Appendix 1 Detailed descriptions of the original procedures and sub-models of IM-LUDAS.
Define Initial SLCP Plot: This procedure defines SLCP plots in the initial year of each simulation according to the local implementation rules of the policy. Through this procedure, the initial SLCP plots of the prescribed quota are randomly located for every household in the two Mongolian villages in the following spatially bounded way: flatland is preferentially selected, and other land is targeted when available flatland is not found within the landholdings. This selection process does not consider land degradation levels, so the selected plots can include both degraded and non-degraded land. However, since the procedure is run at the initialization stage, the landscape agents’ vegetation and soil properties, which are calculated by the biophysical sub-models, are still blank (see Fig. 1). The Land Restoration Dynamicssub-model running in the SLCP plots represents restoration of vegetation and soil conditions, with reference to land history. Hence, the procedure sets the land history of the selected plotsto seven (years) plusa random value from zero to 25: seven indicates the time that has already elapsed since the start of the first round of the SLCP;twenty-five is the time that would be taken for shifting sandy pasture to be restored to the same level of fixed sandy pasture (Miyasaka et al. 2014); and the randomness is added in order to represent the variety of degradation levels among the plots. The labor force required at each SLCP plot is set to the empirically estimated value for tree plantations (Appendix 2).
Implement New Policy: This procedure decides if a hypothetical policy is implemented according to the user’s setting. If the user sets some hypothetical policy, the procedure cancels the second round of the SLCP and instead executes the hypothetical policy from 2011 by performing the Define New SLCP Plot procedure.
Define New SLCP Plot: This procedure annually executes a hypothetical SLCP by selecting participating households and their implementation plots in line with the user’s policy settings. Through the procedure, households that satisfy the economic criterion set by the user are chosen; the plots that satisfy the agricultural criterion set by the user are also selected from the participating households’ cropland; their land-use type is converted to tree plantations; labor force required at each plot is set as in the Define Initial SLCP Plot procedure; their SLCP history is set to zero; and their land history, which is to be used in the Land Restoration Dynamicssub-model, is reset with reference to the degree of their crop yield decline to reflect previous degradation levels in the restoration process.
Land Restoration Dynamics: Planting trees and fallowing land by stopping any land use are major land restoration measures in northeast China. The Land Restoration Dynamicssub-model is run in plots where the SLCP is executed or any land use is ceased. Restoration speed at different topographic locations differs even under the same measures (Miyasaka et al. 2014). This sub-model includes four types of model, differentiated by the combination of two restoration measures (planting trees and stopping land use) and two topographic types (flatlands and sand dunes). Each type of model estimates plant species composition, vegetation cover, plant species diversity (see Appendix 3 for the list of plant species that these vegetation indices are based on), soil organic carbon, total nitrogen, total phosphorous, and sand, silt, and clay content; the explanatory variable is the implementation period of the restoration measures. All these models were empirically constructed by linear, log-linear, or polynomial regression with our field-collected data (Appendix 4). According to Miyasaka et al. (2014), we defined five restoration levels by plant species composition and associated the lower two levels with shifting and semi-fixed sandy pasture and the higher three levels with fixed sandy pasture. Any plot whose land-use type is restored to fixed sandy pasture can be seen as available again by households, but the probability of becoming available varies between the three restoration levels (Appendix 2). State change by this sub-model is linked to land-use change in this manner. The category “tree plantations”does not appear to change by the result of the sub-model. This is because—in contrast to“pasture” sub-categories(i.e., “fixed sandy pasture,” “semi-fixed sandy pasture,” and “shifting sandy pasture”)—we have not specifically defined sub-categories of “tree plantations.”Although the internal conditionsin tree plantations(including plant species composition and soil properties)dochange,this change is considered to bevariation within the category.
Land Use Choice: This sub-model represents heterogeneity of household land-use decisions, including three types of model empirically built for the three predefined livelihood types. The land-use decision models are in the form of multinomial logistic regression models (Appendix 5). The response variable is four typical land-use categories in the area: maize cropland, bean-centered cropland, paddy, and pasture. The explanatory variables are the three topographic types (lowlands, flat sandy lands, and sand dunes), the number of owned livestock, and distance between the location of a household’s house and each pixel of its own land. Every household agent annually performs one of the three models in accordance with its livelihood type for every plot of its own as long as its labor budget has a remaining value. The labor force of each household is allocated in advance in the Allocate Labor procedure to different activities (i.e., crop farming, livestock farming, non-farming, and tree planting), on the basis of the household’slivelihood type. The labor force needed at each plot is set to the empirically estimated value for the land-use type decided (Appendix 2). Every time the land-use type of a plot is decided, the labor force for the plot is deducted from the allocated labor budget. As stated in the description of Individual decision-making, the household agents do not always choose land-use or location options with the highest probability because of the ordered-choice algorithm (Le et al. 2008).
Calculate Grazing Pressure: Grazing pressure (i.e., livestock density) is a main determinant of pasture degradation and an explanatory variable of the Pasture Degradation Dynamicssub-model. This sub-model calculates grazing pressure (GP)as shown in Eq. (1):
GP=Hlvstck/Ppstr (1)
whereHlvstckis the number of livestock each household owns andPpstris the area of pasture it uses. The landscape agents whose land-use type is pasture perform this calculation, considering the area of pasture with the shared ownership ID and the owner households’ number of livestock. In villages where pasture is used collectively, however, the landscape agents calculateGP as shown in Eq. (2):
(2)
where i is household codes in the village. The calculation of the grazing pressure is based on a sheep unit in China (Hu and Zhang 2001).
Pasture Degradation Dynamics: Thissub-model represents different pasture degradation patterns on the basis of grazing pressure and topographic location. Empirical data used here were derived from a grazing experiment conducted in the study area (Okuro 1997), in which temporal changes in vegetation and soil properties were measured under different topographic conditions (flat sandy land and sand dunes), and under different grazing pressures (high as 6 sheep per ha, middle as 4 sheep per ha, and low as 2 sheep per ha). Six types of model for the combinations of the two topographic types and the three levels of grazing pressure were empirically constructed by linear or log-linear regression (Appendix 4). The response variables are vegetation cover, soil organic carbon, total nitrogen, total phosphorous, and sand, silt, and clay content; the explanatory variable is grazing period. We defined three degradation levels based on vegetation cover and associated the levels with the three pasture categories (i.e., fixed, semi-fixed, and shifting) (Appendix 2). State change by the sub-model is thus linked to land-use change as well as the Land Restoration Dynamicssub-model. The sub-model is carried out by the landscape agents that previously performed the Calculate Grazing Pressuresub-model.
Cropland Degradation Dynamics: This sub-model includes three types of model representing land degradation that occurred in the typical cropland types in the area: maize cropland on lowlands, maize cropland on flat sandy lands, and bean-centered cropland on sand dunes. The response variables are soil organic carbon, total nitrogen, total phosphorous, and sand, silt, and clay content; the explanatory variable is the duration of cultivation. The landscape agents whose land-use type falls into the cropland types perform a corresponding model. All of the models were empirically built by log-linear regression with our field data (Appendix 4).
Agricultural Yield Dynamics: This sub-model mainly consists of three types of model representing crop yield deterioration in the above-mentioned cropland types. We empirically estimated the maximum crop yield for each cropland type with our household data (Appendix 2) and the general patterns of temporal decrease in crop biomass in each cropland type by log-linear regression with our field-collected crop data (Appendix 4). The sub-model annually calculates crop yield per unit of area (Pyld),as shown in Eq. (3):
Pyld= aryldPyld-max (3)
where a is an adjustment parameter set for each village, ryld is the reduction rate derived from the estimated temporal patterns, andPyld-max is the maximum crop yield per unit of area.Our household data indicated that average maize yields significantly differed among the five research villages. We therefore set the adjustment parameters to simulate the local economy in a more realistic way by incorporating that variance. We also empirically estimated the lower limits of crop yields for households to decide to abandon each type of cropland (Appendix 2).
When crop yields fall below the limits, cultivation is ceased, and the cropland is changed to shifting sandy pasture. The sub-model also provides the yields of paddy estimated with our household data. The paddy yields fluctuate annually within the range of standard error but do not significantly decrease. The sub-model and the Cropland Degradation Dynamics sub-model are carried out as a set by the landscape agents.
Generate Crop Farming Income: This sub-model calculates income from crop yields estimated by the Agricultural Yield Dynamics sub-model. The coefficients of the transformation from crop yields to monetary income (unit price of crop) are based on our field observations (Appendix 2). In the sub-model, a household’s crop-farming income (Hincm-crp) is calculated as shown in Eq. (4):
(4)
where b is the unit price of corresponding crop species (1 = maize, 2 = bean, and 3 = rice), Pcrplndis cropland area, k is the plot codes for maize cropland, l is the plot codes for bean-centered cropland, and m is the plot codes for paddy.
Generate Livestock Farming Income: Modeling the entire process of earning income from livestock—including breeding and selling strategies based on the number of livestock animals based on sex and kind—is beyond the scope of this research. This sub-model was embedded only to incorporate the effect of land degradation on livestock-farming income constrained by the area of available pasture. “Available pasture” means fixed and semi-fixed sandy pasture, which households practically use. We assumed that household livestock-farming income remains unchanged from the initial values, which are derived from our household data, during a simulation, as long as the number of a household’s livestock does not change. Households feed livestock mainly by grazing them in pasture, so households would have to reduce the number of livestock if available pasture area decreases as a result of land degradation. A change in the number of livestock, however, does not necessarily correspond exactly to the area of available pasture, because a household may assume that the rest of available pasture is still enough or raise livestock without grazing, to some extent, by using a stable and collecting or purchasing forage. The sub-model simply represents the annual change in livestock number using the randomization shown in Eq. (5):
(5)
where rlvstck is the reduction ratio of a household’s livestock number, q is a random number floating evenly over [0, 1], Ppstr is the area of current pasture, and Pur-pstr is the area of initial pasture. A household’s income from livestock farming (Hincm-lvstck) is then calculated, as shown in Eq. (6):
(6)
where Hur-incm-lvstck is the initial value of Hincm-lvstck.
Generate Off Farm Income: This sub-model annually calculates household off-farm income, but modeling it precisely is outside the scope of this study. On the basis of our finding that the average off-farm income per unit of labor differed significantly among livelihood types (Appendix 2), the sub-model estimates household off-farm income (Hincm-off) as shownin Eq. (7):
(7)
where c is the predefined average off-farm income per unit of labor for each livelihood type and Hlbr-off is the labor force allocated to off-farm work.
Generate Subsidy: This sub-model annually calculates the amount of subsidy provided to participating households (Hincm-slcp),as shown in Eq. (8):
(8)
where β is the user-defined amount of subsidy per unit of area and Pslcp is the SLCP plot area. The SLCP plots whose SLCP history exceeds the user-defined implementation period are excluded from the calculation.
Classify Livelihood Type: This sub-model is an automatic classification based on dissimilarities in socioeconomic characteristics between each household and the centroid of each livelihood type.The algorithmis similar totheK-mean clustering procedure, except that the group centroids (i.e., average values) were predefined (Table 1) and constant during a simulation. The details are described in Le (2005) and Le et al. (2008).
Appendix 2 Parameters of the procedures and sub-models of IM-LUDAS except for biophysical and land-use decision regression sub-models.
Parameter / Default/baseline-scenario valueExternal Parameter Setting
Amount of subsidy / 90 yuan/mu
(160 yuan/mu is applied to only the initial year as a fixed value)
Implementation period / 8 years
Total population / 323 households
Define Initial SLCP Plot
Land quota of SLCP-plot area / Mongolian village A: 2 mu/person
Mongolian village B: 3 mu/person
Labor force required for tree plantations / 1.1 ± 0.2 man-day/mu
Define New SLCP Plot
Agricultural criteria:
maize cropland on lowlands
maize cropland on flat sandy lands
bean-centered cropland on sand dunes / N/A
N/A
N/A
Economic criterion / N/A
Land Restoration Dynamics
Restoration levels defined by an index of plant species composition:
1 (shifting sandy pasture)
2 (semi-fixed sandy pasture)
3 (fixed sandy pasture with low chance of being reused)
4 (fixed sandy pasture with middle chance of being reused)
5 (fixed sandy pasture with high chance of being reused) /
1.150
2.034
2.803, 50%
3.578, 70%
4.196, 90%
Land Use Choice
Required labor force:
maize cropland
bean-centered cropland
paddy /
8.5 ± 0.2 man-day/mu
2.9 ± 0.4 man-day/mu
8.3 ± 2.3 man-day/mu
Pasture Degradation Dynamics
Grazing pressure levels:
high
middle
low /
6 sheep/ha
4 sheep/ha
2 sheep/ha
Degradation levels defined by vegetation cover:
1 (shifting sandy pasture)
2 (semi-fixed sandy pasture)
3 (fixed sandy pasture) /
10% >
10% ≤, 50% >
50% ≤
Appendix 2 (continued)
Parameter / Default/baseline-scenario valueAgricultural Yield Dynamics
Maximum yields:
maize cropland on lowlands
maize cropland on flat sandy lands
bean-centered cropland on sand dunes /
900 kg/mu
700 kg/mu
55 kg/mu
Adjustment parameters of crop yields:
Mongolian village A (maize in cropland on flat sandy lands)
Mongolian village B (maize in cropland on flat sandy lands)
Mongolian village C (maize in cropland on flat sandy lands)
Han village A (maize in cropland on flat sandy lands)
Han village A (maize in cropland on lowlands)
Han village B (maize in cropland on flat sandy lands)
Han village B (maize in cropland on lowlands) /
0.71
1.41
1.01
1.59
1.30
1.32
1.12
Lower limits of yields:
maize cropland on lowlands
maize cropland on flat sandy lands
bean-centered cropland on sand dunes /
175 ± 20 kg/mu
140 ± 16 kg/mu
11 ± 7 kg/mu
Paddy yields / 307 ± 27 kg/mu
Generate Crop Farming Income
Unit price of crops:
maize
bean
rice /
1.4 yuan/kg
5.4 yuan/kg
1.4 yuan/kg
Generate Off Farm Income
Average off-farm income:
livestock-farming type
crop-farming type
non-farming type /
76 ± 19 yuan/labor
76 ± 19 yuan/labor
160 ± 23 yuan/labor
Classify Livelihood Type
Classification criteria (average values of socioeconomic characteristics of each livelihood type) / see Table 1
Labor-allocation rules used for updates:
livestock-farming type
crop-farming
livestock-farming
non-farming
crop-farming type
crop-farming
livestock-farming
non-farming
non-farming type
crop-farming
livestock-farming
non-farming /
45 ± 4%
22 ± 2%
3 ± 1%
48 ± 5%
3 ± 2%
4 ± 2%
43 ± 6%
9 ± 3%
12 ± 3%
Update Household State
Range of working age as a householder / 27 ± 1 – 77 ± 1 year
Annual fluctuation of labor force per household / ± 14.4 man-day
Appendix 3 List of plant species thatLand Restoration Dynamics sub-model is based on.