Introduction

Tropical forests play a key role in human cultures, socioeconomic systems, biodiversity, and the operation of the Earth system1,2, with tropical deforestation being considered a key part of the ‘Great Acceleration’ of human impacts on the planet3. Humid montane forests, found at altitudes between 500 and 3500 m, and particularly between 1200 and 2500 m4,5,6, are particularly key in this regard given their unique endemic biodiversity7,8,9, status as carbon sinks10, sensitivity to climate change11, involvement in climate regulation5,12, and role in water conservation. For example, montane tropical forests capture rain and stabilize stream and groundwater flows, critical for supporting downstream ecosystems and communities6,13,14, and host up to 80% of floral plant species in some tropical regions9,15. In many regions in South America and Asia, the 20th and 21st centuries have witnessed large areas of these forests be converted to large-scale agricultural systems13. Furthermore, land degradation, through livestock overgrazing and frequent fires, has led to high losses of biodiversity, biomass, and forest structure16,17,18,19,20, undermining the ecological integrity of these ecosystems. Although there is a growing global concern for monitoring montane tropical forest change, eastern Africa has been relatively neglected in this regard, with few systematic, long-term studies21,22,23.

In the 21st century, tropical montane forest loss is frequently discussed as primarily resulting from post-colonial land use changes linked to encroachment by local small-holders and industrial activities24. In eastern Africa, tropical forest loss has been specifically associated with extraction of wood for fuel and construction, sale of forest products for economic value, and, particularly, through expansion of land-holdings by impoverished communities25. However, such perceptions ignore the centuries of European colonialism that left legacies for environmental challenges and socioeconomic inequalities today. For instance, colonial land uses involved large-scale conversion to plantation forests, livestock ranching land, infrastructural development, and economic extraction for timber and fuelwood26,27. Such activities were often promoted by colonial agents who saw land conversion as a form of economic ‘development’ and productivity28,29,30. Simultaneously, traditional approaches to forest management and land use were often suppressed, while populations were relocated away from more productive areas that were coveted by colonial powers31,32. Even early instances of tropical forest conservation involved a colonial ‘locking out’ of native communities, with natural ‘value’ being placed above millenia of cultural use and care33,34,35.

Within the eastern African region, high trends of forest loss since the 1950s have been estimated based on historical data and more recent Landsat images. For instance, forest estimates in Uganda indicate that forest area reduced from 20,901 km2 in the 1900 to about 11,176 km2 by the end of 196036, representing a 46% forest area cover loss as a result of deforestation through clear cutting of forest areas followed by establishment of subsistence farming37. In Tanzania, through historical map assessment, about 80% of the forest area within the Eastern Arc Mountains is estimated to have been lost before 1955 and another 25% of forest cover lost since 195522. Furthermore, using historical map assessments it is estimated that Madagascar lost about 44% of its natural forest cover between 1953 and 201438. While much attention is focused on recent time periods in these studies, many of these contemporary patterns of montane forest loss and increased fire frequencies, linked to the establishment of large-scale forest plantations, introduction of exotic species such as Eucalyptus sp., railway fires and prohibition of indigenous communities from undertaking traditional fire controls, as well as their associated hydrological feedbacks, likely have roots in colonial alteration and forest policy. Indeed, even in the case of recent small-holder extraction, the location of populations and available land for food production has been shaped by colonial population relocation and impoverishment33.

In Kenya, for example, colonial policies relating to land ownership and resource use altered the traditional norms of native communities and their relationship with the forest ecosystems they inhabited, as they restricted use and ownership of forest resources32. However, existing comprehensive forest assessments for Kenya have been almost entirely focused on the past two to three decades, the period over which Landsat imagery has been available24. Some studies have discussed the role of colonial impacts on Kenyan forestry28,39,40,41,42 and historical perspectives on forest changes and colonial legacies have been seen as increasingly important21,23. However, no study has attempted to undertake a spatial and long-term temporal quantitative assessment of primary montane forests in eastern Africa, and changes in neighbouring land uses and population distributions, at the start of colonial period, during and post-independence.

The latitudinal tropics have overall seen a growing incorporation of archeology, history, and paleoecology to provide quantitative insights into past human-forest relationships26,43,44,45. This archival evidence can provide long-term, spatial records of forest cover and land use for time periods prior to those where satellite imagery is available46,47. Several existing studies in eastern Africa have used historical data and a combination of historical and paleoecological data to estimate changing forest cover extents21,22,23. In addition, in Madagascar combined studies of global forest cover data and historical forest maps have been implemented to assess past forest cover loss38. However, there is still need to improve the accuracy and reliability of historical estimates derived from historical data, which can be done by integrating higher resolution historical maps and relevant historical literature. Here, we focus on georeferencing, digitizing and assessing historical maps from 1910, historical literature from the late 1890s and undertaking land use analysis using Landsat images acquired between 1970s and 2024 to establish land use and forest cover changes in the montane forests of Central Kenya since 1910, as well as their associated socioeconomic contexts. We analyse post-colonial resettlement datasets to assess the impacts of small-scale farmers and large-scale plantations on the primary montane forest within the central montane regions of Kenya. Our goal is to, firstly, establish the possible spatial extent and distribution of montane forest vegetation at the onset of colonialism and subsequent increased human impacts on these ecosystems. Secondly, we seek to establish the extent of land use changes as a result of colonial land allocations, particularly focusing on large-scale cultivation and also the resulting impact on land use dynamics as a result of post-colonial resettlement of small-scale farmers within montane forest zones. Finally, we explore the extent of infrastructural development within the colonial and native areas, with focus on railway and road network expansion during the colonial periods. While our datasets are restricted to Kenya’s montane regions, the methodology and historical perspective applied is of relevance to studies of long-term human interactions with montane tropical forests, and tropical forests more generally, on a global scale.

Our study covers the montane region of central and western Kenya where primary montane forests currently occur at an elevation of between 1800 and 4000 m asl., giving way to an alpine montane zone above 4000 m asl48,49,50 (Fig. 1). These montane forests are located within the five main water catchment areas of Kenya, which include the Mt Kenya ecosystem which currently covers an area of about 284, 590 ha and located at 00°10’ S, and 037°20’ E, The Aberdares (Mt Nyandarua), with an estimated area of 216,200 ha located at 00°39’ S, 036° 42’E, The Mau forest complex with an estimated area of 412,000 ha, located at 00°33’ S, 035°38’ E, Mt Elgon with an estimated area of 107,820 ha, located at 01° 07′ N and 034° 36′ E, and the Cherangany forest zones with estimated area of 110,181 ha, located at 01° 16’ N 035° 26’ E. Kakamega-Nandi forest with an estimated area of 47,000 ha is a mid-altitude tropical rainforest and an eastern extension of the Congo Basin forests, located at N00 ° 17’ E34 ° 53’. (Fig. 1). The vegetation biomes of these zones are defined by the prevailing climatic and altitudinal conditions49,50. Generally, at lower altitudes between 1800 and 2500 m asl, the primary forest is composed of high canopy mixed montane forest dominated by Podocarpus sp. and Ocotea sp. This forest zone is followed by bamboo (Arundinaria sp) and low canopy montane forest (Hagenia sp, Usnea sp) communities between 2500 and 3000 m asl. The zone between 3000 and 3500 m asl is dominated by the sub-alpine vegetation of heath and moorland mainly covered by ericaceous vegetation types. Beyond 3500 m asl, the dominant biome is alpine vegetation with distinct vegetation types of Senecio sp and Lobelia sp49,50.

Fig. 1: The location of the central highlands of Kenya.
figure 1

a Tropical equatorial location of Kenya. b Central highland region of Kenya (gray) and the major cities (blue squares). c The elevation profile and the distribution of the main montane forests within the Central highland region of Kenya above ca. 1800 m asl. 1. Mt Kenya, 2. The Aberdares, 3. The Mau complex, 4. Cherangany, 5. Mt Elgon, 6. Kakamega-Nandi, based on 1925 map (green). Data Source for a: ESRI World Imagery: Esri, DigitalGlobe, Earthstar Geographics, CNES/Airbus DS, GeoEye, USDA FSA, USGS, Getmapping, Aerogrid, IGN, IGP, and the GIS User Community. b Esri, Maxar, Airbus DS, USGS, NGA, NASA, CGIAR, N Robinson, NCEAS, NLS, OS, NMA, Geodatastyrelsen, Rijkswaterstaat, GSA, Geoland, FEMA, Intermap, and the GIS user community. and c NASA STRM Earthdata (https://search.earthdata.nasa.gov/).

Full size image

The climate of these montane ecosystems is generally cool and moist to moderately dry at lower altitudes and on the lee ward side of the mountains. There is a bimodal rainfall pattern, with long periods of rainfall between April and May and a short rainfall period between October and December. Rainfall ranges from 2500 mm in the wetter regions to 800 mm per year on the drier sides of the mountain ranges (KFS & KWS, 2009). These mountain regions have often been historically referred to as the White Highlands and were the preferred areas for the settlement of European settlers during the colonial period 1895–196351,52. The region has a very favorable climate, with cool to warm temperatures ranging between 0 and −18° C on the high montane areas to 18–30 °C in the lower elevation areas53. Currently, the montane forest areas are managed by the Kenya forest service (KFS), Kenya wildlife services (KWS), and the county governments within whose jurisdiction they fall. Due to their importance as biodiversity hotspots and as habitats to endemic plants and animal species, the montane areas have been designated as Key Biodiversity areas (KBAs)(Mau forest complex), Important Bird areas (IBAs) (The Aberdares and Mau forest complex), Biosphere reserves and UNESCO World Heritage sites (Mt Kenya)54.

Results

Colonial and native community land areas

At the onset of colonialism, the preference for settlement locations by the colonial white settlers was greatly determined by the availability of a conducive climate that offered a comfortable living environment, supported large-scale plantation farming, and provided forest resources for building and trade. The colonial settlement area covered an area of about 2,900,000 ha, and about 1,700,000 ha of forest was under colonial control by the end of the historical mapping exercises of the colonial government in the mid-1920s. The region set aside for occupation by the ‘native African communities’ within the Kenyan central highlands area was represented by a total of about 3,900,000 ha, and about 280,000 ha of forest area, with the high-density areas being located within the central region (Mt Kenya and Abedares) of the Kikuyu communities and the western region of the Abaluhya, Kalenjin, and Luo communities (Mt Elgon and Kakamega) (Fig. 2).

Fig. 2: Land area allocation and human impact within the central highland region of Kenya around 1925.
figure 2

The zoning of land areas indicating the boundaries of native reserves, inhabited by native communities (brown), the colonial reserve areas for colonial settlers (blue) and forest reserve areas referred to as crownland/ crown forests (dark green). The human impact is shown by the distribution of native villages (black dots) (digitized from the 1910–1916 maps), motorable road network (red lines), railroads (purple), sawmills (green dots) and areas of forest land excisions (yellow).

Full size image

Infrastructural development

Our analysis shows that extensive infrastructure establishment had occurred by the mid-1920s, with most of this development occurring in colonial settlement areas. The East African railway, that was under construction since 1895, was now well-connected, making accessibility to most farms and industries within the colonial settlement areas easier. In addition, most of the farms within colonial settlements and montane forest areas were connected to motorable and graded roads (Fig. 2). These provided easy access to makeshift sawmills and saw yards, and supported large-scale timber exploitation and delivery to train stations for export. The opposite is however true for regions within the native settlement areas, that seem to have been marginalized from infrastructural development and had no extensive connections to the main roads or railway network.

Primary forest extent during the colonial period (1900–1975)

Our results indicate that the extent of primary montane forest within the Kenyan Central Highland region, including moorland and high alpine vegetation, at 1910 covered an area of about 1,980,000 ha, with most of this occurring at an elevation above 1800 m asl (Fig. 3, and Supplementary Fig. 5). By 1925, the area of montane forest had already been reduced to an estimate of about 1,770,000 ha, representing a 10% loss. By the end of the colonial period in 1963, the total extent of the primary montane forest covered an estimated area of about 1,220,000 ha (Fig. 3 and Supplementary Figs. 5 and 6).

Fig. 3: Land cover/use change maps of the central montane region of Kenya.
figure 3

Land cover/use changes in primary forest cover(green), large-scale plantations (pink) and small-scale mixed farms (light green) in 1925, 1975, 2000, and 2024. Insert bar graphs shows the total areal landcover area estimates in hectares at RMSE of 0.4–0.06 km for the historical map estimates and a 95% Confidence Interval for the 1975–2024 estimates of primary montane forest, large-scale plantations, and small-scale mixed farms between 1910 and 2024.

Full size image

Post-colonial period 1975–2024

The Landsat classification assessment generated acceptable accuracies with an overall Kappa hat accuracy of 0.91, 0.86, and 0.83 for the 1975, 2000, and 2024 landcover classification assessments, respectively at a 95% confidence interval. There was high agreement between the producer and user accuracies in the classification of both primary forest and small-scale farms, while moderate agreement was observed for large-scale plantations (Supplementary Table 7).

From the classification assessment we observe that by 1975, approximately 700,000 ha of primary forest had been additionally lost over a 50-year period since 1925, when the forest reserve boundaries had been established and now covered an area of about 1,077,000 ha, representing a 40 % total loss from 1925. In addition, land under large-scale plantation agriculture had increased to about 605,000 ha, while small-scale farms occupied an approximate area of 95,000 ha. Between 1975 and 2000, the loss of primary montane forest slowed, with a loss of approximately 44,000 ha representing a 4% loss. On the other hand, land under small-scale farming extensively increased to over 600,000 ha while largescale plantations reduced to 340,000 ha due to extensive subdivision. By 2024, the extent of primary montane forest cover had reduced to an area of about 845,000 ha, representing a further 21% loss. Small-scale plantations further increased by about 160,000 ha and large-scale plantations also increased by about 20,000 ha (Fig. 3 and Supplementary Figs. 5 and 6).

Resettlement of native Africans to re-settlement areas

Land redistribution of the large-scale colonial settler farms to small holder communities began after independence between 1962 and 1964 (Fig. 4). The available data indicate that by 1975 approximately 338,000 ha had been reallocated to native farmers (Figs. 5 and 6A). This reallocation was part of the initial resettlement program that targeted the reallocation of approximately 500,000 ha as part of the one million Acre settlement scheme52,55,56. This involved the purchase of previously colonial owned land by native farmers. We assessed 317 settlement schemes established between 1960s and 2012, of which 21 schemes lack data on the size of area allocated and number of settlers. We observe that by 2012, 693,850 ha had been allocated and a total of 124,175 households resettled. The area of land allocation per household had greatly reduced from an average of 10.5 ha to about 2.4 ha per household (Fig. 6b). The resettlement program eased up from the mid-1970s, only to again increase between 1978 and 1984, where a further 134,000 ha were redistributed to native farmers. Between 1990 and 2000, the resettlement into new areas slowed down and then we observe a steady increase of new resettlement from the early 2000s to 2012, where 135,000 ha of new areas were allocated to native small holder farmers with land sizes measuring an average of 2.3 ha per household (Fig. 6b and Supplementary Fig. 7).

Fig. 4: Distribution of the initial resettlement schemes within the central highlands of Kenya.
figure 4

Resettlement areas of the native small-scale African farmers within the central montane areas of Kenya during the initial reallocation program in the 1960s (purple), in relation to the colonial reserves (blue), along the margins of the montane forest reserves in 1975 (green), and the native reserves (brown). Regions in yellow represent areas of montane forest loss between 1910 and 1975.

Full size image
Fig. 5: The total number and land sizes of resettlement schemes.
figure 5

The size of land area in hectares of each of the documented resettlement schemes between 1962 and 2012, with each individual black dot representing a single resettlement scheme established at a given time. Resettlement data retrieved from the Ministry of Lands and planning (MoLPP), a dataset presented by Lukalo and Odari102.

Full size image
Fig. 6: Area of Land allocated during resettlement programmes in the postcolonial period.
figure 6

a Total estimates of land area (left scale) reallocated to native small-scale farmers within resettlement schemes in the Central, Rift-valley and Western region of the colonial reserves’ areas between 1962 and 2012. Red dashed line (right scale) represents the total land area allocated within all regions between 1962 and 2012. b Estimated average area of land allocated per household in the Central, Rift-valley and Western region of the colonial reserves’ areas between 1962 and 2012. Blue line and squares represent the average size of land allocated to native small-scale farmers within colonial reserves between 1962 and 2012. Resettlement data retrieved from the Ministry of Lands and planning (MoLPP), a dataset presented by Lukalo and Odari102.

Full size image

In contrast to the resettlement program during the 1960s to 1970s, that focused resettlement of native farmers within previously occupied lands, the period between the late 1990s to 2000s saw the excision of new primary forest areas in the Mau forest complex and Mt Elgon forest areas, where over 60,000 ha and 20,000 ha were allocated for resettlement, respectively (Fig. 6a, supplementary Fig. 7).

Discussion

Our results indicate an estimated 57 % loss of primary montane forest in the Central highlands of Kenya between 1910 and 2024 and are supported by additional, independent assessments by colonial agents. For example, the first montane forest cover estimate done by the forest department under the colonial government in 1920 estimated the primary montane forest area to be at about 5100 miles2 (1,320,895 ha) which did not include primary forest areas within native reserves (Mt Elgon and Cherangany forest areas ca 320,000 ha) at the time (Troup, 1922). The 1963 forest report indicates that the estimated primary forest area gazetted at that time was approximately 1,069,020 ha (34 % loss between 1920 and 1963)57, while 1,003,528 ha (39% loss between 1920 and 1974) forest cover is estimated for 197458. These estimates corroborate our long-term primary forest estimates in this study. However, despite their great potential in extending the temporal understanding of forest cover dynamics, we acknowledge that the use of historical maps introduces inherent uncertainties resulting from their production, through the georeferencing and digitization process and also in our interpretation. For instance, in our assessment, we note that some sections of the large-scale plantation area in the 1925 map appear to be primary montane forest in the subsequent maps (1975, 2000, 2024) (Fig. 3). Our interpretation is that, during the 1925 large-scale forest excision, sections of these forest areas were not exploited through timber harvesting due to the difficult terrain, accessibility and also lack of merchantable trees for timber extraction. This kind of information was impossible to ascertain using the historical cadastral maps, and, in our case, the assumption is now that in fact the forest area extent in 1925 was under-estimated and the area under actual large-scale plantation over-estimated. Despite these shortcomings, our assessment currently provides the best available long-term quantitative estimates of forest cover over the past 100 years for the montane forests of Kenya.

Our data contributes to a slowly growing number of long-term studies of tropical forest cover on the African continent. Using a combination of paleoecological-data and historical maps to reconstruct the extent of African forest at the start of colonialism, it is estimated that 74–100% of tropical forest in East Africa has been lost since the start of the 20th century21. Similarly, using historical land use maps to investigate the extent of land use change in Tanzania’s Eastern Arc Mountains, research indicates that about 74% of forest was lost between 1908 and 200023. Meanwhile, many short-term studies highlight extensive forest loss in tropical Africa over the past 50 years22,36,37,59,60,61,62,63,64,65. This forest degradation and loss continues to occur, despite these forests being under government and state protection. Although it is estimated that about 75–80% of African forests are under state protection62, this has not seemed to stop accelerations in forest loss. For instance, the rate of total forest loss in Africa between 2010 and 2020 was estimated to be about 3.9 million ha per year62. These recent and historical losses in forest are often looked at purely in the context of changes since the mid-20th century, linked to increasing expansion of industrial activities, urbanism, and demand from growing populations25,66,67. In Africa, impoverished smallholders have been particularly focused on as major drivers of these changes25,68. For instance, it has been argued that the concentration of the local population by the colonisers along roads and bare areas might have contributed to the lower rate of forest cover loss in the Democratic Republic of Congo21. Our study, however, highlights that such a focus ignores the centuries of native land management and forest interactions, and subsequent impacts of colonialism on land use and native demography, that have shaped the presentation of sustainability and economic issues in the 21st century21,23,33,41,69.

The use, management, conservation and status of Kenya’s montane forest ecosystems changed drastically from the early periods of European arrival in Kenya in the late 1800s to the early 1900s. Before the arrival of the colonial regime, scattered native communities of the Kikuyu, Dorobo and highland Kalenjins are said to have inhabited the lower sub montane forest zones70,71. Here they practiced subsistence farming using basic tilling tools and growing tubers and annual crops72. Fallow and shifting agriculture were utilized and native communities cleared forest using small hatchets and fire to open up forest areas to new farming lands, where they would plant yams, sweet potatoes, different types of millets (bulrush millet, foxtail millet), sorghum, beans and edible arrow-roots70,73. The resilience of these strategies is highlighted by the fact that the regions with dense native populations in the early 1900s, the central region (Aberdares-Mt Kenya region) and Western regions (Kakamega and Mt Elgon), show lower rates of forest loss in our dataset compared to the Rift-valley region that had a far lower native population at the start of colonialism. Indeed, where native Africans within native reserve areas had a longer history and ownership rights they established internal conservation and management systems that prohibited excessive forest destructions39,74. For instance, the native Kikuyu community relied heavily on dense forest cover for physical protection from tribal invasions70,75. In addition, the community used forests as areas to lay deceased family members to rest, with forest maintained as a key part of sacred landscape70,74,75.

In contrast, the colonial settlers viewed forests through a primarily exploitative lens76. Primary forests were seen as rich sources of raw timber, with particular focus on the extraction of high density wood for export and cheap fuelwood for the railway industry29,77. Primary montane forests were, however, considered to provide low returns for logging, due to difficulties in terms of milling and access39. This led to pushes to convert forest land to fast-growing plantation forest (often utilizing exotic species) for maximum profits28,77. It is reported that by the early 1920s about 13,000 ha of plantation forests had already been established through the replacement of slow growing primary forest with fast growing exotic Eucalyptus sp., Acacia mearnsii (Black wattle) and Cupressus Sp28. Between 1920 and 1930, the biggest wattle production company had been established in the region, and the area under black wattle plantations used for tannin production had expanded to over 12,000 ha by 193278,79. Moreover, the location of primary montane forest on fertile land with accessible freshwater also made them attractive for large-scale farming of high value crops. The initial forest area loss was a result of the extensive allocation of land within the highland areas to colonial settlers for logging, provision of fuelwood to run train locomotives, and to provide suitable areas for the expansion of agriculture and infrastructure28,41,80. Large areas of primary montane forest were awarded as forest concessions to individuals and colonial settler companies for logging77. Our results indicate that about 210,000 ha of primary forest had been allocated for logging and large-scale farming by 1925 (Fig. 2). As noted in the 1925 colonial annual report, 158,000 ha were already under cultivation at the time. The introduction of coffee and tea plantations in 1904 and 1924, respectively, that required cooler, high-altitude climates, stimulated further conversion of primary forest. By 1936, 42,000 ha were under coffee plantation and, by the mid-1960s, about 105,000 ha of land had been set aside for tea and coffee production.

Much of the primary forest deforestation linked to tea, coffee and exotic tree plantations occurred in Highland Reserves set aside by colonial agents on their arrival. British land ordinance regulations in 1897 specified that European settlers would acquire native land not under cultivation or frequent use by the native Africans. However, due to the increasing number of white settlers, the British government enacted the East Africa Order in the Council of 1902 and later the Crown Land Ordinance 1902 which bestowed the commissioner to the protectorate with the authority to alienate and lease out all ‘native’ lands81. In 1915 the Crown Land Ordinance nullified all legal right to land ownership for the native Africans. (Fig. 7) Ironically, colonial agents claimed extensive destruction of forest ecosystems, especially in the central highland regions77,82,83,84,85, as justification for setting aside the primary forest reserves for use and exploitation by the colonial settlers. However, these forest reserves were actually land set aside for later exploitation40,86. Indeed, in our assessment we observed that by the mid-1920s over 200,000 ha had already been allocated for agricultural and timber logging purposes28,84. These land policies also ruptured the use of forest resources by the native communities32. The native inhabitants were prohibited from forest resource extraction, with hefty fines levied on anyone found in contravention or even six month jail terms28,77,84. This, in turn, critically diminished the urge to conserve the primary montane forests by native Africans. It is well noted that, in the early 1920s, one of the main challenges facing forest restoration efforts in colonial settlement and forest areas was the lack of cooperation from the local communities who found no merit in undertaking replanting exercises in areas on which they had no ownership claims28.

Fig. 7: Timeline of important periods of policy changes.
figure 7

Summary of the key colonial and post-colonial land and forest policy changes affecting land owning rights and management of forests within central highlands of Kenya. Information summarized from references81,84,91,92,94.

Full size image

These colonial processes left legacies for the current crisis facing Kenya’s montane forests. After Kenya’s independence in 1963, native communities were resettled into smallholder farms within land areas previously set aside as colonial reserves. This was initially done through government schemes that, through world bank loans, bought land from the colonial white community and reallocated the subdivided land parcels to native small holder farmers to boost their agricultural production55,56,87. The initial phase of the scheme reallocated about one million acres to native smallholder farmers between 1962 and 1975. From the 1975 land use map we observe that the majority of the primary forest loss since 1925 had been due to expansion of large-scale plantations to an area of approximately 800,000 ha, with small-scale farms expanding to about 95,000 ha. By the late 1970s, however, about 450,000 ha of land previously owned by the colonial settlers had been reallocated to native small-scale farmers, with the average size of land allocated per household ranging between 10 ha and 4 ha in the 1960s and 1970s, respectively. Notably, native, small-scale farmers were now allowed to plant and own coffee and tea farms, and this provided an opportunity to expand and establish small holder tea and coffee farms within the mountain areas. Land area under tea cultivation increased from 25,000 ha to 118,000 ha and coffee increased from 45,000 ha to over 170,000 ha between the 1960s and 1980s88.

In the period between 1975 and 2000 we observe that the land under small-scale farming increased to over 600,000 ha, while land areas under large plantations greatly reduced. This is attributed to extensive resettlement of small-scale farmers in colonial reserves previously owned by colonial settlers. This new wave of resettlement contributed to accelerated loss of primary forest area between 2000 and 2024. Firstly, larger populations needed to be resettled, with land sizes allocated per house hold simultaneously falling from about 10 ha to 2.4 ha per household within the settlement schemes. This caused more forest encroachment by the native settlers in search of fuelwood and forest resources. Secondly, in contrast to the resettlement undertaken in the 1960s, the resettlement of native communities in the early 1990s and 2000s was used to garner political support87,89,90. Many of these resettlement schemes lacked proper documentation to guarantee ownership87, with the majority of the unregistered resettlement schemes represented by allocations made between 1990s and 2000s41,87,89. It is also important to note that the majority of the native inhabitants were resettled into these forest reserves from other regions under political protection and many had no historical practices of traditional forest management41,56,91. Moreover, thanks to colonial policies, the majority of the native population had been disengaged from forest management for over five decades.

The role played by native small-scale farmers in regenerating primary montane forest will be determined by the efforts made by government institutions to genuinely incorporate native stakeholders and traditions in terms of use and benefit sharing. Past efforts have been attempted through programs such as the Shamba system introduced in the early 1900s by the colonial government91,92 (Fig. 7). This program focused on forest regeneration and offered farming land to native farmers for a designated period of time. In exchange the farmers were expected to plant and tend to tree saplings for a period of 3–5 years, after which they would vacate the forest land. The continued vilification of small-scale farmers has fostered distrust and non-cooperation with forest authorities who, historically, and even in recent times, have been associated with forced forest evictions. These challenges reflect deeply rooted historical patterns where state institutions, emerging from colonial systems, have set a precedence for forest degradation while sidelining traditional forest management practices93. In Kenya, the enactment of a new forest act in 2005, followed by its revision in 2020, introduced provisions and guidelines mandating the inclusion of local communities in forest management through the Community Forest Associations (CFAs)94 (Fig. 7). This initiative aligns with the increasing global need for the decentralization of forestry strategies in the past three decades95,96. Looking into the future, the challenge of primary forest loss in tropical Africa will persist as demand for arable land increases and the impact of a changing climate continues to disrupt agricultural production. In order to improve the inclusion of local communities in regional and global forest regeneration efforts, we argue that a greater focus must be placed on ensuring genuine local stewardship and incorporating traditional practices in use and management of forest resources. In addition, empowering native communities as key stakeholders in conservation and forest management will be crucial to safeguarding the remaining ecosystems of primary forest areas.

In addition, there is a big challenge in understanding the long-term drivers and impacts of human interventions within montane areas due to the paucity of long-term quantitative data, especially in the less studied and more inaccessible regions within tropical Africa. As demonstrated by the few long-term studies within eastern Africa to date21,22,23,38, including this study, there is a great potential to improve our understanding of long-term trends in tropical forest cover by incorporating historical maps with more recent datasets and methods. The success of future strategies to slow down and even stop accelerated forest loss will heavily depend on how best we understand the deeper, long-term human and climate driven impacts on tropical forests, which can only be achieved by employing multidisciplinary methodologies and studies, such as those from paleoecology, history, and archeology, analysis which can help extend the temporal period of forest cover shifts and composition within different ecoregions in tropical Africa.

Methods

Data

Notes and descriptions

There are currently no estimates of forest land cover distribution within the central montane region at the start of colonial rule in Kenya in 1900. It is also important to acknowledge that during this period there were no detailed or accurate maps for the region. Therefore, in order to establish baseline estimates of montane forest distribution and extent at the onset of the colonial period, we can only extrapolate from information contained in personal expedition journals and reports that describe the status and extent of forest cover at the time they were recorded. In addition, we can leverage other forms of data that describe the species of fauna and individual plant species and also photographs taken from various locations. For this we used travel documentation within the British East Africa colony that we acquired from the online archives of the Biodiversity heritage library collection and from the Smithsonian Institution Archives under the Smithsonian Field book project (www.biodiversitylibrary.org).

Some of the main datasets relating to past primary forest extent during the colonial period come in the form of the use of elevation points along trekking pathways to define regions of dense forest cover occurrence and commencement, description of various species types, and instances of forest destruction by native communities. Here we explored notes, photographs, and sketch maps from explorers who journaled their discoveries within the montane regions of central Kenya between 1887 and 1910. Thomson (1887) documented his journey through montane areas including Mt Kenya, Aberdares, Mau forest and Mt Elgon and among the tribes of Eastern equatorial Africa72. Gregory (1894) provides a description of physical geography and the sources of the main river systems of the Aberdares and Mt Kenya forest and provides information on the distribution of the vegetation zones and dominant tree species97. Mackinder, (1900), Crawshay, (1902) and Dundas (1908) extensively described the environment of the main native population that inhabited the montane areas at the onset of colonialism70,71,73. Finally, Edmund Heller’s writings between 1875 and 1939 provide an extensive collection of archival notes and photographs taken during his journey through East Africa in the early 1900s describing the traditions and ecosystems he encountered98,99.

Colonial land use allocation and Montane forest extent between 1910 and 1970

We retrieved available historical maps and data reports spanning between 1900 and 1970 from the British Library online map catalog, sourced under the War Office and accessed through the wikimaps Waper Archives-British East Africa (https://warper.wmflabs.org/maps) and also from the League of Nations Library map collection created by the War office, from the United Nations library and archives, Geneva (https://archives.ungeneva.org/east-africa-protectorate). This included topographical maps, cadastral maps, survey maps and sketches, choropleth vegetation, and population maps (Supplementary Tables 1 and 2,). In addition, we used some of the historical literature describing estimated elevation locations along forest edges.

For the creation of the 1910 forest extent map we utilized six historical topographical maps surveyed and published between 1910 and 1913 (Supplementary Table 1). In addition, we also included two other maps from the railways survey expedition to the Mt Kenya region (1908) and also the sketch route map from the Mackinder journeys to Mt Kenya73 (Supplementary Table 4). We first imported the map scans into the QGIS version 3.36.1 platform and then, using linear transformation, we georeferenced all the individual maps using selected ground control points. For each map we selected on average 5 ground control points (GCPs) based on permanent features such as railway bridges, railway stations, river crossings and confluence points and also main road intersections. The accuracy of the georeferenced maps was acceptable based on calculations of the Root Mean Squared Error of the Mackinder map (RMSE ca 35 pixels ~ 2.9 km), railways survey map (RMSE 4.9 pixels ~ 0.18 km) and topographical maps (RMSE 9 pixels ~ 0.4 km) (Supplementary Table 5).

We underlaid the elevation profile using the digital elevation model from the NASA Shuttle Radar Topography Mission Global (STRM-Version 2.1) 30-m horizontal resolution Global Digital Elevation Model (DEM) images (2000), acquired from NASA Earthdata (https://search.earthdata.nasa.gov/). This allowed us to define the extent of the lower mountain forest by adding 12 elevation points which were described as representing the lowest montane forest extent for some of the regions around Mt Elgon98,100,101, Mt Kenya and the Aberdares70,73. We used these points as reference points while manually digitizing the forest zones from the topographical and survey maps (Supplementary Fig. 2).

We then digitized the forest extent areas by creating vector files of forest extent to represent the area of forest extent estimated for 1910. In addition, we manually digitized all visible roads traversable by car and railway networks, as well as the locations of ‘native’ villages within the topographical maps to map their locations and distribution within the study area. Our 1910–1913 maps did not cover some sections of the northern Aberdares and Mt Kenya forest (Fig. 1, supplementary data). To fill in this earlier map, we used the digitized maps from the cadastral 1925 maps which had forest reserve areas within these zones surveyed and mapped out. Our rationale when using these 1925 maps was based on the assumption that these forest areas existed before the 1920s, given that no documented forest plantation had been done in these areas between 1910 and 1925 and there is no record or likelihood of any ‘reforestation’ practices, or climate-induced forest expansion, between 1910 and 1925. In addition, where major discrepancies in extent of forest boundaries occurred amongst the three map types (Mackinder, topographical and survey maps) around the Mt Kenya and Aberdares regions, we relied on the more accurate and extensive topographical maps to define the forest extent. We then calculated the total area of the digitized forest vector polygon layer using the calculate geometry tool option to generate the estimated total area primary forest in 1910.

For the creation of the maps of 1925 forest area extent, we used the 1925 cadastral maps that were created by the British colonial government in the 1920s in an extensive survey and subdivision exercise to define ‘native’ areas, colonial areas, and also forest reserve areas (Supplementary Table 3). All 12 maps with a scale of 1:250,000 which covered our region of interest had already been georeferenced prior to publication on the source website. We loaded the maps into QGIS and checked for deformities and inaccuracies by comparing physical permanent features such as railway bridges, railway stations, river crossing and confluence locations and also main road intersections. All the cadastral maps had a high accuracy with assessment of four random images indicating an average RMSE of 2 pixels ~ 0.06 km (Supplementary Table 5). We then manually digitized all the maps along the defined boundaries to create the vector polygon areas of the forest reserve areas, the native reserve areas, the colonial reserve areas and also largescale areas under forest excisions and plantations areas (Fig. 2). The colonial reserve areas and the forest reserve areas (Crown forests) were the land areas also referred to as the White Highlands51. The total area under each digitized vector polygon layer was then calculated using the calculate geometry tool option, to generate the total land area in hectares for each area.

Post-colonial resettlement of native smallholder farmers between 1960 and 2012

We utilized the resettlement dataset acquired from the Ministry of Lands and planning (MoLPP), a dataset compiled by Lukalo and Odari for the National Land Commission102, and 1964 historical resettlement maps from Surveys of Kenya (SoK) obtained online from the American Geographical Society library, through the University of Wisconsin-Milwaukee Libraries (https://collections.lib.uwm.edu/digital/collection). The resettlement dataset from the Ministry of Lands and planning (MoLPP), contains information on the number of families, the sizes of land areas allocated to each household, and the period of land allocation. We acknowledge that these resettlement data are not entirely exhaustive as some instances of resettlement initiatives lacked proper documentation and are yet to be surveyed and gazetted in the national land database87. However, they provide a representative sample for considering factors behind forest cover and land use change in the region through time.

For our study region we extracted the resettlement datasets focusing on resettlement programs undertaken within colonial areas, native reserve areas and within the montane regions. To provide a context for interpreting the resettlement plans, we grouped the datasets into three regions; the Central region representing land areas around Mt Kenya and Aberdares forests, the Rift Valley region representing land areas around the Mau and Cherangany forests, and the Western region representing the land areas around Mt Elgon. In total, we analysed 317 registered settlement areas that were established between 1960 and 2012 to assess the resettlement patterns during this period. From this dataset we calculated the total land area allocated for resettlement within each defined region, the average size of land area allocated to each household and the number of households resettled within each region over the period between 1962 and 2012.

In addition, following the georeferencing methods describe before, we georeferenced and digitized the 1964 historical resettlement map produced by Surveys of Kenya (SoK) to map areas of the initial resettlement program post-independence.

Post-independence land use transformation

To create the 1975, 2000, and 2024 maps of forest cover, we relied on classification analysis of Landsat images acquired from United States Geological Survey (USGS) Earth Explorer program (https://earthexplorer.usgs.gov)103,104,105. For the 1975 map analysis, we downloaded seven scene images acquired from the Landsat 1–3 Multispectral scanner (MSS) with less than 5% cloud cover (Supplementary Fig. 3). In order to have a complete image mosaic for the 1975 landcover classification, we used one Landsat image from 1973, four images from 1975, and one from 1976 (all images were acquired between January and March of each year). This was necessary in order to create an image mosaic with the least percentage of cloud cover, which peaks during the dry season (generally from January–March and June–October). We undertook pre-processing of the images by performing Top of Atmosphere (TOA) correction through the Dark Object Subtraction (DOSI) correction (EROS, 2018), using the Semi-Automatic Classification plugging (SCP)106 in QGIS version 3.36.1 (QGIS,2024).

All seven images were then combined to create an image mosaic covering the entire region of interest. For the classification analysis we firstly performed band combination to create two visual images (true color, color infrared and false color composite) to help with object identification (Supplementary Table 6, Supplementary Fig. 3). To define our landcover types, we based our classification on three broad groups defined by the dominant landcover type within the study region107: (1) Primary forest; this class groups naturally occurring low and high canopy forests, bamboo forests and high-altitude vegetation including heath and moorland vegetation types. (2) Largescale plantations; includes areas under tea farms, coffee farms and plantation forests. (3) Small scale farms; includes small areas of mixed cropping and grazing lands used for subsistence farming. To define the study extent for the cover analysis we used the forest reserve extent based on the 1910 and 1925 forest reserve map, generated from the historical maps. We then created a training dataset for the supervised classification by using Google Earth Pro to visualize imagery from the 1973–1976 period and topographical maps produced by the survey of Kenya between 1945 and 1975 (DOS, 1945–1975) (supplementary, Table 2). The use of the topographical maps improved our selection of the training data points of the various landcover types as they provided additional information on the occurrence of large-scale plantations and/or forest zones where it was not visually clear on Landsat imagery. We created a training dataset containing 182 samples (polygons) which we used for the supervised classification and a separate dataset containing 91 samples was used as the validation dataset to assess the classification accuracy. Supervised classification was performed using maximum likelihood classifier function, which is an algorithm which uses data information from the assigned training dataset and uses the pixel information to assign similar spectral pixels to the same defined classes108. To validate the classification accuracy, we performed an accuracy assessment using the Semi-Automatic Classification plugging, by creating a confusion matrix using our validation data which generated the user, producer, overall and Kappa Hat accuracy (Supplementary Table 7).

For the classification of the 2000 map we downloaded seven Landsat 7 Enhanced Thematic Mapper Plus (ETM+) images, one from 1999, five from 2000, and one from 2001 (six images were acquired between January and March, while one image was acquired from December of each respective year) in order to create an image mosaic with the least percentage of cloud cover over our region of interest. For the 2024 map, we downloaded seven Landsat 8 Operational Land Imager and Thermal Infrared Sensor (OLI/TIRS) images, three from 2023 and four from 2024 (five images were taken between January and April and two images from June). We then performed image band combinations to generate the false and true color composites which helped us to visualize the different cover classes during classification (Supplementary Table 6, Fig. 3). Landcover classes for both the 2000 and 2024 maps were defined in the same way to the land cover classes in the 1975 classification. We then created training sites with reference to google earth pro, where we reviewed the high-resolution google earth imagery to create the training datasets for both the 2000 and 2024 images. For the 2000 image, we created a training dataset containing 309 samples (polygons) which we used for the supervised classification and a separate dataset with 109 samples (polygons) was used as the validation dataset. For the 2024 image, we created a training dataset containing 383 samples (polygons) which we used for the supervised classification and a separate dataset containing 122 samples (polygons) was used as the validation dataset to assess the classification accuracy. Similar to the 1975 analysis, supervised classification using the maximum likelihood algorithm was performed on both the 2000 and 2024 images108. Accuracy assessment was then performed using the validating dataset to assess the classification accuracy (Supplementary Table 7)

Estimates of landcover changes were then performed by comparing the areal land cover area of the three cover types between the five land cover maps from 1910, 1925, 1975, 2000, and 2024, to estimate the temporal change in primary montane forest, large-scale plantations and small-scale farms. All forest area extents below 1 ha were excluded in the calculation of total areas covered by primary forest, as these areas mainly represented exotic forest areas within private small holder farms.