An Agent-based Approach For Modeling Agricultural Land System Dynamics

Existing at the center of the coupled human and natural systems, an agricultural system is defnedas a complex, human-managed land use system intended to provide foods and services for humans.Promoted by the current global change and sustainability science, several achievements have beenmade in agricultural land system studies, such as the exploration on spatial-temporal characteristicsand their drivers of agricultural land change, and the integrated simulations. In spite of considerableprogress, grand challenges still remain in this emerging field. One critical problem is that currentpractices based on the characterization of different land use and land cover types are overlooking themulti-functions of land systems; and the other is that the effects of human decision-making related tostakeholders in agricultural land management were not addressed properly in the prior studies. Tosolve these problems, the study mainly used household survey data at a typical agricultural area ofNortheast China to explore the spatial-temporal changes and drivers of “agricultural land system” inrespect of land tenure, crop allocation and agricultural intensification. Then, the study tried toconceptualize an agent-based model for simulating crop pattern dynamics at the regional scale. Theconclusions and the innovative points of this study are summarized as follows:(1) A literature review suggests that ABM/LUCCs (Agent-based agricultural land changemodeling) bring theoretical and methodological innovations in land change modeling. They will helpto facilitate the integrated analysis benefited from both natural sciences and social sciences to form abetter understanding on the dynamics and complexity of agricultural land systems. Although there aresubstantial differences between models, the fundamental role of ABM/LUCCs is to express farmers’land use decisions and allocate them on regional level landscapes.(2) The survey shows that land transfer was fairly common across the study area: farmlandacreage per household is almost doubled from an average of1.3ha by early1980s to2.6ha by early2010s. It also indicates an increase in land transfers over time with a sharp decrease of the averageperiod of land transfer contracts. Crop choice displays a trend of decreasing diversity as several cerealcrops such as wheat, sorghum, and millet are no longer grown in the study region and the vastmajority of the beans area has been replaced by maize and tobacco since the early1980s. Landtransfers may be a cause for the increase of the dominance of a small number of crops at the samplelevel, but are not the main driver for changes in cropping structure at the region level. Irrigationintensity is related to the locations of rivers while agricultural inputs, along with land transfer andcrop allocation, show a spatial pattern which is related to the spatial variation in road accessibility andeconomic conditions.(3) Farmer’s attitudes in specific land use decisions were analyzed using the point-scoringmethod, and binary-logit regression models were further used to explain driving forces for landsystem changes. Most factors with significant coefficients in logit results have the higher score in total, indicating that most of the factors perceived as important also play a role in actual land changedecisions at individual level. Although farmers have different attitudes towards agricultural decisions,two family characteristics (education level and the initially allocated land rights) and twosocioeconomic factors (infrastructure and crop prices) proved to be most important in makingdecisions on land transfer, while a number of external factors have substantially influenced theirdecisions on crop choice.(4) The CroPaDy model is developed based on the above achievements. The conceptual model is aclosed-loop comprised by driving forces, decision making processes, and consequences. Farmers’attitudes are the determining mechanism to decision making. The computational model links threesub-models named agents generating module, agent classifying module, and agent decision-makingmodule respectively. Common methods including Monte Carlo, Clustering, Artificial Neural Network,and Probabilistic Approach are used in model parameterization. The CroPaDy model is conceptualizedstrictly according to the ODD Protocol proposed by Grimm et al.(2010) and the GeneralizedFramework for Parameterization of ABM proposed by Smajgl et al.(2011)(5) The main program of CroPaDy model was coded in MATLAB, then the output of matrixcalculation was presented and spatially analyzed in ArcGIS environment. The model results suggestthat land transfer rate in the study area is51.85%and58.90%in2010s and2015s respectively. AndXinDian town gets a relatively high transfer rate than the rest two towns, which is accordant with theprevious characteristic analysis. The crop area of maize, rice, soybean, and tobacco are26055.9、3506.75、5192.2、3983.85ha respectively. When comparing them with the local statistic yearbook, theoverall accuracy of CroPaDy model can reach as high as90%.