Introduction

Since the 1990s, tropical forest landscapes have undergone significant changes, resulting in the degradation and modification of about 930 million hectares (ha) in tropical regions (ITTO, 2020). Within the region, the estimated annual deforestation rate between 2015 and 2020 is 10 million ha. During the 2010s, the largest net losses were recorded in South America and Africa (FAO, 2022). Africa alone has experienced an annual loss of around 3.9 million ha of forest (FAO &UNEP, 2020). Worse still, Africa’s forests, farmland, and pastureland continue to face degradation (Pasiecznik & Reij, 2020). Many anthropogenic pressures are driving these changes, mainly unsuitable agricultural practices, urbanization and industrialization, combined with natural hazards resulting from climate change (Haque & Basak, 2017). These increase the regressive dynamics of land cover, soil degradation, and compromised ecosystem services (Gbedzi et al., 2022; Tsegaye, 2019). Other consequences of land use and land cover (LULC) include biodiversity loss (Hansen et al., 2004) and disruption of biogeochemical cycles (Ramachandra, 2014). Thus, changes in LULC have substantial impacts on the environment and people's livelihoods (Munthali et al., 2022). In Africa, LULC problems are exacerbated by high population pressure and over-reliance on land resources for livelihoods and food, agricultural activities, and illegal logging. As data and insights provided by LULC dynamics are essential for the management of sustainable land resources, examining past changes as well as future trends in LULC of the concerned landscapes are essential to improve sustainable planning, governance, and monitoring (Bunyangha et al., 2021), as well as adopting measures to mitigate climate change (Reddy et al., 2017). LULC reviews are critical elements that can support the implementation of different measures and approaches, such as forest landscape restoration (FLR). The examination of land use to support processes like FLR is critical for effective planning. As per Nedd et al. (2021), studies conducted since 2000 typically acknowledge the need to enhance knowledge of LULC. The studies include, among others, the typology of LULC, the direct and indirect changes associated to LULC, as well as the challenges and knowledge gaps relating to LULC (Nedd et al., 2021). Despite these studies have been carried out to increase awareness of LULC change issues, most FLR initiatives are launched with limited land-use assessments and frequently lack analyses of future trends. Thus, the consideration of successful local practices is often marginalized in planning, which hinders effective FLR action. Hence, it is crucial to evaluate LULC to ensure the sustainability of FLR efforts.

In Togo, the LULC has undergone significant transformations, driven mainly by population growth, urbanization, and agricultural expansion. Agriculture is the main land use in Togo, employing 60% of the country's population (MAEH, 2017). Smallholder farmers practice traditional slash-and-burn agriculture, which involves clearing forests, burning vegetation, and planting crops. These factors are compounded by logging, charcoal production, and overgrazing, which significantly contribute to deforestation and land degradation in Togo. Despite Togo's limited forest cover, which stands at 24.24% (MERF/GIZ, 2016), the country experiences an annual deforestation rate of 0.42% and significant land degradation, estimated at 4.14% between 2000 and 2010 (MERF, 2018, 2020).

To address these issues, Togo has implemented several policies and programs promoting sustainable management practices. The National Land Use Planning Program (PNAT) seeks to promote sustainable land use, including integrated watershed management, sustainable agriculture, and ecosystem-based approaches to natural resource management. Furthermore, Togo has launched several management initiatives to restore degraded forest landscapes. These initiatives include the National Investment Plan for Environment and Natural Resources (PNIERN) (MERF, 2010), the National Reforestation Programme (PNR) (MERF, 2017), the adoption of forest policy matched with a National Forestry Action Plans, which envisages increasing the national forest cover from 24.24% in 2016 to 25%, 26%, and 30% in 2025, 2030, and 2050, respectively. The initiatives have a clear objective to enhance forest governance, increase reforestation to 1 billion trees, and restore 1.4 million ha by 2030 (MERF, 2017, 2021). Additionally, the initiatives aim to restore degraded forest ecosystems in line with the New York Declaration on Forests, the Bonn Challenge, and the AFR100 initiative.

In addition, Togo has implemented several approaches for FLR, including the “community forest management” approach in Tchamba prefecture. This approach aims to give local communities the rights and responsibilities to manage their forests. This is how Tchamba, a district in Togo's central region that is considered a biodiversity hotspot, is a key beneficiary of several projects that aim to address the negative effects of LULC to better implement FLR, such as the Forests4Future (F4F) project. This project is based on the achievements of the Preparation for REDD + and Forest Rehabilitation programme in Togo (ProREDD) and the International Waldpolitik (IWP) programme. These programmes support community forest management, socio-economic characterisation of households, and Tchamba Prefectoral Forest Inventory (IFP) to assess the potential for timber and non-timber products (DFS/GOPA, 2016; IWP/AFR100/GIZ, 2020a, b).

However, despite numerous initiatives operating at different levels and scales, investigations into landscape dynamics have remained relatively limited. Thus, information available on the landscape of this area is incomplete. Besides to the limited knowledge of LULC in the area, there is a lack of studies focusing on the forecasting of future trends. As temporal changes in land use are an equally important component for any planning (Gülersoy, 2014) and as FLR is a lengthy process, it is essential to examine historical land use. Restoration actions and sustainable management of natural resources require a critical understanding of LULC and their impacts on natural resources (Munthali et al., 2022). Against this background, the aim of this study is to (i) analyze changes in LULC in Tchamba Prefecture from 1991 to 2022; and (ii) predict future trends for the 2050s to increased action on FLR issues.

To achieve these objectives, landscape dynamics are investigated using Geographic Information Systems (GIS) and Remote Sensing (RS) (Forkuo et al., 2021). Landsat satellite images are analyzed to achieve the study objectives. The images are proven to provide important radiometric and geometric qualities for LULC analysis. In addition, these images provide a long-term archive of satellite data for historical analysis (Gutman & Masek, 2012). The historical LULC analysis is integrated with the Cellular Automata-Markov modelling to understand the future LULC patterns. Findings inform policymakers and practitioners on land implementation programs, helping them to understand the current state of LULC. As pointed out by Leta et al. (2021), this knowledge can be useful for improving land use policies with a view to long-term land use planning to cope with future changes. The results will also contribute to a greater pool of data on LULC dynamics that can feed into the global data set, thus strengthen ongoing positive actions and properly guiding FLR activities to the benefit of all.

Materials and Methods

Study Area

The district of Tchamba is situated in the central region of Togo, located between latitudes 8°22' N and 9°20' N and longitudes 1°15' E and 1°38' E. It is approximately 300 km from the capital city (Fig. 1). It is bounded to the South by Est-Mono prefecture, on the North and East by Benin Republic, and on the West by Sotouboua and Tchaoudjo Prefectures. The study area covers about 3,166 km2, i.e. 20% of the Central Region’s surface area. The Prefecture is administratively divided into ten (10) cantons (Tchamba, Larini, Alibi, Adjéidè, Affem, Koussountou, Balanka, Kaboli, Goubi, and Bago) grouped in three (3) municipalities (Tchamba 1, Tchamba 2, and Tchamba 3) (INSEED, 2011). The climate in this region is characterized as Sudano-Guinean. It experiences a rainy season that typically lasts from April to October, with an average annual rainfall of 1100 to 1300 mm. The dry season runs from November to March and, during this time, there is a continental north-east trade wind known as harmattan, which blows from November to February. It is a hydrographic system formed by 94 rivers and streams, with Mono and its tributary Ogou being the main streams (IWP/AFR100/GIZ, 2020a). At the Prefectural level, the population is estimated at 156,814 people, with a density of 49 people/Km2 (IWP/AFR100/GIZ, 2020a).The main activities in the districts are agriculture (food crops), livestock (poultry, small ruminants and large livestock), trade (general foodstuffs, cash crops, especially cashew nuts, building materials, etc.), crafts (weaving, sculpture, carpentry, pottery, hairdressing, sewing, etc.), transport (taxis and motorbike taxis), and timber exploitation (timber, energy wood, service wood). This prefecture was selected because of its status as the country’s largest forest area, covering a total of 93,287 ha. It has a great ecosystem diversity comprising the Abdoulaye Wildlife Reserve (30,000 ha), Community Forests (Affem Boussou, Alibi 1, Bago, Goubi, and Koussountou) (MERF/GIZ, 2016) and sacred forests (Hounkpati et al., 2022; Lynch et al., 2018). It not only benefits from several support programs and initiatives, but it is also a target of the "Forests4Future" project, hosting the pilot phase in Togo.

Fig. 1
figure 1

Location of the study area

Full size image

Data Collection

This study combines the use of (i) satellite data and (ii) ground survey data collected using a handheld Geographical Positioning System (GPS) instruments to evaluate the current LULC. The ground survey data enable the creation of a training and evaluation database for the RS classification algorithm. Ground survey data are crucial in understanding land use dynamics and to perform the statistical evaluation of different land use categories (Regasa & Nones, 2022). Landsat data with complete spatial and spectral resolution (30 × 30 m) from the United States Geological Survey (USGS) website (https://earthexplorer.usgs.gov/) were captured during the end of the dry season (January to February) and has less than 5% cloud cover. Multispectral data from Landsat 4–5 (Thematic Mapper), Landsat 7 (Enhanced Thematic Mapper Plus), and Landsat 8 (Operational Land Imager/Thermal InfraRed Sensor) (Bunyangha et al., 2021) were used (Table 1). From the data, the spectral bands used to classify the images are Blue, Green, Red, Near- Infra-Red, Short Wave Infra-Red 1 and 2 (Gyamfi-Ampadu et al., 2020). The bands provide a spatial resolution of 30 m, sufficient to monitor forest dynamics (García-Álvarez et al., 2022). In addition to the data acquisition, Landsat images offer corrected geometric and radiometric data to analyze land use change (Yang & Lo, 2002).

Table 1 The Metadata on the satellite images acquired
Full size table

Data Analysis and Processing

Image Processing and Classification

The Landsat images used in the study provide Level 1 terrain-corrected data, which had already undergone both geometric and radiometric corrections (Bunyangha et al., 2021) and were used directly as inputs in the classification process. However, as the images were acquired on different dates and with different sensors, it was necessary to harmonize them in order to apply a multitemporal approach. For classification, it was necessary to establish a reference that defines how the target land cover types appear on the images. Thus, a definition of the reference points (land use classes) was made by using the high resolution images accessible on Google Earth for the year 2022 (Zhao & Zhu, 2022). Based on the training plots, the images were classified for the four selected dates using verified target land cover types. The satellite images were projected into the WGS 84/UTM 31 N coordinate system prior to classification. Supervised classification using the RandomForest (RF) algorithm, developed by Breiman (2001) was applied to the satellite images. This algorithm was chosen for its good predictive capabilities for LULC (Gislason et al., 2006) and in the case of temporal analysis. It is a non-parametric supervised classification algorithm that combines the decision tree algorithm with an aggregation technique. In order to create multiple small classification trees, this algorithm randomly selects a sample of observations and variables multiple times, according to Breiman (2001) and Grinand et al. (2013). The small trees are then combined and a majority voting rule is applied to determine the final classification category, as per Grinand et al. (2013). Subsequently, validation of the classification was carried out based on 1/3 of the ground survey points taken in the field. Once the classification and validation were done on the 2022 reference map, the same operations were performed on the other scenes. The pre-determined classes delineated were dense forests/gallery forests, open forest/woodland, plantation, tree and shrub savannah, mosaic crop/ fallow land, and built-up (Table 2).

Table 2 LULC classes and their description
Full size table

Accuracy Assessment

Image accuracy assessment is one of the crucial steps in classification. As stated by Bunyangha et al. (2021), no LULC classification process can be considered complete without a thorough accuracy assessment. Assessing the accuracy of the classification provides a certain level of confidence in the results, in the detection of changes, and in the representativeness of the ground features on the corresponding classified image (Leta et al., 2021). For accuracy in this study, it was necessary to establish a reference that defines how target land cover types appear on the images. The reference definition was made based on Google Earth high-resolution images for the year 2022. The kappa index was employed to evaluate the classification accuracy (Jia et al., 2018). By comparing the total corrected and total samples, the overall precision provides an overview of the number of correctly classified sites. The Kappa coefficient, a measure of agreement, compares actual and expected agreement with reference data (Bunyangha et al., 2021). When its value is close to 1, the classification is known as accurate and not accurate when the value is close to 0 (Regasa & Nones, 2021). In this research, the Kappa coefficient was computed using Congalton (1991) equation.

$$\begin{array}{cc}K& =\end{array} \frac{N\sum_{i=1}^{r}\begin{array}{cc}{X}_{ii }& \begin{array}{cc}-& {\sum }_{i=1}^{r}\left(\begin{array}{cc}{x}_{i+}& \begin{array}{cc}*& {x}_{+i}\end{array}\end{array}\right)\end{array}\end{array}}{\begin{array}{cc}{N}^{2}& \begin{array}{cc}-& {\sum }_{i=1}^{r}\left(\begin{array}{cc}{x}_{i+}& \begin{array}{cc}*& {x}_{+i}\end{array}\end{array}\right)\end{array}\end{array}}$$
(1)

where, K = Kappa Coefficient, r = number of rows/columns in the error matrix, N = total number of samples, xii = sum of correctly classified samples, xi +  = row i total, while x + i = column i total.

Land Use and Land Cover Change Detection

The method used to detect changes in LULC is Post Classification Comparison (PCC). This method implies the comparison of sequentially and independently classified images by an overlay to establish how the LULC classification at the initial date has evolved toward a different category at the final date. The matrix rows produced indicate the land use at the initial date (t1) and the columns represent the land use at the later date (t2) (Table 4). The vectors in the rows illustrate the evolution of the land use type between date t1 and t2, while the vectors in the columns describe the land use type at t1 that has converted to the land use type at t2. The data on the diagonal represent areas where LULC remains consistent. In this study, changes for each of the six LULC categories were calculated using 6 × 6 transition matrices.

Annual Rate of Land Use and Land Cover

The annual rate of change measures the extent of change in occupancy classes per year. This variable serves to identify the LULC conversion rate (Bunyangha et al., 2021). In this study, the Puyravaud (2003) formula was used to calculate the annual rate of change (r) in different LULC classes (C1, C2) at specific times (t2, t1).

$$r=\left[\left(\begin{array}{cc}1/\left(\begin{array}{cc}{t}_{2}-& {t}_{1}\end{array}\right)*& \mathit{ln}\left(C2/C1\right)\end{array}\right)\right]*100$$
(2)

where: r = rate of change (%); C1 = area of the class at date t1; C2 = area of the class at date t2 and t = years between the two dates with t2 > t1. A positive value for the rate of change indicates an increase, while a negative value indicates a decrease, and when the value approaches zero, it means an equilibrium in the land use class (Eastman, 2009; Leta et al., 2021).

Future Prediction of LULC

TerrSet Geospatial Monitoring and Modeling System software, specifically the Land Change Modeler (LCM), was used to predict future LULC patterns for the years 2030 and 2050, using historical satellite images classified in this study. This model is proven to have a high capacity for dynamic projection, a deep calibration and the capability to simulate several types of land cover (Eastman, 2009; Leta et al., 2021). Using two dates, t1 as the initial date and t2 as the final date, LCM examines the variation in land use, calculates the modifications, then anticipates the future evolution of the land use (t3, t4) based on the relative transition potential maps (Qacami et al., 2023).

The LCM model produces transition potential maps through three approaches: multi-layer perceptron neural network (MLP), logistic regression, and machine learning via similarity-weighted instances (SimWeight) (Eastman, 2016; Mishra & Rai, 2016). The MLP model is characterized by the use of a minimal number of parameters, it is more accessible and has been improved to offer an automatic mode (Regasa & Nones, 2022). Indeed, three main units make up the MLP neural network: the input layer, the hidden layer containing computing nodes, and the output layer (Eastman, 2009). To model transitions, these units are used. For each transition from one land use type to another, a potential change model is generated as a transition sub-model. In cases where multiple transitions share similar underlying variables and depend on the vulnerability of the LULC to change to other land use types, these sub-models are grouped together. This creates a composite change potential map, also known as transition adequacy, for this land use category (Bunyangha et al., 2021).

For predicting the future LULC scenarios, it is necessary to have access to historical LULC maps and a set of driving forces as input data (Regasa & Nones, 2021, 2022). In this study, the following driving variables were considered for the LULC Change Model (LCM) simulations: the distance from disturbance, the distance from the stream, the distance from urban areas, the distance from the road, and the evidence likelihood. All of these variables are dynamic and differ from static variables like elevation and slope (Regasa & Nones, 2021, 2022). The LCM simulations in this study considered several factors that can affect driving, such as the distance from disturbances, streams, urban areas, and roads, all of which are dynamic variables unlike elevation or slope. This information was provided by Regasa and Nones (2022). The study analysed the changes LULC trends for the years 1991, 2002, 2012, and 2022 in order to predict future LULC changes for 2030 and 2050. The modelling in this study was based on a business-as-usual scenario, which assumes that future land use patterns will follow the trends observed in the past (Clerici et al., 2019).

Future trends in LULC will follow the same pace as historical and recent trends in the region, being driven by influencing factors similar as mentioned by Bunyangha et al. (2021). To assess the predictive performance of the LCM model for future LULC, a map of predicted LULC for 2022 was generated from data for 1991, 2002, and 2022. The multiple objective land allocation (MOLA) and the Cellular automata integrated into the LCM model were used to spatially assign Markov transitions. To validate the Markov model, we compared the simulated land LULC map for the year 2022 with the classified LULC map of the same year (Kundu et al., 2017). The Kappa variations as a basis for comparison was used. The study design is summarized in Fig. 2 below.

Fig. 2
figure 2

Overview of the methodology used for the detection and prediction of LULC

Full size image

Results

Spatio-temporal Dynamics of LULC in Tchamba Prefecture

The results of LULC classification for 1991, 2002, 2012, and 2022 of Tchamba prefecture were obtained with Kappa values of 0.92, 0.83, 0.81, and 0.85, respectively. The results show that, in 1991, dense forests/gallery forests covered 6026.85 ha, which translates to 18.85% of the Prefecture (Table 3, Fig. 3). Open forest/ woodland, and tree and shrub savannah covered 64,496.66 ha (20.17%) and 159,965.14 ha (50.04%), respectively, of the total area of the prefecture. Plantation occupied the least area of 2577.18 ha (0.81%). Considering these classes, the forested area covered 287,306.83 ha (89.87%) of the prefecture with tree and shrub savannah being the predominant class of the forested zones. The non-forest area comprising mosaic crop/ fallow land and built-up areas cover 29,505.14 ha (9.23%) and 2877.02 ha (0.90%), respectively.

Table 3 LULC composition and change trend of Tchamba Prefecture from 1991 to 2022, Area in ha and %, Change and annual change rate in %
Full size table
Fig. 3
figure 3

Classified LULC maps from 1991 to 2022

Full size image

The results further show that, in 2002, there were several changes in the composition of the LULC. Most of the classes decreased slightly except open forest/ woodland, which declined by 32,325.52 ha. This area is more than half of the open forest/woodland coverage in 1991 (Table 3). With an annual loss rate of 6.32%, open forest/woodland decreased to 32,171.14 ha (10.06%). Similarly, dense forests/gallery forests recorded an annual decline rate of 1.30%, decreasing to 52,223.83 ha (16.34%), i.e. a decline of 8043.93 ha. Tree and shrub savannah, however, increased slightly to 179,845.65 ha (56.26%), which is a net gain of 198,880.51 ha and remained the predominant land class. Plantation, mosaic crop/fallow land, and the built-up area also increased slightly to 3512.13 ha (1.10%), 44,888.82 ha (14.04%), and 7047.37 ha (2.20%), respectively. The forest area continued to cover a large part of the area; although it decreased to 267,752.81 ha (83.75%).

LULC coverage in 2012 reveals that dense forests/gallery forests registered an annual loss rate of 1.51% and decreased slightly to 44,893.48 (14.04%), which was a decline of 7330.41. Also, Tree and shrub savannah decreased significantly to 151,732.74 ha (47.46%), a decline of 28,112.90 ha. However, the open forest/woodland increased by 13,818.29 ha. With an annual rise of 3.29%, the area of open forest/woodland increased to 44,688.91 ha (13.98%). Also, mosaic crop/fallow land and plantation increased significantly. The plantation increased to 6160.47 ha (+ 1.93%) which was a net gain of 2648.34. This translates to an annual rate of 5.6% over the period. On the other hand, mosaic crop/fallow land increased to 63,410.51 ha (19.84%), translating to a net gain of 89,395.24 ha and an annual positive rate of 3.45%. Built-up area land increased slightly to 8802.89 (2.75%). All these conversions led to a decrease in forest area to 247,475.59 (77.41%) (Table 3, Figs. 3 and 4).

Fig. 4
figure 4

Trend of LULC in Tchamba Prefecture from 1991 to 2022

Full size image

2022 was marked with the most significant conversions. The LULC dynamics indicated that dense forest/gallery forest stabilised at 43,808.8 ha (13.70%), which translated to a decline of 0.24% from the previous epoch. However, tree and shrub savannahs declined by 16.5% per year, i.e. a drastic decrease of 122,663.10 ha to a coverage of 29,069.64 ha (9.09%). Open forest/wooded land slightly increased to 58,507.20 ha (18.30%), translating to a net gain of 12,517.76 ha and a positive annual rate of 2.69%. In the same period, there was a significant increase in forest plantations and crop/fallow. Forest plantation recorded a net gain of 18,989.13 ha to reach 25,149.60 ha (7.87%). This translates to an annual rate of 14.07%, i.e. crops/fallow recorded a net gain of 89,395.24 ha to coverage of 152,805.74 ha (47.80%). The annual gain rate in this case was 8.80%. However, the built-up area revealed a varied trend. The increase in their surface area remains slight at 10,347.96 ha (3.24%). In terms of forest dynamics, forest area declined from 89.87% to 48.96%, a decrease of a 45.52% between 1991 and 2022.

Composite analysis of the conversion matrix shows that tree and shrub savannah have experienced the highest conversion, followed by open forest/woodland and then dense forest/gallery forest at decreases of 5.50%, 1.03%, and 0.31%, respectively between 1991 and 2022 (Tables 3 and 4). Indeed, 86.05% of Tree and shrub savannah were completely converted to 52.27% of mosaic crop/fallow land, 19.32% of open forest/woodland, 7.03% of dense forests/gallery forests, 7.09% of plantation, and a mere 2.33% of built-up (Table 4, Fig. 5). The matrix shows that 79.40% of open forest/woodland was converted mainly into mosaic crop/fallow land (45.67%), dense forestss/gallery forests (20.28%), and forest plantation (7.87%). As for dense forests/gallery forests, the matrix reveals that 66.42% of this class was converted to mosaic crop/fallow land (37.37%), open forest/woodland (18.78%), and forest plantation (5.21%). Although the conversion of dense forests/gallery forest, open forest/woodland and tree/shrub savannah was significant, the matrix shows that in 2022 dense forest/gallery forest and open forest/woodland retained a significant part of their area respectively at 45,625.59 ha. However, tree and shrub savannah have totally disappeared, i.e. 26,150.67 ha instead of 158,590.8 ha in 1991. The mosaic crop/fallow land, and forest plantation classes, however, have increased significantly, i.e. 153,845.19 ha instead of 29,706.11 ha and 26,234.37 ha instead of 2693.61 ha in 1991 (Tables 4).

Table 4 LULC transition matrix from 1991 to 2022, Area in ha
Full size table
Fig. 5
figure 5

Net change for each LULC category from 1991 to 2022

Full size image

Validation of Future LULC Prediction

To assess the reliability and robustness of the prediction algorithm, LULC in 2022 was predicted based on historical LULC maps from 1991, 2002, and 2012 using the LCM model. The results were then compared to the actual LULC maps in 2022 (Fig. 6). The results reveal that, from a spatial perspective, there are virtually no differences between projected and observed land use maps. The Kappa index is estimated to be 0.91, while the overall error between predicted and actual land use is only 0.001% (Table 5). These values assume that the model and comprehensive LULC maps are well suited to predict future LULC. The slight non-significant overestimates observed in the land cover classes, except for open forest/woodland, tree and  shrub savannah, mosaic crop/fallow land, did not affect the reliability and general acceptability of the prediction of future land use and cover conditions using this method.

Fig. 6
figure 6

Predicted and actual land use and cover map in 2022, (a) predicted LULC map by the model, (b) actual LULC map

Full size image
Table 5 Validation of the LULC prediction based on the actual and simulated 2022 LULC image
Full size table

Future LULC Prediction

Future changes in LULC in Tchamba Prefecture were predicted for 2030 and 2050. The probable percentages of changes in landscape units for 2022–2030 and 2030–2050 were analysed using the transition probability matrix. The coverage highlights the changes observed from the simulated future images between 2022 and 2050, the percentage and the rate of change in the area occupations. The results show significant changes in mosaic crop/fallow land from 47.80% in 2022 to 49.26% in 2030 and 51.66% in 2050. Mosaic crop/fallow land will be the predominant LULC type. The rate of increase observed between 2030, i.e. 3.05% and 2050, i.e. 4.88% shows that mosaic crop /fallow land will increase significantly between 2022 and 2030 compared to changes between 2030 and 2050. Dense forests/gallery forests will slightly stabilise at 13.70% in 2022 and 13.23% in 2030. However, a significant increase will be observed in Dense forests/gallery forests in 2050 with a coverage of 16.17%, which is a rate of 22.24%. On the other hand, a continuous increase in plantation land cover will be observed. Plantation land cover is expected to increase significantly from 7.87% in 2022 to 9.10 in 2030 and slightly to 9.43 in 2050. Additionally, built-up is expected to increase slightly from 3.24% in 2022 to 3.40%. However, Open forests/woodland, as well as tree and shrub savannah, will continue to decline. Open forest/woodland will decline from 18.30% in 2022 to 16.92% in 2030 and 13.36% in 2050. Tree and shrub savannah will decrease from 9.09% in 2022 to 8.16% in 2030 and 5.97% in 2050 (Table 6, Fig. 7). In terms of forest dynamics, forests change only slightly between 2022 (48.96%) and 2050 (47.41%).

Table 6 Area coverage of LULC, percent, and rate of changes between 2022 and 2050
Full size table
Fig. 7
figure 7

Future trends in LULC from 2022 to 2030 and 2050

Full size image

The transition matrix analysis provides the transition probabilities of each land cover class to transform into other classes over 2022–2050. The results show a negative trend for open forest/woodland, i.e. 26.99%, in favour of dense forests/gallery forests and tree and shrub savannah and mosaic crop/fallow land (Table 7, Fig. 8). As for the tree and shrub savannah, they transform at a rate of -34.35% to mosaic crop/fallow land, built-up and plantation. Finally, mosaic crop/fallow land, plantation and built-up will continue to increase at 19.89%, 8.08%, and 5.13%, respectively.

Table 7 LULC transition matrix from 2022 to 2050, Area in ha
Full size table
Fig. 8
figure 8

Trends in LULC of 2022 and 2050 in Tchamba Prefecture

Full size image

Discussion

The study integrates RS approaches and GIS to analyse LULC changes from 1991 to 2022 and predicts future trends through 2050 using the Cellular Automata-Markov modelling technique. The RandomForest algorithm is applied for classification (Breiman, 2001), integrating determining factors relating to dynamic and static variables to enhance the precision of the projections (Regasa & Nones, 2022). The historical analysis of the LULC dynamics was obtained with Kappa values of 0.92, 0.83, 0.81, and 0.85 for the four dates. The quality of the classification confirmed by the Kappa index values is considered statistically acceptable as the image analysis results with a Kappa value above 0.50 are considered good (Pontius, 2000). Similarly, the supervised classification used in this study is among the most appropriate methods for studying LULC by remote sensing because it involves using verification zones (Masoud & Koike, 2006; Nagendra et al., 2006). Regarding the simulation, the value of the kappa index was 0.91, exceeding 80%, and the error was around 0.001%, thus illustrating a solid agreement between the simulated and real LULC maps (Li et al., 2020). Our model’s results show high accuracy and are supported by the few additional studies on Togo as well as be explained by socio-political factors.

Over the past 31 years, the analysis of LULC indicates a gradual decline in dense forests/gallery forests, open forest/woodland, and tree and shrub savannah. Meanwhile, there is an increase in plantation, mosaic crop/fallow land, and built-up areas. The conversion matrix analysis reveals that tree and shrub savannah underwent the highest conversion, followed by dense forests/gallery forests and open forest/woodland. Notably, the most significant losses in dense forests/gallery forests and open forest/woodland occurred between 1991 and 2012. These findings align with regional studies, reflecting similar changes (Adjonou et al., 2010; Kombate et al., 2020). Findings can largely be attributed to socio-political challenges in Togo during the 1990s, leading to the cancellation of international cooperation programs. Additionally, political instability negatively impacted the economy and transactional capacities, contributing to a precarious socio-political climate (Aisen & Veiga, 2013; Bunyangha et al., 2021; Dalyop, 2019). This has affected natural resource management, particularly in protected areas like the hot spot of Abdoulaye Wildlife Reserve (RFA), which experienced degradation due to population invasion and non-conservative agricultural practices (UICN/PACO, 2008). The studies by Adjonou et al. (2010) in the area of the reserve indicates a decrease in forest formations in favour of mosaic crop/fallow land and bare soil between 1972 and 2000. The reasons lie in the non-participatory way these classified areas were created and their management system, which was often based on repression (UICN/PACO, 2008). Additionally, the rise in charcoal and high wood energy production (1000 to 1500 tones of wood per month) in the region, intensified agricultural activities, while the exploitation of river banks further contributes to forest degradation in Tchamba Prefecture (Adjonou et al., 2010; Kaina et al., 2021; Kombate et al., 2020).

However, between 2012 and 2022, dense forests/gallery forests almost stabilised, while open forest/woodland increased to 18.30%, nearing 1991 levels. Political stability, demonstrated by involving local communities in forest management since 2008, can explained these positive changes. The revision of Abdoulaye Faunal Reserve leading to the retrocession of certain parts of the reserve to the communities and vice versa to the recognition of the reserve's state status and the creation of community forests (Alibi-I and Bago) with the retroceded parts of the reserve (MERF, 2016) showcase successful community-based forest management, offering insights for landscape-level restoration in the Prefecture of Tchamba. Through the creation of community forests, the state is accorded legal status to the communities by making them responsible for managing their forest. This model of community forest management, now used as a key restoration model, has allowed for managing several conflicts related to land management and preserving several vegetation resources at the landscape level in the Prefecture. Tchamba remains a successful case where there is at least one community forestry initiative per canton (Issifou et al., 2022). These examples of community forestry have been proven to bring important lessons and insights to the field of FLR and, in particular, to restoration efforts targeted at rural communities (Niraula et al., 2013; Paudyal et al., 2017), as is the case in Tchamba Prefecture.

Between 2012 and 2022, there was a significant decline in tree and shrub savannah, alongside an increase in mosaic crop/fallow land and plantation. Restricted access to the Abdoulaye Wildlife Reserve and community forest areas led to farmers to convert tree and shrub savannah to mosaic crop/fallow land for farming. According to Issifou et al. (2022), to access the new land, over 82% of the farmers clear the forest and even the shallows. The rise in plantations, particularly cashew trees (Anacardium occidentale), is driven by economic value and the cultivation of various plant species (shea (Vitellaria paradoxa), African locust (Parkia biglobosa), Kapok (Bombax costatum), Black Plum (Vitex doniana)) is prioritized for economic, religious, and cultural reasons in the Prefecture (Kokou et al., 2015). Further, Gmelina (Gmelina arbore), Teak (Tectona grandis), and Albizia (Alabizia ferruginea) are increasingly being planted by the people following their promotion by restoration programmes in the area.

The modelled results indicate that mosaic crop/fallow land, the predominant LULC, is expected to dominate in 2050, showing a more significant increase until 2030. This rise is associated with the continued trend of decreasing tree and shrub savannah, open forest/woodland, as shown by the current dynamics in this study and other studies in the region (Adjonou et al., 2010; Kokou et al., 2015; Kombate et al., 2020).This increase is probably linked to ongoing Togolese legislation and the persistent demand for arable land, given the Prefecture of Tchamba's high agricultural potential, attracting people from surrounding areas. The increase in land for agriculture is reported in similar studies in Uganda (Bunyangha et al., 2021), Ethiopia (Leta et al., 2021), Kenya (Kipkulei et al., 2022), Ghana (Kleemann et al., 2017), and many other countries (Phartiyal et al., 2023).

It is remarkable to underline that dense forests/gallery forests are projected to slightly increase in 2050, likely due to the continued rise of the population for the status of the main reserve of the Prefecture and conservation efforts. Community forests initiatives and current resource management policy may contribute to this positive trend through 2050. The rise in planted land reflects a strong commitment to economically viable tree plantations, supported by ongoing restoration projects that contribute to income generation and provide ecological benefits. Changes in land use should not only be considered from a negative perspective but rather how these changes can impact positively upon planning actions, especially in FLR. Trends in LULCs, whether in favour of agricultural land use (Stanturf et al., 2015) or not, always provide learning opportunities that can be used to better plan actions in the FLR context. Moreover, it is critical that barriers to FLR implementation and their understanding are addressed before restoration initiatives are scaled up to meet national and international commitments (McLain et al., 2021).

LULC analysis is a valuable tool for FLR efforts. The continuous expansion of mosaic crop/fallow land shows the need for targeted restoration, particularly agro-ecological innovations, to avoid total degradation. This study underscores the significance of commercial tree plantations for future FLR activities, offering environmental benefits and promoting equity (McLain et al., 2021). The evaluation of restoration success, especially based on the community forest restoration model, should be put at the centre of future actions for scaling up throughout the prefecture and beyond its borders. This will allow for consideration of the socio-economic and technological contributions of local communities (Nielsen-Pincus & Moseley, 2013) that are often neglected in the case of FLR. Yet, participatory processes, including community forest-based restoration models, play a crucial role in achieving faster and more successful FLR implementation at the local level (Fleming & Fleming, 2009; Sayer et al., 2013). While Landsat imagery, with a spatial resolution of 30 m, offers long-term image availability, high-resolution imagery should be used to accurately assess the restoration efforts of local communities. In addition, a complete analysis should incorporate an examination of socio-economic data and scenarios that outline and spatialise various potential futures. These components, combined with a temporal analysis of LULC, as suggested by Bunyangha et al. (2021), are essential for a thorough understanding of the drivers of change in LULC.

Conclusion

LULC is crucial for effective landscape restoration initiatives. This study reveals spatiotemporal changes in LULCs since 1991, with predicted trends for 2050 using the Cellular Automata-Markov modelling technique. Historical analysis identifies two phases. In the first phase, forest lands were converted to farmlands due to socio-economic pressures in the 1990s, with mosaic crop/fallow land dominating the LULC changes. The second phase is marked by the stabilization of dense forests/gallery forests decline through participatory management and community forest recognition. Community forests became a key model for landscape restoration, with the active participation of the population in reforestation, especially in agroforestry plantations. The results of the predictions for 2050 indicate continued dominance of mosaic crop/fallow land due to rich soil attracting agricultural practices. While mosaic crop/fallow land intensifies, there is a conservation of dense forests/gallery forests and an increase in plantation, reflecting policy awareness and local community efforts. These findings inform restoration planning, supporting local communities and ecological integrity. These results provide information for policy makers designing reforms for natural resource preservation and guide stakeholders in prioritizing restoration efforts for sustainable management. The scientific community and other actors can use these results to enrich assessments and achieve FLR objectives.