The impacts of the Anthropocene on climate and biodiversity pose societal and ecological problems that may only be solved by ecosystem restoration. Local to regional actions are required, which need to consider the prevailing present and future conditions of a certain landscape extent. Modeling approaches can be of help to support management efforts and to provide advice to policy making. We present stage one of the LaForeT-PLUC-BE model (Landscape Forestry in the Tropics–PCRaster Land Use Change–Biogeographic & Economic model; in short: LPB) and its thematic expansion module RAP (Restoration Areas Potentials). LPB-RAP is a high-resolution pixel-based scenario tool that relies on a range of explicit land use types (LUTs) to describe various forest types and the environment. It simulates and analyzes future landscape configurations under consideration of climate, population and land use change long-term. Simulated Land Use Land Cover Change (LULCC) builds on dynamic, probabilistic modeling incorporating climatic and anthropogenic determinants as well as restriction parameters to depict a sub-national regional smallholder-dominated forest landscape. The model delivers results for contrasting scenario settings by simulating without and with potential Forest and Landscape Restoration (FLR) measures. FLR potentials are depicted by up to five RAP-LUTs. The model builds on user-defined scenario inputs, such as the Shared Socioeconomic Pathways (SSP) and Representative Concentration Pathways (RCP). Model application is here exemplified for the SSP2-RCP4.5 scenario in the time frame 2018–2100 on the hectare scale in annual resolution using Esmeraldas province, Ecuador, as a case study area. The LPB-RAP model is a novel, heuristic Spatial Decision Support System (SDSS) tool for smallholder-dominated forest landscapes, supporting near-time top-down planning measures with long-term bottom-up modeling. Its application should be followed up by FLR on-site investigations and stakeholder participation across all involved scales.
A range of policy goals have emerged in recent years to tackle the challenges posed by the Anthropocene such as climate change [[
The concept of Forest and Landscape Restoration (FLR) has gained prominence as an important management strategy to restore ecological functionality and enhance human well-being in deforested or degraded landscapes [[
A significant portion of forest restoration opportunities can be found in smallholder-dominated forest landscapes. While smallholders inhabit approximately 33% of forest landscapes globally [[
Modeling approaches are valuable tools for assessing restoration opportunities in such contexts while considering smallholder needs and their impacts on forest landscapes. Dynamic and spatially explicit models have proven useful to support management decisions and have been utilized in Spatial Decision Support Systems (SDSS) [[
A further challenge refers to the consideration of global and local scales in a single simulation approach [[
Against this background, we developed a new modeling approach to support environmental management and policy decision-making processes based on long-term scenario simulations with a particular focus on smallholder-dominated forest landscapes. The approach builds on two conceptual pillars: (
The novel modeling approach integrates deterministic and stochastic functions within a dynamic and spatially explicit cellular automata model inheriting features of the PLUC model. The newly developed model relies like PLUC on the PCRaster Python Monte Carlo framework for probabilistic modeling [[
The presented study is part of the LaForeT project (
For the purpose of this model development study, we present in section 2 the underlying structural skeleton of the LPB-RAP model. This is supported by supplemental materials S1 File, which introduces further background information and further newly incorporated and applied methods. Section 3 showcases modeling results for the case study area of Esmeraldas province in Ecuador (see S2 and S3 Files for case study parametrization and S4 File, as well as S5 File forfurther results) to highlight the model's versatility as a scenario-driven simulation tool for studying smallholder-dominated LULCC and RAP in forest landscapes. The Esmeraldas province is used because it has been under high deforestation pressure in the recent past, but also has several conservation zones [[
In this section, we first introduce the LPB-RAP modeling approach (section 2.1). We then describe essential new model components for extended landscape and LULCC simulation: determinants for long-term simulations (section 2.2), policy enforcement levels and related simulation choices (section 2.3), and allocation procedures (section 2.4).
LPB-RAP, including a detailed manual and the required data, is available open-source via GitHub (see S5 File). Open-source software used comprises SAGA GIS (version 8.0.0+), QGIS (version "Hannover"+), PCRaster (version 4.3.3+), Python (version 3.9+) and R (version 4.2.0+).
This section provides a short overview of the major commonalities and differences between the base model PLUC [[
At the core, LPB-RAP and PLUC both refer to probabilistic LULCC-models that provide information on application-specific area potentials in simulated future landscape configurations. The original PLUC model [[
Altogether, LPB-RAP inherited from PLUC: (
While PLUC provides uncertainty information only for a particular focus aspect (i.e., potentials for bioenergy crops) across all samples of different landscape trajectories, in contrast, LPB delivers uncertainty information for the entire aggregated landscape. This new feature supports the user in understanding simulated landscape configurations in a more coherent fashion and investigating areas of interest in their spatial context. For further information, see also section 3.1.
Many new model methods, parameters and assessment features were implemented in LBP-RAP targeting long-term simulation in a fine-scale spatial resolution, especially relevant for the consideration of smallholder-dominated forest landscapes. Section 2.1.2 describes the most important new main model features (long-term simulation approach, footprint approach, explicit LUTs, differentiated forest dynamics simulation) while further details can be found in the prelude section of S1 File. Table A1 in S1 File lists major components of both models as well as their commonalities and differences. The same table also provides references to sections where more information about LPB-RAP can be found.
The following subsections introduce main model features of model stage 1, including outlooks on further conceptualized development stages; for a brief description see also section 5.
LBP-RAP comprises three modules of LULCC simulation: LULCC_basic functions as the basic probabilistic simulation module, LULCC_mplc as an aggregating module (mplc stands for „most probable landscape configuration") and LULCC_RAP as an interpretation module (see Fig 1).
Graph: The LPB-RAP model flowchart depicts the major simulation steps and outputs within the user-defined baseline and guideline scenario setting. The basis of the new model design is the aggregation and re-interpretation of probabilistically simulated datasets of the general landscape (LULCC LUTs Monte Carlo averages) and selected thematic aspects (accompanying maps). The three main modules (LULCC_basic, LULCC_mplc, LULCC_RAP) are executed consecutively, i.e., each module simulates all time steps (and in LULCC_basic all samples) before the next module is applied. For details regarding the allocations in modules LULCC_basic, LULCC_mplc and LULCC_RAP, see Figs 2 to 4.
Graph: The flowchart presents the LULCC_basic allocation based on the redesigned PLUC model engine's basic functionality (blue). The extended LPB-RAP concept includes a variety of adapted (light grey) or new (dark grey) simulation components. These are new inputs and accompanying maps, allocation rules per guideline scenario and user settings, as well as differentiated succession modeling based on the provided baseline scenario information. The cascading allocation for a restriction policy measure in LULCC_basic is steered within the available landscape extents, LUTs and slope classes (restricted and excluded areas, excluded LUTs and inaccessible terrain cannot be used). Accessible terrain is differentiated into favorable and difficult terrain. Cascading allocation refers to coding where the next level of accessible terrain, i.e., the next pool of available cells, is only used if there is still unsatisfied demand. The cascading allocation for weak conservation is: (
Graph: The diagram displays the essential components of the LULCC_mplc procedures to approximate the discrete anthropogenic impact during a simulation (factual mplc) within scenario assumptions based on prior derived probabilities (Monte Carlo, short: MC, averages) and aggregated results (mathematical mplc) per time step. In LPB-RAP, this module serves as an intermediary step to provide the basis for possible landscape configurations of potential FLR as simulated in LULCC_RAP. This is based on the probable landscape configurations in the factual mplc calculation simulated in LULCC_mplc.
Graph: The diagram shows the RAP allocation procedure within the landscape previously simulated in LULCC_mplc per consecutive time step. The algorithm procedure is additionally steered by the user in the form of the user-defined targeted net forest increment in percent and the defined additional other ecosystem pixels map, besides user-defined and user-selected degradation classes to be simulated with RAP-LUT25. Targeted net forest and other ecosystems input maps derived from the initial simulation year depict the target status which should be restored (e.g., incorporating a 3% increase for net forest extents). To incorporate the potential climate change impact element, the calculation is conducted for each time step according to the current drawn potential natural vegetation map. Note that this approach does not simulate afforestation for non-forest biome pixels.
We implemented several innovations in LPB-RAP, which expanded model code and algorithm functionality (for a comprehensive overview see prelude in S1 File).
For the long-term simulation approach, LPB-RAP features spatially explicit annually interpolated population data and climate-based datasets on a twenty year average. The latter were derived from five climate periods that serve as guiding posts for the long-term simulations until 2100. In its present form, LPB-RAP can simulate various scenario combinations of SSP/RCP scenario data, consideration of restricted areas in different scenario environments, and gradual simulations of terrain parameters and regional systemic simulation choices. The model can be adjusted to a diversity of parametrized initial conditions and scenario variations depending on the scope of the study and region.
For the depiction of land use dynamics, LPB-RAP can, besides the basic PLUC demand/yield approach, utilize a newly developed land footprint per smallholder capita approach, similar to the footprint approach as depicted by Kastner and Nonhebel, 2009 [[
For the depiction of the entire landscape, LPB handles explicit LUTs while maintaining granular high-resolution at the hectare scale instead of aggregated mosaic LUTs, as for example done for the case of PLUC and its considered crop LUTs. Therefore, LPB-RAP comprises n = 23 LUTs, divided into 18 basic LUTs and up to five RAP-specific LUTs (see Table A3 in S1 File for the available basic LUTs and section 2.5 in S2 File for their further descriptions in the context of the applied case study). Due to the generic approach of LPB for global applications, the model was optimized for using modified Copernicus land cover-based maps [[
To enhance the differentiated simulation of forest landscapes the user can further parameterize (forest) succession stages and succession rules (see section 2.2 in S1 File and section 3.1.10 in S2 File for further explanations).
The deterministic simulation part of LPB-RAP builds on spatially explicit datasets of climate data-based and population scenario projections as well as scenario-based long-term time series projections of land use until 2100. These projections provide the underlying long-term guiding posts, or baseline scenario, for the dynamic simulation of landscape and land use change. Datasets can be substituted for completely diverging scenarios or only gradual variations depending on user needs.
We refer to sections 2.7, 2.8 and 3.1.12 in S2 File for further information on the applied SSP2-RCP4.5 scenario assumptions in the case study context of the Esmeraldas province.
Incorporating long-term climate data-based projections is one key innovation in LPB-RAP compared to the base model PLUC. The climate-based determinants serve as the scenario-based guiding posts for simulating future landscape development. They steer and limit the simulation of succession and maximum woody above-ground biomass (AGB) development, as described in the following sections.
Building on climate reference data and projections, potential natural vegetation (PNV) per climate period was computed to derive recent and potential future biome distributions. We rely on ensemble-based stacked model generalizations [[
We estimated the potential maximum undisturbed forest AGB as an approximation to the climax stadium to not overestimate woody AGB dynamics in the forest-related LUTs. For this, we combined the ESA biomass dataset version 3 for 2018 [[
The current model design can accommodate maps of annual increments per selected potential forest type (undisturbed, disturbed, plantation) to enable a spatially explicit approximation of AGB development. The model is conceptualized to use ESA AGB V3 or similar high-resolution data if available. This could currently not be realized for the Esmeraldas region due to quality issues of the increment from 2017 to 2018 within the ESA AGB V3 data (quality flag 3, i.e., "improbable change"). Accordingly, for this study, we solved the simulation of AGB development based on a stochastically derived increment between values reported in the literature (see section 2.2.5 in S1 File and section 3.1.4 in S2 File for further details). For the case of LUT agroforestry, this was already simulated stochastically within the range of the disturbed forest minimum and maximum values because this LUT can potentially depict a broad range of different systems and development stages in space and time during the simulation.
For the long-term simulation within the PLUC demand/yield approach, it is necessary to also provide potential yields per simulated climate period of representative crops and livestock density per agricultural LUT to describe potential land use development. This feature is already implemented in the provided LPB-RAP model and will be subject to a forthcoming publication but is not further described here.
The second component of the underlying baseline scenario is a SSP projection of long-term population development [[
The third pillar of the user-defined baseline scenario refers to the land use component of the presented dynamic footprint approach. For each simulation, land use is depicted by the actively simulated LUTs that represent anthropogenic demands in built-up area (representing streets and building structures in LPB-RAP), agricultural land and demand in woody AGB (see section 1.6 in S1 File for an overview of all components). For long-term simulations and subnational regional modeling, global narratives may have to be adjusted to fit the regional context and current trends (see for the implementation example differences between the global and regional SSP2 narrative in section 3.1.12 in S2 File Table B13 in S2 File). The resulting projected demands for a particular primary active LUT are allocated based on suitability maps within LPB-RAP. The resulting landscape configuration for a time step t is thereby an interplay of methods as depicted in S1 File and the configured time series of demand (see S3 File). Subsequently, land use allocation incorporates inter alia the newly developed dynamic settlement algorithm (see section 2.3.4 in S1 File), the guideline scenario allocation rules (i.e., how restricted areas can be used, see section below) and the terrain inclination levels (see section 2.3.5 in S1 File).
Three forms of potential policy measures and assumptions about underlying pressure aspects can be simulated with LPB-RAP, all of which display plausible what-if scenarios in their own sense. This conceptual scenario level steers the distribution of anthropogenic land use demands by the definition of available areas. Therefore, the location of restricted conservation areas in the case study region is required as model input. Restricted areas are treated differently in the three implemented scenarios.
Scenario weak conservation depicts land use change in restricted areas when land-use demands cannot be satisfied in favorable terrain (i.e., more gentle slopes) or difficult terrain (i.e., steeper slopes) in unrestricted areas. It uses a four-step procedure of a cascading allocation per time step. This scenario assumes that restricted areas are respected by the population as long as unrestricted land is available. The model will record unallocated demands if they occur despite the usage of restricted areas for user information purposes.
Scenario enforced conservation, on the contrary, simulates enforced protection of restricted areas. Land use expansion to and logging in protected areas in this setting is strictly prohibited. Land use is only simulated until land outside restricted areas is available. Any unallocated land demands are registered as an indicator for potential transregional leakage or food security risk.
Scenario no conservation allows land use expansion in restricted areas, thus also potentially amplifying pressure on forests located within such areas. It assumes that such restricted areas are either repealed or ineffective. The scenario is implemented as part of model parametrization, based on a user-defined time step X ≥ initial simulation year of a weak conservation or enforced conservation simulation run. The scenario is realized in a two-step procedure of first considering favorable terrain use and second by the opportunity to expand land use on difficult terrain for the entire case study area. By this design, land use is allocated until the topographic boundary conditions (e.g., for user-defined inaccessible slopes of >45%) for the considered simulation area are reached.
This section introduces the allocation procedures considered in the three consecutively executed modules of LPB-RAP.
LULCC_basic offers the basic functionality of the PLUC model, i.e., the dynamic modeling of LULCC based on suitability factors and anthropogenic demands. It further calculates probabilities based on the embedded PCRaster Monte Carlo framework for all employed basic LUTs and further thematic aspects. Within LPB-RAP, the PLUC base model was substantially expanded to allow for the simulation of all new simulation targets (see Fig 2).
For more detailed information, the reader is referred to prelude information in S1 File
The second module is an extension of the module LULCC_basic. LULCC_mplc aggregates the simulated probabilistic outputs by the highest probability per pixel to derive one discrete landscape configuration for each time step. These landscape configurations can also be considered as a standalone result if the focus is on probabilistic simulations (mathematical mplc). For the simulation of RAP within the module LULCC_RAP, however, it is required to run LULCC_mplc with a corrective allocation (factual mplc within scenario assumptions) based on the a priori derived probability maps per LUT (Monte Carlo averages per primary or secondary active LUTs) to discretely simulate the provided and simulated anthropogenic demands (see Fig 3).
Subsequently, the module runs an extensive analysis on the mplc landscape configuration (depending on the user's choice using the mathematical or the factual mplc), deriving 300+ variables of different thematic categories per time step (including simulated land use area, landscape share, and a variety of pressure and forest condition aspects; for full scope, see mplc log-file in S5 File). Further information is provided in the prelude section of S1 File.
The LULCC_RAP module is the final calculation step in the LPB-RAP modeling framework and serves both as a form of impact assessment and as a scenario of its own: first, the mplc landscape is evaluated and area potentials according to the land use impact are calculated, i.e., the derivation of potential FLR measures in accordance with the simulated land use; second, the module suggests RAP-LUTs which follows an ecological restoration context when considered as a scenario. These RAP-LUTs describe by intent overarching land use categories, as the model has no information to depict the FLR potential in greater detail, such as the description of specific agroforestry or timber plantation systems. LULCC_RAP provides estimates that potentially would result from FLR implementation (65+ variables are derived additionally for the resulting log-file, see RAP log-file in S5 File). To do so, the LULCC_RAP module evaluates the most probable landscape configurations derived from LULCC_mplc for each time step. LULCC_RAP uses the mplc information to identify available areas for restoration purposes after all anthropogenic demands are satisfied. Area availability is simulated by three different approaches with diverging algorithm procedures: RAP maximum, RAP minimum, and RAP suggested additional restricted areas (see for further information below). Caution has to be taken as annual outputs of LULCC_RAP cannot be interpreted as a sequence of implemented restoration measures over a period of t+n but instead represent potential entry points for future FLR projects.
Aside from the suggested conversions of LUTs cropland-annual and pasture to agroforestry, LUT21–LUT24 (RAP agroforestry, RAP plantation, RAP, reforestation, RAP other ecosystems) evaluate pixels that lost their initial ecosystem condition (forest or other ecosystems) while RAP-LUT25 describes the Restoration Area Potential of degraded but not entirely deforested forest pixels. LULCC_RAP allocates the module-specific LUTs and considers pixels which are not simulated under certain land use any longer (RAP other ecosystems, RAP reforestation, RAP plantation) or qualify for adaptation measures (RAP agroforestry, RAP restoration of degraded forest) and thereby become available for restoration purposes (see Fig 4 and section below).
Of importance in this case is the total available area which is defined as suitable for restoration by LULCC_RAP after anthropogenic land use needs are satisfied (LULCC_mplc). RAP-LUT25 (RAP restoration of degraded forest) in contrast evaluates pixels which are still considered as forest by the model. For more detailed information regarding RAP-LUT25 see prelude in S1 File.
The first algorithm in the module LULCC_RAP evaluates the entire modeled study area following a broader perspective of FLR towards Landscape Restoration [[
Graph
Table 1 Rationale of LULLC_RAP-specific LUTs suggested for envisioned landscape transformation in the FLR context (building on Temperton et al., 2019, in the dynamic LULCC landscape modeling framework for a general smallholder-dominated forest landscape).
RAP LUT Envisioned realization Envisioned tree cover outcome Envisioned forest site quality outcome Suggested on climate-period-based biome information and initial targeted net forest Envisioned goals achieved RAP-LUT21 agroforestry rehabilitation efforts by use of fruit or fertilizer trees in combination with cropland or pasture systems (any kind of animal or husbandry) or other agroforestry systems > 0% to approx. 70% depending on per plot targeted agroforestry system to sustain yields (equivalents) minor to medium (envisions some forest site qualities, e.g., for soil parameters) Independent from underlying biome and targeted net forest information as areas remain under land management within scenario assumptions. This LUT is conceptualized as a compromise between mitigation and production goals. RAP-LUT22 plantation reforestation efforts by use of sustainable multifunctional mid- to long-term rotation periods mixed tree species for timber plantations ≥ 70% high (resembles secondary forest naturally developed) Only suggested on forest biome pixels where net forest is not targeted. This LUT is conceptualized as a compromise for economically used plantations which resemble natural potentials under Sustainable Forest Management until harvest to combine the restoration targets of mitigation and production (in time limited potential). RAP-LUT23 reforestation reforestation efforts by passive or active reforestation with the goal of a status of almost all recovered primary forest traits ≥ 70% full (resembles almost primary forest) Only suggested on forest biome pixels where net forest is targeted (resembling forest characteristics at terrestrial surface level at timestep t+n). This LUT envisions the combined restoration goals of forest (habitat) re-establishment and mitigation, hence, simulated on sites connected to the initial net forest extent or located within targeted net forest. RAP-LUT24 other ecosystems Landscape Restoration efforts, which denote, for example, the potential need for the recreation of riverbanks and other small ecosystems for Landscape Restoration, including full functionality and biodiversity. The LUT incorporates target pixels up to succession or net forest deforested, as the conversion for other ecosystem pixels cannot be distinguished. This application covers, for example, the case that a pixel of herbaceous wetland is converted to agricultural land use (e.g., by drainage), was abandoned and is in succession to undisturbed forest. The initial landscape mosaic would demand herbaceous wetland, though. Corresponding pixels and associated information exceed biome and net forest target information due to microclimate, plot conditions and habitat et cetera (e.g., also for user-defined cultivated landscape ecosystem types) and represent the initial and potentially later still given landscape mosaic within the general forest landscape. This LUT envisions the necessary re-establishment of other ecosystems within the general forest landscape. RAP-LUT25 restoration of degraded forest enrichment planting in degraded forest plots towards growth for potential maximum AGB ≥ 70% high (in regard to AGB, under use again due to proximity) Suggested on all user-defined degradation stages, considered as impacted by wood AGB extraction, on simulated disturbed forest pixels This LUT envisions the recovery of forest site conditions and thereby the achievement of mitigation goals.
LPB-RAP calculates the RAP minimum potential of mitigation measures when FLR measures shall be initiated neither landscape-wide nor short-term. To achieve this, LULCC_RAP firstly evaluates the projections from the population data to identify the year of the population peak when considering an internally calculated static footprint approach. On the contrary, it considers the year of summarized peak demands when using the dynamic footprint approach using external time series as presented here. Secondly, LULCC-RAP draws on the mplc landscape for the same year and evaluates all areas under anthropogenic use to derive the population peak or the peak demands land use in case of the external footprint approach. This information is used in a third step to eliminate all overlapping RAP areas per time step t until the respective year is reached. As a result, only those areas remain which are not projected to satisfy the maximum anthropogenic demand in the simulated future setting, i.e., pixels within the LUTs of RAP plantation, reforestation, other ecosystems and p.r.n. degraded forest.
For the sake of policy development, LULCC_RAP simulates a spatially explicit and user-defined single output of potential areas p.r.n. suited for a short-term rededication as additional restricted areas. This output may serve as a potential counterpart of conservation compared to restoration under the consideration of population peak demands or summarized peak demands. In this case, the model user can define a list of LUTs, which will be evaluated for the peak year, where the anthropogenically used areas are subtracted before. This approach allows for an approximation of the maximum demand and the available maximum area suited for restoration and conservation until the population or demand stabilizes or declines. The computed information provides an estimate of long-term regional additional potentials for conservation areas (and partly restoration if parameterized), which ideally will be established based on further plot-specific investigations and stakeholder elucidations. Therefore, this output refers to a top-down measure and is only produced in a weak or enforced conservation scenario setting.
The results section showcases results of the LPB-RAP model as derived for the implementation case study Esmeraldas province of Ecuador (see S2 File for further information on background and parametrization). The basic message of the LPB-RAP simulation approach is the estimate of impact and impact mitigation options of land use patterns in the investigated smallholder-dominated forest landscape in the long term. The model, therefore, provides the user with the scenario information of the landscape without and with potential FLR per simulated time step. Further information on potential long-term annual development are available as visual outputs (e.g., maps and gif) and numeric outputs (e.g., time series) for a broad range of variables. These variables, which include for example pressure aspects, or potential biomass and carbon sequestration, are presented in detail for the case study Esmeraldas in S4 File (selected) and S5 File (full spectrum). This information can be, for example, also used for short- to mid-term measures in regional landscape planning or policy development. In the following sections, we present major modeling results of the simulation that can support decision-making processes; these are (
The resulting information on simulation probability is an essential element of this model approach. It can be used for landscape planning and policy development because it provides information on implied uncertainty of the future LULCC configurations. As shown in Fig 5, depending on the varying trajectories of the Monte Carlo samples, the overall class "100% probability" declines gradually over time. As patterns of land use expansion slightly vary between samples, uncertainty increases partially within the simulation. The stochastic simulation of forest age also leads to increased uncertainty in later simulation stages (change of LUT08 = disturbed forest to LUT09 = undisturbed forest). Overall trends are similar when comparing all three guideline scenarios.
The given probability and uncertainty information refers to the technical simulation within the applied scenario information only, i.e., the probability is based on the number of occasions a pixel is simulated with the same LUT over all samples per time step. Therefore, the model simulates the highest probability for steady land cover or land use, i.e., mainly for LUTs such as water and built-up as well as land use near settlements and undisturbed forest areas not impacted by anthropogenic demands. Uncertainty mainly occurs (
Graph: The diagram depicts the classified probabilities in LULCC_mplc, here following the corrective allocation to the factual mplc, i.e., for particular pixels, smaller probabilities than derived from the mathematical mplc can be in place. The class 0% does not occur; it is only implemented to track potential simulation errors. Otherwise, the distribution of probability classes is related to the chosen number of samples; here, the pseudo-random sampling approach was chosen, resulting in 49 samples for the Esmeraldas region. Further uncertainty elements are the stochastically based simulations of plantations and succession besides the development of previously unused areas. For an example visualization of the spatial distribution of uncertainty for a selected probing date, see Fig 6; for all probing dates of the three guideline scenarios marked on the x-axis see section 1.1.2 in S4 File; for discrete numeric values in annual resolution see S5 File.
Graph: Spatial distribution of landscape modeling probabilities in 2100 (long-term) in the SSP2-4.5 based policy scenario no conservation. Visible are the differentiated uncertainty classes and their spatial distributions in accordance with simulation rulesets in interplay with dynamic modeling, e.g., for the no conservation scenario land use development in restricted areas (see section 2.3 in S2 File) is depicted and increased uncertainty at the land use development frontiers. For aspects of to be considered model internal and external plausibility please refer to the discussion in section 4.3.
The visualization of the spatial distribution of landscape modeling probabilities supports the model user in the interpretation of simulated landscape configurations. The provided formats are, besides PCRaster map outputs, GIF files (for the entire simulation time frame) and R-based maps (probing dates, see S4 File Background information). Uncertainty increases per simulated time step as expected. Nonetheless, the simulation rulesets, for example, whether restricted areas are used for land use expansion purposes or not, are clearly visible in the emerging patterns. These datasets are provided for each simulated year and defined probing dates for the entire study landscape. They can be subject to further inspections and GIS operations to identify smaller areas of interest for a particular year, for example, for identifying restoration or conservation zones.
The following sections present the results of selected comparative model scenario outcomes for demonstration purposes, i.e., the concluding juxtaposition of simulated scenarios "no FLR" (mplc) and "potential FLR" (RAP) landscape configurations for the baseline scenario SSP2-4.5 setting.
For further selected results depicting all probing dates and for all three simulated policy scenarios the reader is referred to S4 File, which illustrates results for mplc (section 1) and RAP (section 2), such as SSP-related simulated urbanization, derived probable pressure aspects (e.g., indicated pressure on the population by use of difficult terrain and unallocated demands; pressure on restricted areas; pressure on forests by deforestation, conversion or degradation; pressure on habitats by remaining forest quality regionally and in restricted areas), possible mitigation measures (RAP minimum and RAP suggested additional restricted areas). For all covered landscape components, we refer to S5 File.
As expected, within the applied scenario assumptions (same demands and largely unchanged parametrization) all three policy guideline scenarios show very similar emerging land use patterns during the course of simulation (see section 1.3 in S4 File and S5 File).
For the Esmeraldas region, scenarios weak conservation and enforced conservation perform de facto identical under consideration of the same parametrization to approximate the starting conditions (see S2 File). This occurs due to the decreasing number of oil palm plantations, a decreasing wood demand and the large share of parameterized unprotected (forest) areas, which in the case of the Esmeraldas province largely denote suitable areas for land use expansions. In case of Esmeraldas province, all projected SSP2-based demands can be allocated throughout the simulation time frame 2018/2025–2100 within the parametrized scenario assumptions. The landscape development is firstly offset by decreasing oil palm plantations area, which allows for the expansion of agricultural LUTs until the population peak (2060) despite an increase in kilocalorie intake per person. Afterwards, the continuously simulated societal shift towards a higher animal-based diet causes the expansion of pastures increasing forest conversion towards 2100, which here coincides with the year of peak demands (due to scenario assumptions in conjunction with data limitations). In the no conservation scenario, land use expands to the formerly restricted areas, increasing the area of land use in (formerly) restricted areas from 12% in 2025 to 23% in 2100. This is in contrast to both conservation scenarios, which did not reveal such patterns.
The juxtaposition of LPB mplc and RAP as the last conceptual scenario stage is tied to an FLR context. The difference can best be visualized for the potential forested area and spatial landscape configurations for a probing date per scenario mplc (no FLR) and, within this scenario, the RAP scenario (potential FLR) as a counter-proposal.
The main model result is the detection of forested areas that have a likelihood to remain despite anthropogenic use (mplc) and possible suitable area potentials to reinstate forested areas (RAP) with either more forest cover (LUT21 = RAP agroforestry and LUT22 = RAP plantation) or even forest site characteristics (LUT23 = RAP reforestation and LUT25 = restoration of degraded forest) per time step following the implemented scenario assumptions (see Fig 7).
Graph: Potential forested area per time step during simulation based on the mplc landscape configuration and RAP interpretation of forest at terrestrial surface level (LUT08, LUT09 and LUT23), of tree cover (LUT04 and LUT21), as an in time-limited potential (LUT22), or as existing disturbed forest area that can be improved in quality after anthropogenic wood extraction (LUT25). Note that the y-axes differ. Scenario weak conservation is not explicitly depicted here as its simulation, under the case study area conditions in interplay with scenario assumptions, is identical to the enforced conservation scenario.
Mplc describes the potential of remaining forested areas in the user-defined scenario setting. It follows here the SSP2-4.5 baseline scenario rationale by simulating population and smallholder demands under changing climate conditions. RAP denotes this aspect per time step by additionally considering possible areas of tree to forest cover and p.r.n. locations resembling forest site characteristics. Agroforestry (LUT04 as well as LUT21) in this context depicts a spectrum. Moreover, the initially simulated peaks in enforced conservation for LUT22 = RAP plantation are mostly based on larger initial areas of herbaceous vegetation and shrubs which are located on potential forest biome pixels outside the targeted net forest distribution. LUT25 (RAP restoration of degraded forest) is also covered in LUT08 (disturbed forest) and in this context implies the potential of forest pixels to fully regenerate under supported management action (for a visualization of the spatial configurations of the selected probing dates see section 2.2 in S4 File or S5 File). Note that the counterintuitive larger share of RAP-LUT25 in the no conservation scenario results from the different spatial use of the landscape due to the assumed repealed or ineffective restricted areas. Due to the thereby possible different spatial allocation of the same demands other forest pixels are affected. This leads to different remaining forest pixels after land use allocation with diverging AGB contents. The higher share of LUT25 thereby documents, that (
Conceptually, each time step depicts a potential entry point for FLR measures, which could result in larger areas of potential forested areas. Within the parameterized scenario, LPB-RAP delivers for each time step the theoretical maximum potential under the assumption that landscape-wide FLR has not been implemented yet.
To showcase the new model functionality, we present model outputs for the simulation year 2050 based on LULCC_mplc (see Fig 8) and Restoration Areas Potentials computed in LULCC_RAP (see Fig 9), exemplified for the enforced conservation scenario.
Graph: Mplc "no FLR" spatial distributions of basic LUTs in 2050 (mid-term) for SSP2-4.5 policy scenario enforced conservation. For further probing dates and scenarios see section 1.3 in S4 File. Note that by steadily increasing demands no abandoned LUTs are simulated and only in t = 1 already existing land use in restricted areas occurs.
Graph: Spatial distributions of RAP-LUTs in 2050 (mid-term) for SSP2-4.5 policy scenario enforced conservation (based on mplc landscape configuration depicted in Fig 8). To highlight the potential distribution of RAP the remaining regional landscape area is here only depicted as a hill shade. Note that LUT21 dominates the case study landscape due to the demand in food production. LUTs 22 to 25 also occur but only in very minor quantities and mostly scattered in the landscape. Note also that RAP only describes area potentials for severely impacted pixels that require active measures while conservation potentials are not depicted. For further probing dates and scenarios see section 2.2 in S4 File.
RAP simulation in this scenario case showcases that due to agricultural demands for food production interests in non-food RAP-LUTs can be only marginally fulfilled.
Within the combined scenario assumptions, there still remain areas that can be potentially protected in the long term as they are not simulated under use for the peak demands year. This coincides here with the final simulation year of 2100 (see Fig 10).
Graph: Potential additional restricted areas (petrol) compared to existing restricted areas (yellow) within the simulated regional extent (purple). These are area extents that remain available despite land use of the peak demands year (here 2100, enforced conservation scenario).
This is one example of how LPB-RAP can be used to support top-down landscape planning and policy development while building on information from long-term bottom-up scenario simulations. In the case of Esmeraldas province for example, existing restricted areas in the simulated extent comprise 414,987 ha or 24.72% of the landscape. Additionally, 261,660 ha could be protected without risking smallholder demands. This would amount to a total area of 676,647 ha or 40.31% of the entire simulated study landscape (Esmeraldas province plus buffer). These scenario-based suggested areas comprise a range of to be individually weighted cases of different levels of protection and management choices and should be further investigated in situ.
This section discusses the novel LPB-RAP modeling approach with a particular emphasis on the RAP modeling components. This is followed by a discussion on the model's internal and external plausibility as well as current caveats and limitations for the presented model design.
To our knowledge, this is the first time a sub-national region such as a province was modeled dynamically and long-term following a SSP2-4.5 narrative in a fine granular 100 m resolution with a particular focus on smallholder land use patterns and to derive RAP estimates. The LPB-RAP modeling approach in its present form approximates RAP of formerly converted, deforested or degraded areas in a forest landscape context. The model does not focus on recent global potential hotspots of restoration [[
For the application of future, especially long-term, scenario-based RAP estimates, we did not encounter a model in the literature to date that (
Additionally, LPB-RAP provides information on pixel-level land use conditions for each consecutive year when aiming to enter into landscape-wide FLR measures. Thereby, LPB-RAP results have the potential to support relevant stakeholders for a long-term planning horizon, especially based on different scenario combinations of SSPs/RCPs, area restrictions and land use.
For the general application of LPB-RAP in the presented form and the depiction of SSP2 assumptions, model results overall indicate the need to break "glocal" (i.e., local to global) patterns near time towards SSP1 measures; this may entail explicit measures for climate protection such as (
We can build for internal model plausibility on two aspects: firstly, the considered base LULCC modeling framework embedding features of the PLUC model thereby relying on the same principle of suitability maps and bottom-up modeling. Secondly, on the assessment of endogenous parameters such as the computed suitability maps, based here on a detailed representation of settlements and corresponding LULCC in the smallholder simulation context. Here we rely on empirical data of settlement points and parameters derived from in situ investigations of land use demands and walking distances. This altogether depicts plausible approximations of the case study context for the initial simulation year as presented here.
Regarding external plausibility, the main aspect of caution is the scenario character of the model as it operates not on the perpetuation of historic trends as common for models such as CLUE but on a set of current empirical data based on a
For this study, we chose the SSP2-4.5 scenario narrative "Middle of the Road". In comparison, the overall SSP1 scenario "Taking the Green Road" would, on the contrary, indicate a more sustainable global to local pathway with a fundamentally more positive impact on climate and population development and land management. However, given the current debates on the impact of climate and global environmental change [[
The question if perpetuated trends of historic and recent developments are plausible for a chosen case study is another subject and should be answered externally if desired [[
The following section illuminates, for such a model external plausibility assessment, backgrounds of the four major modeling components to establish the "most probable landscape configuration" on trends for Ecuador and the simulated case study province of Esmeraldas. It further discusses how well they fit to implemented methods and what-if-scenario assumptions for long-term simulation:
(
(
(
For the overall historical trends and recent developments perpetuation, it should be noted that Ecuador is assessed as being on the brink of the transition to an emerging nation [[
While section 3.1. of the results covered the internal simulation uncertainty implied by the model internal recording of landscape modeling probabilities, this section focuses mainly on external uncertainty for the developed LPB-RAP model. Three aspects should be pointed out in detail for the overall modeling approach and its limitations as presented in this study.
Firstly, limitations of depictions of land use systems must be addressed. Currently, the general model is capable of smallholder land use simulations, which depends on input data for parametrization and as such acknowledges commercial demands to a certain degree only. However, the explicit dynamic simulation of commercial demands with its own parametrization, in addition to the smallholder land use simulation, is not yet part of the LPB model due to lacking primary or secondary regional data, and points to a future task for regional landscape studies as well as the further LPB-RAP model development. For example, in the case of timber demands, according to recent studies, smallholder subsistence demand in the Esmeraldas is rather marginal [[
Secondly, the aspect of validating the model's spatial simulation representation should be clarified [[
Thirdly, the aspect of forest degradation and deforestation must be considered in the forest landscape simulation [[
However, all of the discussed issues may apply to other land use models too or are not yet acknowledged at all. Likewise, other caveats and limitations representing external uncertainty are mainly based on model parametrization and calibration data of the chosen local to global data products. This may comprise (
The presented LPB-RAP modeling approach presents a valuable new tool for supporting local to regional landscape planning and policy development focusing on smallholder-dominated forest landscapes at the level of a sub-national administrative unit such as a province. In its current design, LPB-RAP raises awareness of the consequences of projected land use development in a changing landscape framework and offers compromise solution suggestions. For future applications, we expect that model performance in regard to the approximation of spatial patterns will further improve with the growing availability of more detailed secondary and current primary data sources, p.r.n. also by the extension of incorporation of regional commercial demands. Further analysis steps are possible to extend the range of potential research questions, e.g., deforestation hotspot analysis or the simulation of different baseline scenarios and subsequent analysis of overlapping areas for restoration.
LPB-RAP model stage 2 aims to extend the simulation of potential future multifunctional landscapes in the Anthropocene with ecosystem fragmentation analysis and the modeling of potential habitat corridors for a user-defined regional umbrella species. Further model development stages may include the incorporation of a) economic aspects in the form of opportunity costs, as well as b) incorporation of potential sustainable forest management.
The model in its present form demonstrated its usefulness of simulating long-term dynamic land use change patterns in combination with potential climate change, population dynamics and developing land use demands. Overall, LPB-RAP proves to be a suitable scenario tool for long-term projections and RAP investigations. It can be characterized as a top-down planning tool based on bottom-up modeling, which requires local follow-up investigations, stakeholder participation across scales and local expertise for the implementation and continued realization of restoration.
Having depicted the two dichotomous different plausible destinations of a continuance of the status quo (mplc) or the entry into landscape-wide FLR (RAP) as a counter proposal on each time step landscape for the simulated region, and likely others like it, we ask all involved stakeholders again: Quo vadis, smallholder forest landscape?
S1 File
LPB-RAP new generic and user-defined methods.
(PDF)
S2 File
Case study modeling: parametrization and calibration.
(PDF)
S3 File
Case study modeling: parametrization of demand time series and calibration to scenario assumptions.
(XLSX)
S4 File
Case study modeling: Selected results.
(PDF)
S5 File
GitHub repository containing open-source model, user-manual and required data.
Additionally, all secondary result outputs for this study are deposited.(PDF)
We are grateful for the invaluable input of colleagues of the LaForeT-R
By Sonja Holler; Daniel Kübler; Olaf Conrad; Oliver Schmitz; Carmelo Bonannella; Tomislav Hengl; Jürgen Böhner; Sven Günter and Melvin Lippe
Reported by Author; Author; Author; Author; Author; Author; Author; Author; Author