagner’s original report.

2.2. The Data Set

The AOD dataset used in this study was createad by “datafying” [34] the original archival sources held in the Bavarian State Archives. This involved transcribing, annotating, and classifying over 5400 brief prose hand-written descriptions submitted by the foresters and compiled by Wagner [18].

This dataset enables the analysis and mapping of selected vertebrate species across Bavaria, offering insights into biodiversity prior to industrial transformation of the landscape. Figure 1b and Figure 3 present an example analytical outcome derived from the dataset: the distribution of the number of positively reported species across the country. For instance, the only recorded sighting of a brown bear (Ursus arctos) in the period leading up to 1845 originates from the Tegernsee office, located in the Bavarian–Austrian Alps:

The bear appears singly, probably dispersed from Illyria, but very rare in the Bavarian high mountains: the last one was from 1826–28 partly in the local, partly in the neighbouring Tyrolian mountains, where it was shot.

(E_01278, Tegernsee, Salforste)

Similarly, the dataset documents the final sightings of the Eurasian beaver (Castor fiber) in Bavaria prior to its extinction around 1867 [35]. Reports mention its presence along, e.g., the river Amper (“Only very few families can be found along the river Amper”, E_04904, Freising, Upper Bavaria), as well as its disappearance from former habitats (“They used to be common on the banks of the Danube, but have been completely lost in recent times”, E_00185, Neustadt, Lower Bavaria; “The last one was shot 20 years ago at Kratzmühle an der Altmühl, Kipfenberg district court”, E_01554, Eichstätt, Swabia).

These textual descriptions offer insights into landscape ecology, the impact of human activity, and 19th-century perceptions of the environment. For example, in the Rhine-Palatinate administrative district, foresters linked the decline of the common snipe (Gallinago gallinago) to habitat loss: “rare due to the loss of wet meadows” (E_03168, Pirmasens, Palatinate) and, even more clearly, “[It is] seldom since the meadows have been drained everywhere” (E_03300, Winnweiler, Palatinate).

For the case studies that follow, computational tools were applied to the AOD1845 data to generate exemplary findings. While full details are documented in the dataset publication [18], a brief overview follows: AOD1845 comprises 5467 entries, each representing information about one of the 44 species within one of 119 forestry offices. Due to additional species reported by some foresters, the total number of entries exceeds $44\times 119=5236$. Three reports from the sources were excluded: a second report from Landau (Isar) and two reports from senior foresters of Thurn and Taxis and the Löwenstein/Kreutzwertheim private estates. The dataset also excludes data on fish.

Each record of the dataset is encoded in binary format to indicate presence or absence. Geographic coordinates (longitude and latitude) of the administrative seat of each forestry office enable rough spatial analysis and mapping. Links to digitised images of original sources and their transcriptions are embedded, organised by office and species.

Figure 4 illustrates how a typical archival statement was transformed into dataset form, using the example of the Eurasian otter reported from Deggendorf.

2.3. Enriching and Annotating the Data

The text field of the AOD1845 dataset contains rich yet unstructured qualitative information. To enable deeper analysis, we annotated specific features of the text and classified particular statements within it. These included the following:

• Habitat descriptors;
• Geographical names (toponyms);
• Phrases or expressions that indicate mankind’s ecological impact or human–nature interaction, relation, and understanding;
• Categorizable statements about population size: ABSENT, EXTINCT, VERY RARE, RARE, COMMON TO RARE, COMMON, ABUNDANT, NOT QUANTIFIABLE;
• Categorizable projections about population development: INCREASING, STEADY, DECREASING.

Table 2. Sample textual entries and their annotations and classifications.

ID	Text	Habitat Descriptors	Toponyms	Human-Nature	Quantifiers	Population Development
					Inferred Classification	Inferred Classification
E_00005	[The otter:] In all relevant rivers and creeks, but is also becoming increasingly rare. More are caught or shot in and near the creeks of the Bavarian Forest where they come from the Danube than on the larger rivers.	‘rivers’; ‘creeks’	‘Bavarian Forest’; ‘Danube’	‘caught’; ‘shot’	‘in all’	‘becoming increasingly rare’
					COMMON TO RARE	DECREASING
E_00047	[The beech marten is] very common in barns and buildings in all localities, also in dense young forests. Lives under walls and in hollow trees.	‘barns’; ‘buildings’; ‘dense young forests’; ‘walls’; ‘hollow trees’	—	—	‘very common’	—
					ABUNDANT	—
E_00147	[The roe deer is] still everywhere in the forests, but is quite diminished by unfavourable weather conditions, especially during the winter of 1844/45.	‘forests’	—	—	‘still everywhere’	‘quite diminished’
					COMMON	DECREASING
E_00930	On the south-western border of the forest district, starting at the Linglmühle mill as far as Loma, into the Pfreimd river outwards to the Pfrentschweiher pond, and from Loma to the Bohemian border in Zottbach, the otter seeks its food and residence.	‘river’; ‘pond’	‘Linglmühle’; ‘Loma’; ‘Pfreimd’; ‘Pfrentsch- weiher’; ‘Zottbach’	—	—	—
					COMMON	—
E_00968	Since the desolation of several ponds, especially the draining of the Pfrentschweiher, the ducks are very few, and rarely come to the few small ponds during the mating season.	‘ponds’; ‘small ponds’	‘Pfrentsch-weiher’	‘draining’	‘very few’; ‘rarely’	—
					RARE	—

presents examples of annotated entries. Some information was explicit (e.g., “lives under barns”), while some required inference (e.g., interpreting “in all” as indicating widespread presence). We manually annotated features 1–3 across the full dataset. We restricted the annotation to selected species—roe deer, beaver, capercaillie, and otter for feature 4 and capercaillie for feature 5—for case study purposes.

To prepare for textual analysis, we normalized textual features 1–3 as well as text of the responses. We also removed stop words, using the spaCy Natural Language Processing (NLP) library and its de_core_news_lg model for German language (https://spacy.io/models/de/ (accessed on 8 April 2025)). For further statistical testing, we then counted tokens for each textual feature and calculated mean averages of the text lengths per species and per reporting office.

Species data were enriched by the following nominal classification scheme:

• Class: mammalia (mammal), aves (bird), reptilia (reptile);
• Conservation status according to the German Red List: EXTINCT OR LOST, THREATENED WITH EXTINCTION, HIGHLY THREATENED, THREATENED, NEAR THREATENED, EXTREMELY RARE, NOT THREATENED;
• Whether the species is commonly subject to human hunting: yes or no;
• Whether the species was associated with aquatic habitats: yes or no.

The annotations are publicly available alongside the dataset [36]. For the statistical analyses that follow, we excluded the brown bear and the ducks from the data; the bear because of its singular only appearance, the ducks because Wagner asked here about the entire bird family, rendering these entries incommensurable with the specific species of the rest.

2.4. Textual Analysis

The questionnaire requested information not only on the occurrence status of species but also on their “places of residence” (Wohnorte). Many foresters complied with this request, but their responses varied significantly in how this information was provided. This variability is a distinctive feature of the survey and necessitates particular caution when applying text analytical methods to the dataset.

Consider examples E_00047 and E_00147 from Table 2. We used textual data of this kind to find patterns of similarity among species. Both examples describe species presence, but they differ in how much habitat detail is given. While E_00047, concerning the beech marten, contains several such descriptors that make up more than a third of the whole text (8/21 = 38.1%), E_00147, describing the roe deer, contains just one token out of 17 overall (5.9%). Due to unbalanced weighting of such habitat descriptors across the data and the overlay of too much non-habitat-related information as in E_00147, a similarity analysis over the complete texts did not yield any meaningful results.

We hence used only the habitat descriptors (not the full text) for analysis. We set up a text analysis pipeline involving the following steps:

• Extract habitat descriptors from the text;
• Group texts by forestry office or species;
• Apply term frequency–inverse document frequency (TF-IDF) analysis, which yields a complex vector representation for each of the 119 offices and the 44 species, respectively;
• Reduce dimensions using Principal Component Analysis (PCA);
• Cluster results using k-means (with cluster number determined by the elbow method);
• Visualise the output using colour-coded plots.

Additionally, quantitative data were generated from the lengths of the textual responses assigned to the three annotated categories in Section 2.3. These were then brought into relation with other metadata about species and offices. A one-way analysis of variance (ANOVA) [37] was used to triangulate between the quantitative text metrics and the qualitative classification in categories 4 and 5. This enabled a mixed-methods approach that connected textual nuance with ecological interpretation [38,39].

3. Results

3.1. Describing Animal Presence and Absence in 1845

The historical analysis reveals that, while the foresters generally possessed basic zoological knowledge, the quality and depth of their response varied—also reflecting personal interest in the regional fauna or particular species. This seems to be particularly true for animals not central to forest industry and hunting concerns.

As seen in examples, foresters such as Jak. Reverdys in Berchtesgaden appear to have fulfilled their duty with minimal effort (cf. Section 2.1.2), offering short, generalised statements—for instance, describing the badger (Meles meles) as “to be found more frequently” and the stone marten (Martes foina): “likewise” (E_01371 and E_01372). In contrast, Max Schenk of Tegernsee provided richly detailed accounts. His report on the badger included multiple named mountains and valleys where the animal was observed: “More frequent in some years, less frequent in others on the foothills /: Holzeberge, Brand, Kogel district Tegernsee, Enknel[?] [and] Schinderberg, Breitenbachthal, Rohnberg [and] Leitnergraben, district Miesbach”. For the stone marten, he noted: “There used to be many animals of this species in the villages as well as on the farmsteads; - gradually they have diminished so much that they are now quite rare” (E_01279 and E_01280). Tegernsee, therefore, exemplifies detailed geographic and temporal reporting, while Berchtesgaden sits at the opposite end of the spectrum.

The Wernberg report in Upper Palatinate is another case of engaged and knowledgeable authorship. Heinrich Drexel also demonstrate his involvement and interest in nature, environmental issues, and state-of-the-art science. He lists 46 additional animals—mainly birds—and provides supplementary descriptions, including scientific names and distinctions between migratory and breeding species on an extra page. By its accuracy, the Wernberg report also supplements (and contradicts) the sparse report from the adjacent Vilseck office at one point. While Jos. Zölch from Vilseck marked the common crane (Grus grus) as absent with a simple “—”, Drexel noted: “Is not to be found here, but in the neighbouring forest district of Vilseck /: Rödelweichermoos :/ however rarely.” (E_01136, Wernberg; E_00916, Vilseck).

The variation in quality is also evident when comparing the average word count per species. Excluding ducks (as this question was about a whole family and motivated some officers to list and describe particular species within this family), it ranges from an overall of 313 tokens for the great bustard (Otis tarda) to 1471 tokens to describe the European otter (Lutra lutra). As expected, there is a strong correlation (r = 0.94) between the overall text length per species and the number of offices in which it was reported to be present—people wrote decisively more about the species that were present in their area than about those that were absent.

Statistical analysis also reveals other notable patterns (see

Table 3. Pivoted summary table of one-way ANOVA by F-distribution (N = 42) ^a.

Group	Class	Threat Status b	Occurrences c	Hunt	Aquatic
Geographical names (mean)	1.44 (0.248)	1.18 (0.341)	0.44 (0.946)	7.29 ** (0.010)	0.01 (0.921)
Habitat descriptors (mean)	2.03 (0.145)	0.36 (0.920)	0.88 (0.633)	0.05 (0.824)	0.37 (0.545)
Human Influences (mean)	15.50 *** (0.000)	0.76 (0.621)	1.71 (0.236)	4.18 ** (0.047)	1.51 (0.226)
Full textual description (sum)	1.03 (0.367)	0.95 (0.483)	21.09 *** (0.000)	1.84 (0.182)	0.05 (0.823)

^a Significance thresholds: *** (p < 0.01), ** (p < 0.05). ^b According to the German Red List (as of 2024). ^c By the number of offices that report presences of the species.

). Species that were hunted were described with significantly more geographic detail (1.93 vs. 0.92 average tokens) and human–nature interaction (0.18 vs. 0.08). The descriptions of mammals also contained more references to human influences than those of birds (0.23 vs. 0.05). However, no clear correlation was found between response detail and the modern conservation status of species, suggesting that predicting today’s status categories from the 19th-century perceptions alone is not easily feasible.

3.2. Species Distribution

Figure 3 maps the location of all 119 forestry offices, with the colour of each marker reflecting the number of species reported as present in that district. This basic statistic takes the binary classification of the occurrence statuses for each of the 44 surveyed species per office. It does not account for quantitative remarks on abundance statuses, however (see Section 3.5 for those).

Despite this limitation, the map offers a preliminary indication of biodiversity patterns within the study area. In particular, the Alpine region (Salforste as well as the southern parts of Upper Bavaria and Swabia) stands out for high reported species richness. Some endemic or range-restricted species, such the marmot and chamois, were only observed in these mountainous areas. The sole recorded bear sighting also originates from this region.

However, in the administrative district of Lower Bavaria, located in the southeast, an important exception to this statement must be acknowledged. Lower Bavaria has an exceptionally high proportion of privately owned forest (see Table 1), which likely fell outside the scope of the survey. As a result, the underrepresentation of species in this area may reflect incomplete data coverage rather than actual lower biodiversity.

3.3. Species Reporting

In the following, we explore correlations between the present-day threat level—according to the categories given by the German Red List [40]—and the species’ distribution patterns recorded in the 1845 dataset. However, caution is required: the number of positive reports alone does not provide a reliable measure of population size or conservation status. A species may be widely reported, yet rare in each location.

This is illustrated by comparing the roe deer (Capreolus capreolus) and the Eurasian otter (Lutra lutra), both of which were reported as present in nearly all districts. However, as Figure 5 shows, the otter’s presence was often classified as rare, suggesting it was already under pressure in 1845—unlike the deer, which was more frequently described as common.

Conversely, some species that were reported only rarely in 1845 are not considered threatened today. The marmot, for example, appears in just five reports but is a habitat specialist adapted to Alpine regions, which are limited in spatial extent. Thus, a low number of reports does not necessarily imply a historical threat.

Despite these caveats, a z-score analysis (threshold: 1.5) reveals that certain species are widely reported in the 1845 survey but are threatened with extinction in 2024. These include the common snipe (104 reports), black grouse (82), capercaillie (74), and hamster (27). A similar deviation can be seen for the otter (116 reports in 1845 as discussed above, today threatened) and the common sandpiper (118 reports in 1845, today classified as extremely rare). Likewise, some species with low report numbers in 1845—such as the mute swan (5) or the wild boar (22)—are not currently threatened. These examples underline the complexity of interpreting historical data and the importance of contextual ecological knowledge.

3.4. Describing Habitats

Running the text analysis pipeline described in Section 2.4, we first grouped all survey entries by species and applied clustering to the extracted habitat descriptors. The analysis yielded six distinct clusters with a silhouette score of 0.57, indicating reasonably clear groupings (see Figure 6).

The results reveal that habitat similarity does not necessarily align with taxonomic proximity. For instance, the beech marten and the European pine marten—both members of the same genus Martes—do not share the same cluster, indicating that their habitats are different. In contrast, the beech marten clusters closely with the white stork, highlighting that both species favour proximity to human settlements. This is evident in the words typically used to describe their “places of residence”, which often mention buildings, villages, or human-made structures.

Birds, being ecologically diverse, appear scattered across all clusters. However, aves species that prefer aquatic habitat—such as waterfowl—are clearly clustered, alongside aquatic mammalia like the Eurasian beaver or the Eurasian otter. This and the other observations align with modern biological knowledge, but one should keep in mind that these results are based purely empirically on single observations brought together to display the manner in which habitats were perceived and described by people in 1845.

Deer species—roe deer, wild boar, red deer, and fallow deer—also form a tight grouping, reflecting their shared forest-based habitats and roles in traditional hunting practices.

We applied the same clustering approach to entries grouped by forestry offices, again using habitat descriptors. The clustering is even slightly better as with the species (silhouette score: 0.71). Instead of analysing the cluster by the offices’ names, however, we now enriched our data by the geographic coordinates of the offices’ seats and projected them on a map (Figure 7). A tendency becomes visible: Cluster 2 (green) largely covers the Alpine region and partly the natural areas of the Bavarian and Upper Palatinate forested mountains in the east, but the other two clusters are hardly distinguishable based on basic geographic information.

3.5. Species Studies

3.5.1. The Eurasian Otter (Lutra lutra)

The Eurasian otter stands out in the dataset for the richness of contextual information provided across reports. It is one of the most frequently annotated species, with detailed references to geography, habitat, and human impact. The otter was also reported as present almost country-wide—abscent only in Kirchheim, Zweybrücken (both Palatinate), and Markt Einersheim (Middle Franconia). However, as noted in Section 3.1, its frequency classification skews towards rare presence, indicating that although widespread, it was already declining in many areas.

Today, the Eurasian otter is classified as “threatened” on the German Red List, with a sparse distribution and a projected long-term population decline [41]. Data from the Global Biodiversity Information Facility (GBIF) record only 59 sightings in Bavaria and (modern) Rhineland-Palatinate between the years 2000 and 2024 [42], compared to, for example, 3755 sightings for the Eurasian beaver in the same period and area [43].

Figure 8 overlays the 1845 report with modern GBIF sightings. Contemporary observations are concentrated in the Alps, the southeastern, and the northeastern parts of the country, partly overlapping with the historical range. However, many areas of historic presence, even those classified as abundant—such as Neustadt an der Saale (Lower Franconia), the Vohenstrauss region in the Upper Palatinate, or parts of central Bavaria—now show no recent sighting. Particularly, in the Rhön Mountains of the country’s north, the forester gave a detailed report about the geographies of the otter’s presence in 1845: “On the [river] Saale /: between Kissingen and Neustadt :/ on the [river] Brend /: between Neustadt and Bischofsheim :/ on the [river] Streu :/ from Heustreu up to Fladungen :/ quite common; also in the Elzbache” (E_04503). For none of these areas, a recent sighting is recorded in GBIF.

Of the 116 positive reports from 1845, 95 include detailed habitat descriptions—ranging from general references to streams and rivers to named water bodies. This allows for precise mapping of past occurrences (Figure 9). One example is the forestry district of Vohenstrauss (Upper Palatinate), where the otter was said to frequent a sequence of streams and ponds including the historic Pfrentschweiher (east of the village Pfrentsch). This same pond also features in reports of declining duck populations (see Section 1). Historical maps from around 1836 (the so-called Uraufnahme, cf. ([44], pp. 31–45)) confirm the existence of the Pfrentschweiher, now largely converted to agricultural land. This shift in land use likely contributed to the disappearance of both ducks and otters in the region.

3.5.2. The Wolf (Canis lupus)

The wolf belongs to the rarest species recorded in Wagner’s survey, with only eleven districts reporting occasional occurrence. Ten of these were located in the western Rhine Palatinate (Figure 10). The single exception comes from the office of Selb in the Upper Franconian Fichtel Mountains, where the presence of wolves was attributed to the particular harsh winter:

With the exception of the winter of 1844/45, when two of these were felt for a long time, it has not been noticed for more than 100 years.

(E_02482, Selb, Upper Franconia)

Nine additional offices reported the wolf as extinct, with estimated disappearance dates ranging from around 1650 (E_02261, Goldkronach, Upper Franconia) to a more recent 1825 (E_04330, Heidingsfeld, Lower Franconia).

The only district where wolves were described as appearing regularly was Homburg (Palatinate):

When the snow is deep and lasts for a long time, a few wolves appear almost every winter from the neighbouring Prussian high forest, even sometimes in summer when they are being pursued on the other side. Their stay is usually very short, however, because they are immediately pursued here too.

(E_02878, Homburg, Palatinate)

This account not only suggests migration triggered by harsh winters but also highlights potential differences in wildlife management between Prussia at the western border and the Bavarian Palatinate. Six Palatinate reports refer to the wolf as Wechselwild (migrating game), as opposed to Standwild (resident game). These sightings are frequently tied to the severe winter of 1844–45.

As many reports mention both origin and direction of movement, we were able to reconstruct approximate migration routes (Figure 11), providing a rough sketch of 19th-century dispersal dynamics.

3.5.3. The Eurasian Beaver (Castor fiber)

The beaver has long been a subject of interest in human–animal studies and historical ecology, owing to its close interactions with humans and its cultural, ecological, and economic relevance. In addition to being hunted for its fur and the so-called Castoreum (a secretion once valued in medicine), the beaver was at times classified as a fish to permit its consumption as fasting food during Lent [45,46]. It was also viewed as an “unwanted visitor” and nuisance species due to its damage caused on trees and agricultural land [47].

In recent decades, however, its ecological value has been recognized increasingly [48], and successful attempts have been made to reintroduce it across Europe including our study area in a controlled manner [47,49].

Despite this renewed attention, relatively little is known about the historical occurrence of the beaver [5]. Among historical studies, Sander Govaerts’ examination of the medieval Netherlands is particularly noteworthy. Govaerts reconstructed the beaver’s presence using administrative records. For Bavaria, Volker Zahner compiled and mapped historical occurrences using corresponding place names and toponymic evidence [35].

Zahner’s chronological description dates the extinction of the beaver in Bavaria approximately to the year 1867 ([35], p. 15), making Wagner’s survey probably the final evidence of its presence in Bavaria before its reintroduction in the later 20th century. In 2024, GBIF recorded 184 sightings in Bavaria [50].

In 1845, the beaver was still present in 16 forestry districts–though always described as rare or very rare. The Munich report, which, like others, indicates the decline of the beaver’s population, can be seen as typical: “The same appears only as a great rarity on the banks of the Isar towards Freysing, but without really staying there” (E_05036, Munich, Upper Bavaria). Several reports noted its recent disappearance: “Were otherwise frequent on the banks of the Danube, but have disappeared completely in recent times” (E_05124, Neustadt a.D., Upper Bavaria); “The last one was shot 20 years ago near Kratzmühle an der Altmühl […]” (E_01554, Eichstätt, Swabia).

We classified reports into categories RARE or VERY RARE, EXTINCT (indications of earlier but no longer occurrence), and NEVER (indications that the beaver never occurred in this area, such as “There are no reports that any were ever here” (E_02483, Selb, Upper Franconia)). Where place-specific details were included—such as in the Munich example above (“an den Isarufern gegen Freysing”)—locations were geocoded and incorporated into GIS mapping. The resulting distribution is shown in Figure 12 for southern Bavaria.

3.5.4. The Western Capercaillie (Tetrao urogallus)

The western capercaillie is highly sensitive to ecosystem change and human disturbances. Like the beaver, it was both a target of (aristocratic) hunting and a species affected by the forestry practices that became increasingly dominant around 1800. As such, the capercaillie is considered as a loser in the face of ecosystem change by habitat degradation.

Nevertheless, Wagner’s survey not only shows signs of a dwindling capercaillie population. In 1845, it still appears to occur in many places, sometimes frequently or very frequently, and in one place there is even talk of an increasing population. One forester noted: “These species have been reproducing for 3 years and only a few cocks are shot during the mating season” (E_02236, Geroldsgrün, Upper Franconia). Other reports explicitly linked local declines to human interference:

Now [it] only occurs singly, but used to be more common. The cause of its disappearance is probably primarily to be sought in the current frequent visits to the forests by wood collectors;

(E_02104, Bayreuth, Upper Franconia)

Breeds in the Annweiler Bürgerwald, but is not infrequently eaten by wood and dead wood gatherers during the breeding season;

(E_02633, Annweiler, Palatinate)

is widespread throughout the district in fairly large numbers, but reproduction suffers due to the many hunting grounds (E_04261, Goßmannsdorf, Lower Franconia).

Table 4. Classification of the frequency of occurrences of selected species: Eurasian beaver, western capercaillie, Eurasian otter, roe deer, wolf.

Number of Reporting Offices	Eurasian Beaver	Western Capercaillie	Eurasian Otter	Roe Deer	Wolf
ABSENT	94	44	2	1	98
EXTINCT	9	1	0	0	11
VERY RARE	12	8	17	1	5
RARE	3	18	42	11	4
COMMON TO RARE	1	23	24	28	1
COMMON	0	13	18	42	0
ABUNDANT	0	6	3	36	0
NOT QUANTIFIABLE a	0	6	13	0	0
TOTAL	119	119	119	119	119

^a An occurrence of this is reported, but the report does not contain (enough) information to quantify its presence.

and Figure 13 provide an overview of the occurrence status of the capercaillie inferred from the textual data with available trend annotations, inferred from ten reports. Only one—Geroldsgrün—projected a positive population trend.

We also integrated the AOD1845 data with recent and historical research on the historical capercaillie population in Bavaria. In particular, a written account by Andreas Johannes Jäckel from 1891 offers additional information on the species’ distribution and decline in the Palatinate [51,52]. This allows a more comprehensive mapping by merging these two sources. Consistent with the 1845 survey, Jäckel notes that the capercaillie was found throughout much of Bavaria, particularly in areas with large, contiguous, and undisturbed forests, as well as in hilly and mountainous regions. He specifically highlights the Upper Bavarian and Swabian mountains, along with the Bavarian Forest.

However, for the Rhine Palatinate, which Jäckel describes as “formerly famous for capercaillie courtship”, he reported a steep decline of population “since the mid-1850s”. Figure 13 maps his account alongside the 1845 data, together providing one of the few existing sources of historical insight into capercaillie habitats in this region.

Today, the capercaillie is considered extinct in the Palatinate [51], and no recent sightings have been recorded [53].

4. Discussion

4.1. Data-Centric Historical Ecology

Historical ecology and environmental history have long relied on narrative records, naturalist notes, and early scientific surveys to study past ecosystems and human–nature interactions [54,55,56,57,58,59]. Within this broader scholarly context, our study aligns with emerging calls for more systematic and data-driven approaches in these fields. In particular, we apply computational approaches to the domain of historical ecology, a methodological direction that is present in ecology [60,61,62] and commonly used in digital and computational history [34,63,64,65], but remains comparatively underexplored in historical ecological research yet [66,67,68,69,70,71].

By combining Natural Language Processing (NLP), similarity analysis, clustering, and spatial mapping, we present a replicable workflow for transforming textual historical data into structured ecological information [72]. While these methods are increasingly used in humanities-based disciplines, their application to historical biodiversity data—especially in relation to pre-industrial European contexts—remains limited. Our approach seeks to bridge that gap.

This data-centric strategy not only enables the extraction of spatial and semantic patterns from archival prose, but also facilitates interoperability with contemporary biodiversity frameworks. Importantly, it does so without replacing the contextual depth of historical research. On the contrary, it highlights the value of integrating computational precision with the interpretive strengths of humanities scholarship—a combination we believe is essential for advancing historical ecology in the digital age.

4.2. Mobilising New Sources for (Historical) Ecology

Our approach highlights the value of administrative sources as empirical foundations for historical ecology and biodiversity studies. The systematic annotation and enrichment of prose-style survey responses enabled us to extract quantifiable information on species occurrences, habitats, and human–wildlife interactions. Although historical texts have been used in ecological research before, they often serve a primarily illustrative purpose. By contrast, our study presents a structured pipeline for converting archival prose into formats suitable for integration with modern ecological data. This process creates new opportunities for longitudinal analysis across historical and contemporary baselines.

The success of such an approach, however, depends critically on the expertise of historians [73,74,75]. Historical research methods are essential for locating and identifying relevant sources, assessing their reliability and context, and avoiding misinterpretations that may arise from anachronistic readings or uncritical data extraction. Historians bring the necessary skills to evaluate provenance, interrogate silences or biases in the record, and interpret language and categories specific to a given period. Their involvement is equally vital in the ’datafication’ of sources—ensuring that annotations preserve meaning and reflect the epistemologies of the time. Without this grounding, efforts to mobilise archival materials for ecological research risk misrepresenting both the historical and environmental realities they seek to illuminate.

We follow earlier calls for closer collaboration between archivists, historians, and ecologists to unlock the full potential of historical biodiversity archives [76,77,78].

4.3. Data Integration

A key contribution of this research lies in making historical biodiversity data interoperable with modern ecological standards and modelling techniques [79]. By enriching 19th-century species reports with metadata such as Red List status, georeferencing, and threat classifications, we demonstrate how archival sources can be brought into productive dialogue with the principles of biodiversity informatics [80,81,82].

Biodiversity informatics aims to structure, standardise, and mobilise biological data across time and space, typically relying on contemporary observations or modern literature [83,84,85]. Integrating historical records into these frameworks presents specific challenges: historical taxonomies may differ from modern ones; geographic references may have changed; and species’ ecological roles may have shifted over time. Successful integration thus requires careful harmonisation between historical and present-day data structures [86,87] and connectivity and interoperability of metadata standards and authority files from various domains, such as VIAF, ABCD, DarwinCore, or ontologies for historical placenames.

In this process, historical expertise is indispensable. Historians ensure that metadata enrichment does not impose present-day concepts anachronistically onto historical records. They provide critical insights into the language, administrative boundaries, and scientific understanding of the 19th century, helping to avoid misinterpretations. Without this grounding, efforts to link archival materials to current biodiversity data infrastructures risk distorting the ecological and cultural realities of the past.

While we do not claim to have fully resolved the complexities of harmonising historical and modern ecological data, the study offers a practical illustration of how such integration can be approached. Our work highlights that meaningful interoperability is achievable when computational methods are guided by historically informed annotation and interpretation. Future research will benefit from building interdisciplinary standards for integrating archival biodiversity data into global information systems. One of these paths may lead to a biodiversity knowledge graph [88] with improved data integration and inference [89].

4.4. History of Science and Knowledge Practices

The AOD1845 survey not only offers ecological data but also provides valuable insights into the scientific culture and knowledge practices of mid-19th-century Bavaria. Situated at the intersection of local expertise and emerging scientific inquiry, the survey reflects how knowledge about nature was produced, structured, and communicated within administrative and proto-scientific frameworks [68,90].

The compilation of observations by foresters illustrates an early form of systematic data “bottom-up“ collection that anticipates features of today’s citizen science initiatives [91,92,93,94]. Typical for this time, the survey embodies a transitional phase between traditional, experience-based ecological knowledge and the increasingly formalised, standardised methods of scientific data gathering that would characterise later ecological surveys.

Moreover, the responses reveal how ecological knowledge was shaped by human priorities, perceptions, and land management interests. Animals were described not only in terms of presence or absence but also in relation to hunting value, agricultural conflicts, and local climatic conditions. These factors influenced how species were observed, recorded, and understood. The dataset therefore encapsulates both environmental realities and contemporary ways of knowing nature—a reminder that historical biodiversity records are embedded within specific cultural, economic, and epistemological contexts.

Recognising these embedded knowledge practices is crucial for historical ecology. Without critically engaging with the assumptions, omissions, and classifications present in historical sources, there is a risk of misreading past ecological conditions. Here again, the historical method proves essential: it enables researchers to interrogate the production of knowledge itself, rather than treating historical data as a transparent reflection of natural phenomena.

Thus, historical ecology benefits not only from the empirical information contained in archival sources, but also from an understanding of how and why that information was gathered in particular ways. The AOD1845 survey stands as both a rich ecological archive and a testament to the historical processes that have shaped biodiversity knowledge over time.

4.5. Enriching Biodiversity Knowledge

Establishing reliable historical baselines is crucial for understanding long-term biodiversity change and for informing contemporary conservation strategies [4,76,95,96,97,98,99,100,101,102,103]. The AOD1845 dataset offers a valuable pre-industrial reference point for species distributions across Central Europe, capturing conditions before large-scale industrialisation and modern agricultural intensification reshaped the landscape.

Our case studies—focusing on species such as the Eurasian otter, beaver, wolf, and capercaillie—illustrate the potential of historical data to reveal shifts in species distributions and population dynamics. Although necessarily selective, these examples demonstrate how archival sources can identify past habitats, document inferred trends such as local declines or extirpations, and trace human impacts on ecosystems over time.

Importantly, these historical insights are not merely of antiquarian interest. They can directly inform current conservation efforts, such as species reintroductions, habitat restoration, and the setting of realistic ecological targets. Understanding the historical range and abundance of species helps distinguish natural absences from anthropogenic losses and can guide more ecologically sound restoration practices.

Nonetheless, interpreting historical biodiversity records requires caution. Differences in observational practices, reporting biases, and shifting taxonomic definitions must all be critically evaluated. Here, interdisciplinary collaboration between historians, ecologists, and conservation practitioners proves essential to ensure that historical data are both accurately understood and appropriately applied to modern contexts [104,105].

In this way, enriched historical datasets like AOD1845 not only expand our temporal perspective on biodiversity change but also contribute substantively to present-day debates in conservation science and environmental policy.

4.6. Human–Nature Perception and Anthropocenic Studies

Beyond providing ecological data, the AOD1845 survey offers important insights into how 19th-century observers perceived and interpreted nature. The foresters’ reports contain reflections on factors influencing species populations, including land use changes, habitat loss, climatic variability, and hunting pressure. These observations illustrate that ecological awareness—albeit framed through the lens of contemporary priorities and knowledge—was already present among local officials.

Such historical perceptions are valuable for understanding the cultural dimensions of human–nature relationships. They reveal how natural environments were viewed through the practical concerns of forest management, agricultural productivity, and game regulation, rather than conservation for its own sake. Nevertheless, the documentation of habitat decline and species rarity shows an emerging sensitivity to environmental change that prefigures conservationist thinking (see e.g., [106]).

Reading these archival records alongside modern environmental discourses highlights both continuities and shifts in how humans conceptualise their relationship with the non-human world. Today’s Anthropocene debates, which frame humanity as a dominant geological force, often call for a deeper historical understanding of the gradual processes that have led to current ecological crises [107]. Historical ecology, informed by archival sources like AOD1845, can contribute to this broader perspective by tracing how human actions and perceptions have shaped ecosystems over centuries.

Moreover, acknowledging the historical entanglement of social, economic, and environmental factors helps to challenge simplistic narratives of ecological decline or recovery. It encourages a more nuanced view that situates biodiversity change within broader patterns of land use, governance, and cultural values over time.

Thus, the AOD1845 dataset not only documents species occurrences but also provides a window into historical human–nature interactions, offering a valuable resource for Anthropocenic studies that seek to integrate ecological and cultural histories [106,108,109,110,111].

4.7. Towards Computational Historical Ecology

Although our annotations were conducted manually, they offer a ground truth for future machine learning models and (semi-)automated analyses. Scaling this approach will require further methodological development, however, including domain-specific models for historical language and ecology. In particular, advances in Natural Language Processing, Machine Learning, and Geospatial Analysis would support the extraction, classification, and integration of ecological information from a wide variety of archival materials once they are discovered and digitised (see [112] for a first example using the AOD1845 dataset). This may lay ground for historical applications in various fields of ecology and Earth system studies such as land use, climate, and biodiversity. We would like to frame this cross-disciplinary research programme that combines historical, ecological, and computational thinking and practices in a data-centric approach as Computational Historical Ecology.

However, the potential of computational methods can only be fully realised if they are understood as a joint and integrative effort. Historical language variation, regional terminologies, and the epistemological frameworks of earlier periods require careful interpretation. Computational models must be trained not merely on linguistic features but on understandings, grounded on both history and natural sciences. Thus, interdisciplinary collaboration remains central to scaling Computational Historical Ecology while preserving interpretative rigour. As such, it offers an opportunity to bridge humanities and environmental sciences by a shared understanding of data and their processing, bringing new temporal depth and cultural context to biodiversity research at a time when historical perspectives are urgently needed [57,66,113,114].

5. Conclusions

This study introduces a computationally informed approach to historical ecology, demonstrating how archival sources can be transformed into structured ecological data. By applying text mining, spatial analysis, and structured annotation to a 19th-century Bavarian survey, we offer a new pathway for investigating past biodiversity patterns.

Albeit a single case-study, our findings show that digitised administrative records, when carefully prepared and enriched, can contribute to current biodiversity knowledge and conservation efforts. While limitations remain, especially regarding the completeness and standardisation of historical data, we see promise in this direction. In particular, historical baselines derived from sources like AOD1845 may support reintroduction planning, habitat assessment, and long-term biodiversity monitoring.

We conclude by encouraging greater investment in digitising ecological archives, developing computational tools suited to historical language, and fostering collaboration between disciplines. Computational Historical Ecology, as proposed here, has the potential to bridge humanities and natural sciences and to offer a fuller understanding of biodiversity change over time.