Research article Special Issues

Global dynamics for a multi-group alcoholism model with public health education and alcoholism age

  • A new multi-group alcoholism model with public health education and alcoholism age is considered. The basic reproduction number R0 is defined and mathematical analyses show that dynamics of model are determined by the basic reproduction number. The alcohol-free equilibrium P0 of the model is globally asymptotically stable if R01 while the alcohol-present equilibrium P of the model exists uniquely and is globally asymptotically stable if R0>1. The Lyapunov functionals for the globally asymptotically stable of the multi-group model are constructed by using the theory of non-negative matrices and a graph-theoretic approach. Meanwhile, the combined effects of the public health education and the alcoholism age on alcoholism dynamics are displayed. Our main results show that strengthening public health education and decreasing the age of the alcoholism are very helpful for the control of alcoholism.

    Citation: Shuang-Hong Ma, Hai-Feng Huo. Global dynamics for a multi-group alcoholism model with public health education and alcoholism age[J]. Mathematical Biosciences and Engineering, 2019, 16(3): 1683-1708. doi: 10.3934/mbe.2019080

    Related Papers:

    [1] Shaoyung Chen, Hsinyi Mao, Pinhua Tu, Naichen Cheng, Jiashing Yu . Fabrication of decellularized adipose tissue/alginate composite microspheres with pASCs encapsulation for tissue engineering. AIMS Bioengineering, 2017, 4(3): 351-365. doi: 10.3934/bioeng.2017.3.351
    [2] Mark Weingarten, Nathan Akhavan, Joshua Hanau, Yakov Peter . A Place to Call Home: Bioengineering Pluripotential Stem Cell Cultures. AIMS Bioengineering, 2015, 2(2): 15-28. doi: 10.3934/bioeng.2015.2.15
    [3] Velislava N Lyubenova, Maya N Ignatova . On-line estimation of physiological states for monitoring and control of bioprocesses. AIMS Bioengineering, 2017, 4(1): 93-112. doi: 10.3934/bioeng.2017.1.93
    [4] Matthew Hogan, Glauco Souza, Ravi Birla . Assembly of a functional 3D primary cardiac construct using magnetic levitation. AIMS Bioengineering, 2016, 3(3): 277-288. doi: 10.3934/bioeng.2016.3.277
    [5] Stephen W. Sawyer, Kairui Zhang, Jason A. Horton, Pranav Soman . Perfusion-based co-culture model system for bone tissue engineering. AIMS Bioengineering, 2020, 7(2): 91-105. doi: 10.3934/bioeng.2020009
    [6] Daniele D’Arrigo, Marta Bottagisio, Silvia Lopa, Matteo Moretti, Arianna B. Lovati . Tissue engineering approaches to develop decellularized tendon matrices functionalized with progenitor cells cultured under undifferentiated and tenogenic conditions. AIMS Bioengineering, 2017, 4(4): 431-445. doi: 10.3934/bioeng.2017.4.431
    [7] Joseph M. Chambers, Robert A. McKee, Bridgette E. Drummond, Rebecca A. Wingert . Evolving technology: creating kidney organoids from stem cells. AIMS Bioengineering, 2016, 3(3): 305-318. doi: 10.3934/bioeng.2016.3.305
    [8] Satoshi Migita, Kunitaka Araki . Effect of nanometer scale surface roughness of titanium for osteoblast function. AIMS Bioengineering, 2017, 4(1): 162-170. doi: 10.3934/bioeng.2017.1.162
    [9] Marcello Rubessa, Andrea Ambrosi, Dianelys Gonzalez-Pena, Kathryn M. Polkoff, Scott E. Denmark, Matthew B. Wheeler . Non-invasive analysis of bovine embryo metabolites during in vitro embryo culture using nuclear magnetic resonance. AIMS Bioengineering, 2016, 3(4): 538-551. doi: 10.3934/bioeng.2016.4.538
    [10] Olga M. Tsivileva, Inna M. Uchaeva, Nikolay A. Yurasov . Biotesting of technologically important carboxy containing acridones with solid-state fungal culture. AIMS Bioengineering, 2021, 8(1): 1-13. doi: 10.3934/bioeng.2021001
  • A new multi-group alcoholism model with public health education and alcoholism age is considered. The basic reproduction number R0 is defined and mathematical analyses show that dynamics of model are determined by the basic reproduction number. The alcohol-free equilibrium P0 of the model is globally asymptotically stable if R01 while the alcohol-present equilibrium P of the model exists uniquely and is globally asymptotically stable if R0>1. The Lyapunov functionals for the globally asymptotically stable of the multi-group model are constructed by using the theory of non-negative matrices and a graph-theoretic approach. Meanwhile, the combined effects of the public health education and the alcoholism age on alcoholism dynamics are displayed. Our main results show that strengthening public health education and decreasing the age of the alcoholism are very helpful for the control of alcoholism.


    1. Introduction

    Global land-use/land-cover (LULC) change projections generated by the integrated assessment and climate modeling communities are used for a wide variety of applications including modeling carbon dynamics, water use, and land-change [1,2,3]. While increasingly sophisticated models are becoming available, applying global projections of LULC change at regional scales remains a challenge requiring further research [4,5]. Typically, global LULC models produce LULC projections at broad regional scales or at coarse grid resolutions [4,6]. However, modeling at local to regional scales is often of greater relevance for informing management decisions and mitigation strategies. For LULC change in particular, local factors have important and direct influence on land-use patterns necessitating modeling at finer scales where local data is available [7].

    The climate modeling community has been moving towards a scenario framework in an effort to explore implications of alternative climate futures and mitigation outcomes [8]. Applying global-scale scenario data to regional-scale scenario modeling of LULC change has two important benefits: (1) it allows for the incorporation of global climate scenarios into regional-scale research, and (2) facilitates relevant comparisons between LULC modeling efforts [9]. Using global LULC data in regional assessments requires data downscaling to scales relevant for analysis. Downscaling refers to the process of translating data from a coarse scale to a more detailed scale while maintaining a degree of consistency between datasets [10,11].

    Previous approaches to downscaling LULC projections have relied largely on finer scale classified LULC products derived from remote sensing sources to facilitate the downscaling of LULC data. Dendoncker et al. [12], used the Coordinated Information on the European Environment (CORINE) land cover mapas a starting point for projecting land-use at fine spatial scales. West et al. [13], used Moderate Resolution Imaging Spectroradiometer (MODIS) land-cover data to facilitate land-use downscaling from a 0.5 degree grid to a 0.05 degree grid for the United States. Other approaches have used finer scale models to downscale global land-cover data [14,15]. Sleeter et al. [16] and Sohl et al. [17] used an integrated assessment model combined with land-use histories and expert knowledge to drive downscaling.

    Representative concentration pathways (RCPs) are the latest set of climate scenarios used for the Intergovernmental Panel on Climate Change’s (IPCC) Fifth Assessment Report (AR5) [18]. The RCPs are a set of trajectories of land use, air pollutants, and greenhouse gas levels leading to total radiative forcing targets of 2.6, 4.5, 6.0, and 8.5 in W/m2 in the year 2100 [9]. Each RCP is named after the radiative forcing target value (e.g. RCP 2.6, etc. ). Global RCP land-use transition projections (i.e. LULC change amounts) are available annually between 2005 and 2100 at 0.5 × 0.5 degree grid cells [9]. While LULC transition values at this resolution may be suitable for global modeling efforts, local to regional scale land-use modeling often requires RCP LULC transition values summarized at the regional scale to maintain data consistency.

    Regional scale RCP LULC transition values can be created by simply aggregating RCP grid cell LULC transition values to a regional boundary of interest. However, a strategy is needed for dealing with coarse grid cells overlapping into two or more regions. For these cells, LULC transition values must first be downscaled to region boundaries before they can be summarized by region of interest. Since RCP cells are quite large (~ 3000 km2) the method used to downscale RCP boundary cells to regions may have a large impact on the resulting LULC projections.

    We present three approaches for downscaling RCP land-use projections to Level III Ecoregions [19] in the Pacific Northwest, USA: (1) using area proportions (“area-based”), (2) using proportions of LULC from a classified remote sensing dataset (“composition-based”) and (3) using a historic LULC transition dataset (“transition-based”). Ecoregions are useful for regional scale analysis as they represent areas of similar biotic, abiotic, physical and aquatic characteristics [19,20] and have been proven useful in examining historical [3] and potential future land use change [21]. For RCP 6.0 [22] we used LULC projections from the RCP database [4] to derive ecoregion-scale projections of LULC change for each of the three downscaling approaches. We then used a state-and-transition simulation model (STSM) to project changes in LULC between 2005 and 2100 based on each downscaling outcome. Below we describe the three downscaling methods and compare the results of each simulation. Our goal was to assess the most effective downscaling technique for regional applications of RCP data.

    2. Methods

    2.1. Study area

    Seven Level III Ecoregions in the Pacific Northwest, USA were analyzed in this study including the Puget Lowland, Coast Range, Eastern Cascades Slopes and Foothills (hereafter East Cascades), Willamette Valley, Klamath Mountains, Cascades, and North Cascades. The total study area is comprised of approximately 266, 734 km2 (Figure 1). The LULC mosaic in the Pacific Northwest is characterized by the forested mountainous terrain of the Cascades and Coast Range, agriculture and urban dominated land-use in Willamette Valley and Puget Lowland and grass/shrub dominated areas of the East Cascades. Historical LULC change rates within the Pacific Northwest are high with the greatest change rates found in the Puget Lowland, Coast Range and Willamette Valley ecoregions [23]. Logging, urbanization, and changes in agriculture are all important regional land change processes.

    Figure 1. A) Study area with ecoregions and RCP cells (in orange). B) Study site in the context of the USA.



    2.2. Methods overview

    Global LULC projections from 2005 to 2100 for the RCP 6.0 scenario were available from the RCP database hosted by the International Institute for Applied Systems Analysis (IIASA) [24]. RCP 6.0 is a medium to high emission pathway scenario where radiative forcing stabilized at 6.0 W/m2 in the year 2100. Mitigation efforts in this scenario were implemented around 2060 in an effort to reduce global greenhouse gas emissions, resulting in a decrease in fossil-based fuels and an increase in renewable energy sources [22]. Globally, land use change was characterized by increased area devoted to agriculture and development; however regional variation occurred in land-use patterns. A complete description of RCP 6.0 can be found in Masui et al. [22]. A total of 21 RCP land-use transitions were considered for downscaling (Table 1). RCP projections provide LULC transitions as a percentage of total cell area within a 0.5 × 0.5 degree cell.

    Table 1. RCP transitions considered for downscaling and the average annual amount of transition over the study region between 2005 and 2100
    RCP “from” Class RCP “to” Class Average Annual Transition (km2)
    Crop Pasture 57.9
    Crop Primary land 0.0
    Crop Secondary land 25.4
    Crop Urban 30.7
    Pasture Crop 144.3
    Pasture Primary land 0.0
    Pasture Secondary land 551.4
    Pasture Urban 62.3
    Primary land Pasture 348.7
    Primary land Crop 8.1
    Primary land Secondary land 1278.1
    Primary land Urban 0.0
    Secondary land Pasture 158.8
    Secondary land Crop 4.6
    Secondary land Urban 135.2
    Urban Pasture 0.0
    Urban Crop 0.0
    Urban Secondary land 0.0
    Mature secondary forested land Harvested biomass 21871.8
    Primary forested land Harvested biomass 0.0
    Young secondary forested land Harvested biomass 0.0
     | Show Table
    DownLoad: CSV

    RCP grid cells within or partially overlapping the study area were selected for use in our downscaling methods. Large bodies of water within the study site were extracted and consolidated as a separate region so that water areas could be excluded when applying area-based reapportioning. Ecoregion and RCP cell layers were merged so that sub-cell zones were created at ecoregion boundaries. For the purpose of explaining our three downscaling approaches, we define the following terms in relation to Figure 2. Cells are 0.5 × 0.5 degree RCP cells (Figure 2A), regions refer to ecoregion boundaries (Figure 2B), and zones are sub-cell areas created by the intersection of a cell and region (Figure 2C‒D).

    Figure 2. A representation of A) an RCP 0.5 × 0.5 degree cell grid, B) two adjacent regions, C) a cell grid overlaid and merged with regions, and D) an RCP cell with sub-cell zones including a water zone.

    Regional level transition values were created by aggregating cell transition values to regional boundaries. However, for RCP cells overlapping more than one ecoregion, RCP cell LULC transitions must be downscaled to zones before summarizing at the ecoregion level. Three methods: area-based, composition-based, and transition-based downscaling were tested. In order to use RCP transition values in our STSM, broad RCP classes were cross-walked into LULC classes consistent with remote sensing imagery and historical data. For the area and composition-based reapportioning methods, this process was done following aggregation of RCP transitions to the ecoregion level. For the transition-based method, cross-walking was required prior to RCP transition reapportioning, so that transition classes were consistent with the historical Land Cover Trends dataset [25] used in the downscaling process. Table 2 provides an overview of our three downscaling methods and crosswalking approaches.

    Table 2. Overview of methods for the area-based, composition-based, and transition-based downscaling approaches
    Area-based Composition-based Transition-based
    1. Areas for zones covered by water proportionally redistributed to land-based zones within each RCP cell.
    2. RCP transitions apportioned to zones based on each zone’s proportion of area, relative to its cell.
    3. Zone transitions summed across regions.
    4. RCP transition classes
    cross-walked to STSM transition classes.
    1. Landcover composition derived for each cell and zone from a classified remote sensing product.
    2. RCP transitions apportioned to zones based on each zone’s landcover area of the "from" class of the transition, relative to its cell.
    3. Zone transitions summed across regions.
    4. RCP transition classes
    cross-walked to STSM transition classes.
    1. RCP transition classes
    cross-walked to STSM transition classes.
    2. Region level transition data from the Trends dataset distributed to zones based on zone area relative to the region.
    3. RCP transitions apportioned to zones based the proportion of Trends transition occurring within each zone relative to its cell.
    4. Zone transitions summed across regions.
     | Show Table
    DownLoad: CSV

    2.3. Area-based downscaling

    For the area-based downscaling approach, cell transition values were downscaled based on the proportional area of each zone within a cell. Since LULC transitions cannot occur where water bodies are present, the relative proportion of land area (not total area) of each zone within a cell was used to determine RCP downscaling. RCP cell transitions were downscaled to zones as follows:

    Ai,j,z(t)=Pi,j,c(t)Ac(AzWz)(AcWc)
    (1)

    Where Ai,j,z(t) represents the area transitioning from class i to class j for zone z in timestep t. Pi,j,c(t) represents the RCP proportion of area transitioning from class i to j for cell c in timestep t. Variables AC, AZ, WC, and WZ represent the area of cell c, the area of zone z, the water area of cell c, and the water area of zone z respectively. Thus, Pi,j,c(t) AC represents the total area of transition in a cell, and @\frac{{\left( {{A_z} - {W_z}} \right)}}{{\left( {{A_c} - {W_c}} \right)}}@ represents the ratio of zone land area to cell land area. Transitions downscaled to zones were then aggregated to regions.

    2.4. Composition-based downscaling

    Our composition-based downscaling approach assumes the relative presence of a particular LULC type determines the share of corresponding RCP transition apportioned between ecoregions. This method relied on a classified LULC map to facilitate the downscaling of RCP transitions to zones. The baseline LULC composition map must have a sufficiently high resolution to disaggregate RCP cells, offer a classification scheme capturing all of the RCP classes, and provide good spatial and temporal correspondence with the study region. The National Land Cover Database (NLCD) 2006 dataset with a resolution of 30 m and 16-class classification scheme met these requirements (Table 3) [26].

    Table 3. NLCD 2006 classes and percent of study region area
    NLCD Class Percent of Study Area (2006)
    Open water 1.32
    Perennial ice/snow 0.22
    Developed, open space 2.76
    Developed, low intensity 1.55
    Developed, medium intensity 0.69
    Developed high intensity 0.25
    Barren land (rock/sand/clay) 1.68
    Deciduous forest 1.54
    Evergreen forest 53.36
    Mixed forest 4.56
    Shrub/scrub 18.52
    Grassland/herbaceous 6.50
    Pasture/hay 3.37
    Cultivated crops 1.93
    Woody wetlands 1.02
    Emergent herbaceous wetlands 0.72
     | Show Table
    DownLoad: CSV

    Land-cover composition was derived for each cell and respective zone from the NLCD map. NLCD landcover classes were generalized to RCP LULC classes. RCP cell transition values were then downscaled to zones based on the NLCD area of the RCP transition “from” class. For example, the RCP transition “crop to urban” was reapportioned based on the area of the NLCD class “crop”. Each transition is apportioned as follows:

    Ai,j,z(t)=Pi,j,c(t)AcCi,zzcCi,z
    (2)

    where Ai,j,z(t) represents the area transitioning from class i to class j for zone z in timestep t. Pi,j,c(t) represents the RCP proportion of area transitioning from class i to j for cell c in timestep t. Ac is the area of the cell and is the NLCD area of the LULC class i in zone z. Transitions downscaled to zones were then aggregated to regions.

    2.5. Transition-based downscaling

    The transition-based downscaling approach utilized the USGS Land Cover Trends (hereafter referred to as “Trends”) dataset [25] of ecoregion-level LULC transitions to disaggregate RCP cells to zones. Trends data include LULC transition values for transitions between 11 LULC classes for four temporal periods between the years 1973 and 2000, for each Level III Ecoregion in the conterminous United States. The dataset was created using a statistical sampling method and manual interpretation of Landsat imagery [3,27]. RCP land-use transitions were crosswalked to transition classes used in our STSM as described in section 2.6. Average annual transition values for 9 Trends transitions corresponding to STSM transition classes were used to facilitate the disaggregation of crosswalked RCP cell values (Table 4).

    Table 4. Trends transitions and average annual transition amount over the study region between 1973 and 2000
    Trends “from” Class Trends “to” Class Annual Transition (km2)
    Agriculture Developed 25.9
    Agriculture Forest 7.6
    Agriculture Grass/shrub 7.5
    Forest Agriculture 13.8
    Forest Developed 55.0
    Forest Grass/shrub 59.2
    Grass/shrub Agriculture 5.1
    Grass/shrub Developed 2.7
    Grass/shrub Forest 0.0
     | Show Table
    DownLoad: CSV

    Ecoregion-based transition values from Trends were distributed to zones based on the proportional area of each zone relative to its region. Crosswalked RCP transitions were matched with Trends transitions and apportioned to zones as follows:

    Ai,j,z(t)=Pi,j,c(t)AcTi,j,zzcTi,j,z
    (3)

    where Ai,j,z(t) represents the area transitioning from class i to class j for zone z in timestep t. Pi,j,c(t) represents the RCP proportion of area transitioning from class i to j for cell c in timestep t. Ac is the area of the cell and Ti,j,z is the Trends area transitioning from class i to j in zone z. Transitions downscaled to zones were then aggregated to regions.

    2.6. Crosswalk to STSM classes

    In order to utilize the downscaled RCP projections within our STSM approach, we first had to convert the broad RCP classification scheme into transition classes more consistent with remote sensing derived LULC maps. For example, the RCPs classify vegetation as either “primary” or “secondary”, where “primary” land refers to land that has not been altered by human activity and “secondary” land is land that is recovering from human disturbance [4]. On the other hand, remote sensing-based LULC maps most often classify vegetation according to basic functional types, such as whether a cell is dominated by forest or grass/shrub. To account for discrepancies between classification schemes, we cross-walked the RCP scheme to the classification system used for the STSM (Table 5).

    Table 5. RCP classes and corresponding STSM classes excluding forest to biomass transition classes
    RCP Class STSM Classes
    Pasture Agriculture, Grass/shrub
    Primary Land Forest, Grass/shrub, Wetland
    Secondary Land Forest, Grass/shrub, Wetland
    Crop Agriculture
    Urban Developed
     | Show Table
    DownLoad: CSV

    Translating from RCP transitions to STSM transitions would have been straight forward had each RCP class corresponded to a unique STSM class. However, the “primary-land”, “secondary-land”, and “pasture” RCP classes potentially encompass more than one STSM class. The “primary-land” and “secondary-land” RCP classes were split between STSM classes “grass/shrub”, “forest”, and “wetland”. Cross-walking the “pasture” RCP class was especially problematic since the RCP definition of “pasture” included rangeland, which encompasses both grazing lands dominated by natural vegetation as well as croplands producing forage. Since no unique STSM equivalent was available, the RCP class “pasture” was split between STSM classes “grass/shrub” and “agriculture”.

    RCP transitions into and out of “pasture”, “primary-land”, and “secondary-land” classes were allocated among possible STSM transitions based on landcover area derived from an NLCD 2006 map. RCP transitions were first matched to corresponding STSM transitions. For example the RCP transition primary-land to pasture potentially encompasses five possible STSM transitions: wetland to agriculture, wetland to grass/shrub, grass/shrub to agriculture, forest to agriculture and forest to grass/shrub (Figure 3). The RCP transition was allocated among STSM transitions considering both the “from” class and “to” class of each crosswalked STSM transition. For each STSM transition the proportion of NLCD area of the “from” class of the transition relative to the NLCD area of all STSM classes crosswalked from the RCP “from” class was calculated. Next the proportion of NLCD area of the “to” class of the transition relative to the NLCD area of all STSM classes crosswalked from the RCP “to” class was calculated. RCP transition was allocated to STSM transitions based on the “from” class and “to” class proportions of each STSM transition.

    Figure 3. Crosswalking primary-land to pasture RCP transition to all potential STSM transition classes.

    The three RCP forest to harvested biomass transitions uniquely represent the STSM transition forest harvest and were simply aggregated into a single STSM forest harvest transition. All transitions out of “developed” land were not relevant to our effort and were ignored. Primary-to-secondary transitions were also ignored since the intent of our modeling effort was to examine LULC change, rather than natural environmental transitions.

    2.7. State-and-transition simulation modeling

    State-and-transition simulation models (STSMs) have been used in a variety of applications including simulating landscape level vegetation and fire dynamics, modeling the progression of invasive species, and projecting habitat suitability [28,29,30]. STSMs use an adaptive Markov chain approach to simulate changes in state class (e.g. vegetation, land use class) over time and can be used to model succession, disturbance, and management interactions [31]. An STSM splits the landscape into a set simulation cells which are each given an initial state class. State class transition pathways can be assigned either stochastic probabilities of occurrence or deterministic transition values [31]. More recently STSM’s have been designed to incorporate a spatial dimension, such as a region level stratification, or to be run in a spatially explicit fashion where transition probabilities are assigned by cell location [32]. While STSMs are suitable for modeling probability driven scenarios, only a few efforts have explicitly incorporated climate change scenarios into STSMs [33,34].

    We developed three, probabilistic LULC change downscaling scenarios (i.e. Area scenario, Composition scenario, and Transition scenario) for use in an STSM in order to test our three downscaling approaches. We converted the STSM transition class data from annual area targets into annual transition probabilities from 2005-2100 for each scenario. Annual transition probabilities were calculated from transition target values by dividing the transition target value by the total amount of the transition “from” class within each ecoregion for each timestep. Unlike transition targets, probabilities are driven by the amount of land available for transition.

    LULC change scenario modeling was done in the ST-Sim environment [35]. All scenarios were run using Level III Ecoregions as the spatial strata. Simulations were run annually from 2005 to 2100 using a 10 km2 cell size and 100 Monte Carlo replicates over the 266,730 km2 study area. Initial land-cover state class values were derived from a harmonized classified LULC product. The harmonized dataset was developed using a pixel-based data fusion process combining a collection of fifteen land classification datasets into a single product and validated with remotely sensed imagery [36].

    3. Results

    3.1. Pacific Northwest

    Under RCP 6.0, the Pacific Northwest was projected to see increases in developed land through the year 2100, with gains highly concentrated in the Willamette Valley, Puget Lowland, and Coast Range ecoregions. Large declines in forest cover and agriculture were largely the result of demand for developed land use, although agriculture increases were found in the East Cascades, North Cascades, Klamath Mountains, and Coast Range ecoregions (Figure 4). Forest harvest increased sharply by the year 2100 across all downscaling scenarios.

    Figure 4. Projected LULC change by downscaling scenario between 2005 and 2100 over all ecoregions for A) Agriculture, B) Developed, C) Forest, and D) Grass/Shrub. Error bars represent the 95th percentile of Monte Carlo simulations.

    Downscaling scenarios were compared based on the mean and 95th percentile of Monte Carlo replicates. The difference between two scenarios was only considered important if each scenario’s mean value of Monte Carlo replicates fell outside the range of variation of the other scenario’s Monte Carlo replicates. Average LULC change totals for the Pacific Northwest between the Area and Composition scenarios were within each other’s ranges of variation and thus were not found to be large enough to be important (Figure 4). However, several differences in projected LULC emerged between the Transition scenario and the Area and Composition scenarios. Projected agriculture loss, as well as developed and grass/shrub and gains were greater under the Transition scenario (Figure 4). Important differences between projected forest changes were not found between scenarios (Figure 4).

    Differences between Transition scenario projections and Area and Composition scenario projections over the study site were driven by the downscaling of transitions for cells on the study site border. The impact of the Columbia Plateau ecoregion, an ecoregion outside the study site, explains a great deal of the difference seen in agriculture and grass/shrub change between downscaling methods. The Columbia Plateau ecoregion has a much higher Trends transition value for the grass/shrub to agriculture transition than neighboring ecoregions. Both the North Cascades and East Cascades ecoregions were apportioned less grass/shrub to agriculture transition than the neighboring Columbia Plateau when applying transition-based downscaling compared to the composition or area-based downscaling (Figure 5). Since the Columbia Plateau ecoregion lies outside of the study site, grass/shrub to agriculture transition is effectively removed from the study site. This contributed to an increase in total grass/shrub and a decrease in total agriculture area for the Transition scenario compared to the Composition scenario. When combined, differences between reapportionment of RCP cells bordering ecoregions outside of the study site accounted for most of the differences in projected LULC totals.

    Figure 5. Total zone transition as a percent of total cell transition for the grass/shrub to agriculture transition in the North Coast, Eastern Cascades and Columbia Plateau ecoregions between 2005‒2100, for A) composition-based downscaling, B) transition-based downscaling.

    3.2. Ecoregion variability

    3.2.1. Developed change

    Developed land-use totals increased over all ecoregions by the year 2100, with the largest increases occurring in the Cascades, Coast Range, Puget Lowland and Willamette Valley ecoregions (Figure 6). Of these ecoregions, the largest difference between downscaling methods occurred in the Puget Lowland ecoregion, where the Transition scenario projected nearly twice the amount of developed land than Area or Composition scenarios (Figure 6). The Transition scenario also differed from the Area and Composition scenarios in the Cascades, North Cascades and Willamette Valley ecoregions (Table 6). Between Area and Composition scenarios no important differences were found since the means of simulation runs were within the range of variability of each method (Figure 6).

    Figure 6. Projected total developed area change by ecoregion and downscaling scenario between 2005‒2100 (average over 100 Monte Carlo replicates). Error bars represent the 95th percentile of Monte Carlo simulations.

    The transitions forest to developed and agriculture to developed drove most of the change in developed land over our study site as transitions out of developed were not included in the scenarios and grass/shrub to developed transitions were small. For composition-based downscaling, the amount of the “from” class of the transition determines how border cells are reapportioned. Therefore, the amount of forest and agriculture land at border cells drives reapportioning, not the amount of developed land. However, in the Puget Lowland we expected the presence of existing developed land in the greater Seattle Metropolitan area would influence patterns of future development (Figure 7) since new development is likely to occur near existing developed areas [37]. Compared to the Transition scenario, the Composition scenario led to less forest to developed transition in the Puget Lowland (Figures 7BC). Apportioning transitions into developed land without incorporating historic patterns of development may lead to an under-allocation of transition in the Puget Lowland.

    Figure 7. Total zone transition as a percent of total cell transition for the forest to developed transition in the Coast Range, Puget Lowland, Cascades, and North Cascades ecoregions, for A) area-based downscaling B) composition-based downscaling and C) transition-based downscaling.

    Area-based downscaling split RCP transitions between ecoregions based solely on the proportion of cell area within respective ecoregions. This method does not follow any spatial pattern; instead, the relative size of each zone determines the amount of transition apportioned to each ecoregion. Again, the high apportionment of transitions into developed land that we would expect to see in the Puget Lowland based on the historical presence of developed land did not occur in the area-based scenario (Figure 7A). Like composition-based reapportioning, area-based reapportioning leads to an under-allocation of transitions into developed land in the Puget Lowland.

    When applying transition-based downscaling, greater Trends transition values for transitions into developed land in the Puget Lowland compared to adjacent ecoregions led to a greater apportionment of the RCP transitions into developed in the Puget Lowland and subsequently a higher amount of change into developed land (Figure 7C). Transition-based downscaling led to an apportionment of developed land more consistent with historical, empirical patterns.

    3.2.2. Agriculture change

    Between 1973 and 2000, agriculture in the Pacific Northwest was largely concentrated in the Eastern Cascades, Puget Lowland, and Willamette Valley ecoregions [23]. Under each of the three downscaling scenarios, agriculture was projected to increase in the Eastern Cascades, and decline in the Puget Lowland and Willamette Valley ecoregions (Figure 8). Differences between downscaling scenarios were small for most ecoregions. In the Cascades and Willamette Valley ecoregions, the Composition scenario mean was outside the range of variability of both the Area and Transition Scenarios (Figure 8). Differences were also found between the Area and Transition scenarios in the Coast Range and Puget Lowland ecoregions.

    Figure 8. Projected total agriculture change by ecoregion and downscaling scenario between 2005‒2100 (average over 100 Monte Carlo replicates). Error bars represent the 95th percentile of Monte Carlo simulations.

    In the Willamette Valley the Composition scenario led to the highest projected decline in agriculture. Agriculture in the Willamette Valley constitutes a much higher percentage of the landscape (16% from NLCD 2006), than in surrounding ecoregions, and follows the ecoregion border closely (Figure 9). As a result, most of the agricultural loss in border cells (e.g. agriculture to forest) was allocated to the Willamette Valley (Figure 9B). The apportionment of agriculture transitions in this case demonstrates a typical problem with composition-based downscaling approach. Composition-based downscaling apportions transitions based on the presence of a transition’s “from” class. However, the presence of a land-use is not necessarily indicative of future change. In some cases transitions out of a land-use will more likely occur in areas where land-use is rare since conditions are likely to be less suitable. Composition-based reapportioning may lead to an over-apportionment of the agriculture to forest transition to the Willamette Valley and subsequent overestimation of the agricultural decline.

    Figure 9. Total zone transition as a percent of total cell transition for the agriculture to forest transition in the Coast Range, Willamette Valley, and Cascades ecoregions, for A) area-based downscaling, B) composition-based downscaling, and C) transition-based downscaling.

    Another problem with composition-based downscaling is that all transitions out of a LULC class are reapportioned equally. While agriculture is available for transition to forest in the Willamette Valley ecoregion, agriculture will likely transition to developed land given a high demand for developed land. In contrast, in the forest-dominated Coast Range, agricultural land may be more likely to transition back to forest. Potential differences between the likelihood of transitioning into the “to” class are not well distinguished in composition-based downscaling and may lead to implausible results.

    The Transition scenario resulted in lower amounts of agriculture returning to forest in the Willamette Valley, the result of capturing recent historical LULC conversions occurring in the neighboring Coast Range ecoregion [23].

    3.2.3. Forest change

    Large reductions in forest were projected for all ecoregions except for the Willamette Valley where forests were projected to increase by 339 km2, 619.1 km2 and 131.0 km2 for the Area, Composition, and Transition scenarios respectively (Figure 10). The Transition scenario projected higher total forest loss in the Puget Lowland ecoregion and less forest loss in the Cascades ecoregion than the Area or Composition scenarios (Figure 10). In the Willamette Valley the Composition scenario led to greater forest gains than the Area or Transition scenarios (Figure 10).

    Figure 10. Projected total forested area change by ecoregion and downscaling scenario between 2005‒2100 (average over 100 Monte Carlo replicates). Error bars represent the 95th percentile of Monte Carlo simulations.

    In the Puget Lowland greater forest loss projected by the Transition scenario was largely driven by two factors. First, a high Trends transition value for the forest to developed transition in the Puget Lowland, relative to surrounding ecoregions, led to a greater apportionment of the transition into the Puget Lowland when applying transition-based downscaling opposed to area or composition-based downscaling. Secondly, forest area constitutes a smaller percentage of total area in the Puget Lowland (40% from NLCD 2006) than in the forest dominated Coast Range (66% from NLCD 2006) and Cascades (75% from NLCD 2006) ecoregions [23]. Since, composition-based downscaling apportions transition values based on the “from” class of the transition, less transition out of forest was apportioned to the Puget Lowland when applying composition-based downscaling than for area or transition-based downscaling. Combined, these factors led to a large difference in projected forest change between the Transition and Composition scenarios.

    Large amounts of forest harvest are projected by 2100, with most harvest occurring the in Cascades, Coast Range and Klamath Mountains ecoregions. However, differences between downscaling scenarios were within each other’s range of variability and not large enough to be considered important (Figure 11).

    Figure 11. Projected total forest harvest by ecoregion and downscaling scenario between 2005 and 2100 (average over 100 Monte Carlo replicates). Error bars represent the 95th percentile of Monte Carlo simulations.

    3.2.4 Grass/shrub change

    Grass/Shrub area was projected to increase in all ecoregions with the exception of the Eastern Cascades and North Cascades (Figure 12). Only in the North Cascades and Willamette Valley between the Transition and Composition scenarios and in the Willamette Valley between Composition and Area scenarios were differences in projected grass/shrub seen.

    Figure 12. Projected total grass/shrub area change by ecoregion and scenario between 2005‒2100 (average over 100 Monte Carlo replicates). Error bars represent the 95th percentile of Monte Carlo simulations.

    While grass/shrub makes up a relatively small proportion of landcover within our study region, the Columbia Plateau, Northern Basin and Range and Blue Mountains ecoregions to the East of the study region are largely composed of grass/shrub (Figure 13A). However, looking closely at cells on the border of the Columbia Plateau and the Northern Cascades, actual grass/shrub boundaries are not as abrupt as ecoregion boundaries. In some places dense grass/shrub areas spill over into the Northern Cascades (Figure 13B).

    Figure 13. A) The Northeast edge of the study site with grass/shrub area from NLCD 2006 and RCP cell boundaries. B) A close up of an RCP cell spanning the North Cascades and Columbia Plateau Ecoregions.

    For transitions out of grass/shrub, transition-based downscaling may underestimate the amount of transition that should be apportioned to ecoregions such as the Northern Cascades because Trends values are derived for ecoregions as a whole. Within ecoregion variation, not captured in the Trends transition value, may warrant higher rates of transition near ecoregion borders.

    The composition-based downscaling approach may be beneficial in this case because the NLCD map can indicate where gradual change between LULC occurs (Figure 13B). The impact of this type of spill over is not always clear, however, since historical drivers of LULC change such as management practices may be more important than land-cover in determining how LULC change should be reapportioned between ecoregions.

    4. Discussion

    Of our three downscaling approaches, the area-based approach is the simplest to implement because it requires no additional datasets. This method may be entirely sufficient when splitting transitions across a homogenous landscape and would also be suitable if land-use history data or a land-use composition data was not available or reliable. However, area-based downscaling has several key disadvantages. In some cases, regions have distinct differences in the amount of expected LULC transition. Area-based downscaling fails to incorporate the regional heterogeneity of the landscape in the downscaling method. As demonstrated above, this can lead to large differences in land-use projections and reduced confidence in the scenario itself. Results from area-based downscaling may be especially questionable when a large proportion of RCP cells fall on the edge of two or more regions of interest. Such is the case with most of the ecoregions within our study site. The Willamette Valley, for example is made up of only edge cells.

    Differences between the Area and Composition scenarios were mostly within each other’s range of variability, indicating that they were not large enough to be important. One main benefit of composition-based downscaling is that it accounts for regional heterogeneity between LULC. However, the composition-based downscaling assumes LULC composition is a suitable proxy for LULC change. This may prove inaccurate for transitions where land management practices are more important drivers of change than landcover presence [38]. Furthermore, since composition-based downscaling is determined only by the relative zone composition of the “from” class for each transition, transitions can be apportioned to ecoregions where the LULC of the “to” class is not historically present. LULC composition datasets, such as the NLCD, are available for the conterminous United States, and coarser datasets such as the Global Landcover 2000 dataset are available globally [39]. Datasets like the MODIS dataset have been effective in downscaling scenario data and may be applicable in this case as well [13].

    The transition-based downscaling method has the inherent advantage of using the same data type as the original data to downscale. Differences between the Transition scenario and the Area and Composition scenarios were found for most LULC classes, and were especially pronounced for projections of developed land. Historical transition data can be used to guide future patterns of transitions not evident in land-cover composition data. Land-use management and ownership patterns are inherently represented in historical transition datasets, providing an added benefit over area and composition-based downscaling methods. However, one potential problem with transition-based downscaling is the process inherently assumes regional patterns and distribution of land-use change will continue on a similar trajectory into the future. In reality LULC change is driven by a host of processes, both physical and anthropogenic, with interactions varying across space and time [40]. In addition, land-use transition datasets are less likely to be available at desired regional scales. The most important advantages and shortcomings of each of the three downscaling methods are outlined in Table 6.

    Table 6. Summary of advantages and disadvantages of area-based, composition-based, and transition-based downscaling approaches
    Downscaling Method
    Area Composition Transition
    Advantages 1. No additional data needed.
    2. May be applicable when region boundaries do not follow LULC patterns.
    1. Easy to find composition dataset like “NLCD”.
    2. Does a better job than Area based downscaling at accounting for transition differences between regions.
    1. Using transition history data to downscale incorporates historical management and ownership patterns.
    2. Historical transition data takes into account both the “from” and “to” class of transition.
    Shortcomings 1. Reapportions transition data without regard to regional land use pattern or history.
    2. Simplistic approach, can lead to large discrepancies.
    1. Since only the “from” class of the transition is considered, in some cases, may not accurately reapportion transition data.
    2. The presence of a land-cover is not necessarily indicative of future change.
    1. Low availability on large scale or globally.
     | Show Table
    DownLoad: CSV

    Several other factors impact the outcome of the downscaling method for reapportioning RCP values. Total region size determines the relative influence of edge cells since reapportioned edge cells have relatively less impact when aggregating to larger regions. Also, the shape of the region, specifically the ratio between area and perimeter of the region will impact the proportional contribution of edge cells to the total transition amount within the ecoregion (e.g. Willamette Valley). Ecoregions do a very good job of spatially organizing LULC change, but do not perform well when using an area-based downscaling approach. If spatial strata or regions are a) sufficiently coarse, b) represent large homogenous regions, or c) based on administrative boundaries or some other non-ecological framework, area-based downscaling might be a more suitable approach.

    Since our STSM was driven solely by RCP transition probabilities and available land-use for transition, model outputs closely reflect RCP projections. It should be noted that an initial inventory of RCP transition totals at the ecoregion level suggests that a validation process is needed to ensure values are within reasonable bounds.

    5. Conclusion

    Here we present the results of three downscaling methods used to downscale RCP transition values to scales relevant for analysis in an STSM model. While all three methods may be sufficient in certain circumstances, downscaling using a historical LULC transition data such as the Trends dataset has some distinct advantages. Most importantly using a historical transition dataset to downscale RCP transition values accounts for regional patterns in both the “from” and “to” class of the RCP transition. Historical LULC change data also often captures LULC change information related to land ownership and management. When historical transition data is not available it may be necessary to apply an area or composition downscaling approach instead.

    To our knowledge this is the first attempt to use global RCP LULC transition data to drive a region level STSM of LULC change. This paves the way for more sophisticated STSM analysis of LULC change using the latest global climate data. One such refinement will be to run the model in a spatially explicit manner as was done by Wilson et al. [21]. In this case a raster of LULC was used to initialize the model spatially. Also, spatial multipliers were included to further spatially constrain allowable transition areas. The use of downscaled RCP data in an STSM allows modelers to incorporate climate change scenarios at the regional scale. It will also enable comparisons between study areas, scales, and land cover types within a larger framework.

    Our approach was developed specifically for reapportioning RCP transition values to region level scales, but it may be applicable to several other global gridded LULC change products. Our reapportioning techniques are immediately applicable to the HYDE dataset which provides global gridded LULC change data for the past 12,000 years [41]. Downscaling historic and future data projections will provide a consistent and continuous set of LULC change data that can be used within a STSM.

    Acknowledgments

    The research was supported by the U.S. Geological Survey’s Climate and Land Use Research and Development Program. We would like to thank reviewers Tamara S. Wilson and Miguel Villarreal.

    Conflict of Interest

    The authors declare no conflict of interest.

    Disclaimer

    Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.



    [1] J. Rehm, C. Mathers and S. Popova, et al., Global burden of disease and injury and economic cost attributable to alcohol use and alcohol-use disorders, Lancet, 373 (2009), 2223–2233.
    [2] J. Rehm, The risk associated with alcohol use and alcoholism, Alcohol Res. Health, 34 (2011), 135–143.
    [3] C. Lin, C. Lin and L. Mo, et al., Heavy alcohol consumption increases the incidence of hepatocellular carcinoma in hepatitis B virus-related cirrhosis, J. Hepatol, 58 (2013), 730–735.
    [4] WHO, Global Status Report on Alcohol and Health 2014, Geneva, Switzerland, 2014.
    [5] J. S. Gill, Reported levels of alcohol consumption and binge drinking within the UK undergraduate student population over the last 25 years, Alcohol, 37 (2002), 109–120.
    [6] L. Smith and D. Foxcroft, Drinking in the UK: An exploration of trends, 2009.
    [7] O. Sharomi and A. B. Gumel, Curtailing smoking dynamics: a mathematical modeling approach, Appl. Math. Comput., 195 (2008), 475–499.
    [8] H. F. Huo and X. M. Zhang, Complex dynamics in an alcoholism model with the impact of Twitter, Math. Biosci., 281 (2016), 24–35.
    [9] E. White and C. Comiskey, Heroin epidemics, treatment and ODE modelling, Math. Biosci., 208 (2007), 312–324.
    [10] G. Mulone and B. Straughan, A note on heroin epidemics, Math. Biosci., 218 (2009), 138-141.
    [11] B. Benedict, Modelling alcoholism as a contagious disease: how "infected" drinking buddies spread problem drinking, SIAM News, 40 2007.
    [12] J. L. Manthey, A. Aidoob and K. Y. Ward, Campus drinking:
    [13] an epidemiological model, J. Biol. Dyn., 2 (2008), 346–356.
    [14] 13 A. Mubayi, P. Greenwood and C. Castillo-Chavez, et al., The impact of relative residence times on the distribution of heavy drinkers in highly distinct environments, Socio. Econ. Plan Sci., 44 (2010), 45–56.
    [15] 14 R. Bani, R. Hameed and S. Szymanowski, et al., Influence of environmental factors on college alcohol drinking patterns, Math. Biosci. Eng., 10 (2013), 1281–1300.
    [16] 15 B. Buonomo and D. Lacitignola, Modeling peer influence effects on the spread of high-risk alcohol consumption behavior, Ricerche mat., 63 (2014), 101–117.
    [17] 16 A. Mubayi and P. Greenwood, Contextual interventions for controlling alcohol drinking, Math. Popul. Stud., 20 (2013), 27–53.
    [18] 17 G. Mulone and B. Straughan, Modeling binge drinking, Int. J. Biomath., 5 (2012), 57-70.
    [19] 18 H. F. Huo and N. N. Song, Golobal stability for a binge drinking model with two stages, Discrete Dyn. Nat. Soc., 2012, Article ID 829386, (2012), 15 pages.
    [20] 19 C. P. Bhunu and S. Mushayabasa, A theoretical analysis of smoking and alcoholism, J. Math. Model. Algor., 11 (2), 387–408.
    [21] 20 C. E. Walters, B. Straughan and R. Kendal, Modeling alcohol problems: total recovery, Ricerche Mat., 62 (2013), 33-53.
    [22] 21 H. Xiang, C. C. Zhu and H. F. Huo, Modelling the effect of immigration on drinking behaviour, J. Biol. Dyn., 11 (2017), 275–298.
    [23] 22 X. Y. Wang, H. F. Huo and Q. K. Kong, et al., Optimal control strategies in an alcoholism model, Abstr. Appl. Anal., 3 (2014), 1-18.
    [24] 23 C. C. Zhu and J. Zhu, Stability of a reaction-diffusion alcohol model with the impact of tax policy, Comput. Math. Appl., 74 (2017), 613–633.
    [25] 24 H. F. Huo, R. Chen and X. Y. Wang, Modelling and stability of HIV/AIDS epidemic model with treatment, Appl. Math. Model., 40 (2016), 6550–6559.
    [26] 25 H. F. Huo and M. X. Zou, Modelling effects of treatment at home on tuberculosis transmission dynamics, Appl. Math. Model., 40 (2016), 9474–9484.
    [27] 26 H. Xiang, Y. L. Tang and H. F. Huo, A viral model with intracellular delay and humoral immunity, Bull. Malays. Math. Sci. Soc., 40 (2017), 1011–1023.
    [28] 27 H. Xiang, Y. Y. Wang and H. F. Huo, Analysis of the binge drinking models with demographics and nonlinear infectivity on networks, J. Appl. Anal. Comput., 8 (2018), 1535–1554.
    [29] 28 Y. L. Cai, J. J. Jiao and Z. J. Gui, et al., Environmental variability in a stochastic epidemic model, Appl. Math. Comput., 3(2018), 210–226.
    [30] 29 Z. Du and Z. Feng, Existence and asymptotic behaviors of traveling waves of a modified vectordisease model, Commun. Pur. Appl. Anal., 17 (2018), 1899–1920.
    [31] 30 X. B. Zhang, Q. H. Shi and S. H. Ma, et al., Dynamic behavior of a stochastic SIQS epidemic model with levy jumps, Nonlinear Dynam., 93 (2018), 1481–1493.
    [32] 31 W. M. Wang, Y. L. Cai and Z. Q. Ding, et al., A stochastic differential equation SIS epidemic model incorporating Ornstein-Uhlenbeck process, Phys. A, 509 (2018), 921–936.
    [33] 32 H. F. Huo, F. F. Cui and H. Xiang, Dynamics of an SAITS alcoholism model on unweighted and weighted networks, Phys. A, 496 (2018), 249–262.
    [34] 33 X. Y. Meng and Y. Q. Wu, Bifurcation and control in a singular phytoplankton-zooplankton-fish model with nonlinear fish harvesting and taxation, Int. J. Bifurcat. Chaos, 28 (2018),1850042, 24 pages.
    [35] 34 A. K. Misra, A. Sharma and J. B. Shukla, Modeling and analysis of effects of awareness programs by media on spread of infectious diseases, Math. Comput. Model., 53 (2011), 1221–1228.
    [36] 35 H. F. Huo and Q. Wang, Modeling the influence of awareness programs by media on the drinking dynamics, Abstr. Appl. Anal., 2014 (2014), 1–8.
    [37] 36 H. Xiang, N. N. Song and H. F. Huo, Modelling effectss of public health educational campaigns on drinking dynamics, J. Biol. Dyn., 10 (2016), 164–178.
    [38] 37 S. H. Ma, H. F. Huo and X. Y. Meng, Modelling alcoholism as a contagious disease: A mathematical model with awareness programs and time delay, Discrete Dyn. Nat. Soc., 2015 (2015), 1–13.
    [39] 38 H. Guo, M. Y. Li and Z. Shuai, Global stability of the endemic equilibrium of multigroup SIR epidemic models, Can. Appl. Math. Q., 14 (2006), 259–284.
    [40] 39 H. Guo, M. Y. Li and Z. Shuai, A graph-theoretic approach to the method of global Lyapunov functions, Proc. Amer. Math. Soc., 136 (2008), 2793–2802.
    [41] 40 M. Y. Li and Z. Shuai, Global-stability problem for coupled systems of differential equations on networks, J. Diff. Equs., 248 (2010), 1–20.
    [42] 41 A. Berman and R. J. Plemmons, Nonnegative Matrices in the Mathematical Sciences, Academic Press, NewYork, 1979.
    [43] 42 Y. Muroya, Y. Enatsu and T. Kuniya, Global stability for a multi-group SIRS epidemic model with varying population sizes, Nonlinear Anal. RWA, 14 (2013), 1693–1704.
    [44] 43 R. Sun, Global stability of the endemic equilibrium of multigroup SIR models with nonlinear incidence, Comput. Math. Appl., 60 (2010), 2286–2291.
    [45] 44 M. Y. Li, Z. Shuai and C. Wang, Global stability of multi-group epidemic models with distributed delays, J. Math. Anal. Appl., 361 (2010), 38–47.
    [46] 45 Y. Muroya and T. Kuniya, Further stability analysis for a multi-group SIRS epidemic model with varying total population size, Appl. Math. Lett., 38 (2014), 73–78.
    [47] 46 J. Wang and H. Shu, Global dynamics of a multi-group epidemic model with latency, relapse and nonlinear incidence rate, Math. Biosci. Eng., 13 (2016), 209–225.
    [48] 47 F. Hoppensteadt, An age-dependent epidemic model, J. Franklin Inst., 297 (1974), 325–338.
    [49] 48 X. Z. Li, J. Wang and M. Ghosh, Stability and bifurcation of an SIVS epidemic model with treatment and age of vaccination, Appl. Math. Model., 34 (2010), 437–450.
    [50] 49 X. C. Duan, S. L. Yuan and X. Z. Li, Global stability of an SVIR model with age of vaccination, Appl. Math. Comput., 226 (2014), 528–540.
    [51] 50 M. W. Shen and Y. N. Xiao, Global stability of a multi-group SVEIR epidemiological model with the vaccination age and infection age, Acta Appl. Math., 144 (2016), 137–157.
    [52] 51 J. H. Xu and Y. C. Zhou, Global stability of a multi-group model with generalized nonlinear incidence and vaccination age, Discrete Contin. Dyn. Syst. Ser. B, 21 (2016), 977–996.
    [53] 52 F. V. Atkinson and J. R. Haddock, On determining phase spaces for functional differential equations, Funkc. Ekvacioj., 31 (1988), 331–347.
    [54] 53 G. F.Webb, Theory of Nonlinear Age-dependent Population Dynamics, Marcel Dekker, New York, 1985.
    [55] 54 G. Gripenberg, S. O. Londen and O. Staffans, Volterra Integral and Functional Equations, Cambridge University Press, New York, 1990.
    [56] 55 O. Diekmann, J. Heesterbeek and J. A. Metz, On the defnition and the computation of the basic reproduction ratio R0 in models for infectious diseases in heterogeneous populations, J. Math. Biol., 28 (1990), 365–382.
    [57] 56 P. van den Driessche and J. Watmough, Reproduction numbers and sub-threshold endemic equilibria for compartmental models of disease transmission, Math. Biosci., 180 (2002), 29–48.
    [58] 57 J. K. Hale and P. Waltman, Presistence in infinite-dimensional systems, SIAM J. Appl. Math., 20 (1989), 388–395.
    [59] 58 A. Berman and R. J. Plemmons, Nonnegative Matrices in the Mathematical Sciences, Academic Press, NewYork, 1979.
    [60] 59 J. P. Lasalle, The Stability of Dynamical Systems, in: Regional Conference Series in Applied Mathematics, SIAM, Philadelphia, 1976.
    [61] 60 J. K. Hale and S .M. V. Lunel, Introduction to Functional Differential Equations, Appl. Math. Sci., 99, Springer, New York, 1993.
    [62] 61 H. I. Freedman, S. Ruan and M. Tang, Uniform persistence and flows near a closed positively invariant set, J. Dyn. Diff. Equat., 6 (1994), 583–600.
    [63] 62 M. Y. Li, J. R. Graef and L. Wang, et al., Global dynamics of a SEIR model with varying total population size, Math. Biosci., 160 (1999), 191–213.
    [64] 63 N. P. Bhatia and G. P. Szego, Dynamical Systems: Stability Theory and Applications, Lecture Notes in Mathematics, 35, Springer, Berlin, 1967.
    [65] 64 H. L. Smith and P. Waltman,The Theory of the Chemostat: Dynamics of Microbial Competition, Cambridge University Press, Cambridge, 1995.
  • This article has been cited by:

    1. Na Dong, Lan You, Wenjia Cai, Gang Li, Hui Lin, Land use projections in China under global socioeconomic and emission scenarios: Utilizing a scenario-based land-use change assessment framework, 2018, 50, 09593780, 164, 10.1016/j.gloenvcha.2018.04.001
    2. Gregory Giuliani, Denisa Rodila, Nathan Külling, Ramona Maggini, Anthony Lehmann, Downscaling Switzerland Land Use/Land Cover Data Using Nearest Neighbors and an Expert System, 2022, 11, 2073-445X, 615, 10.3390/land11050615
    3. Zhenwang Li, Quan Tang, Xu Wang, Baorui Chen, Chengming Sun, Xiaoping Xin, Grassland Carbon Change in Northern China under Historical and Future Land Use and Land Cover Change, 2023, 13, 2073-4395, 2180, 10.3390/agronomy13082180
    4. Bojana Horvat, Nino Krvavica, Disaggregation of the Copernicus Land Use/Land Cover (LULC) and Population Density Data to Fit Mesoscale Flood Risk Assessment Requirements in Partially Urbanized Catchments in Croatia, 2023, 12, 2073-445X, 2014, 10.3390/land12112014
    5. Wei Guo, Yongjia Teng, Jing Li, Yueguan Yan, Chuanwu Zhao, Yongxing Li, Xiang Li, A new assessment framework to forecast land use and carbon storage under different SSP-RCP scenarios in China, 2024, 912, 00489697, 169088, 10.1016/j.scitotenv.2023.169088
  • Reader Comments
  • © 2019 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(5097) PDF downloads(730) Cited by(5)

Article outline

Figures and Tables

Figures(1)  /  Tables(1)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog