Climate-induced Land Use Change in France: Impacts of Agricultural Adaptation and Climate Change Mitigation

Interaction between mitigation and adaptation is a key question for the design of climate policies. In this paper, we study how land use adaptation to climate change impacts land use competition in the agriculture, forest and other land use (AFOLU) sector and how a mitigation policy in agriculture might affect this competition. We use for this purpose two sector-specific bio-economic models of agriculture and forest combined with an econometric land use shares model to simulate the impacts of two climate change scenarios (A2 and B1, 2100 horizon), and a greenhouse gas emissions from agriculture policy consisting of a tax of between 0 and 200 €/tCO2 equivalent. Our results show that both climate change scenarios lead to an increase in the area devoted to agriculture at the expense of forest which could have a negative impact on reducing greenhouse gas emissions responsible for climate change. The mitigation policy would curtail agricultural expansion, and thus could counteract the effects of land use adaptation to climate change. In other words, accounting for land use competition results in a reduction of the abatement costs of the mitigation policy in the agricultural sector.


Introduction
According to the International Panel on Climate Change (IPCC) (2013), the average 1 global temperature has increased by about 0.85 • C during the period between 1880 to 2012.  However, their model does not consider other land-demanding economic sectors or their 71 future evolution. In contrast, our methodology allows for LUC not only among sectors 72 but also within the agricultural and forestry sectors (choice of crops and/or pasture, and 73 choice of tree species). This aspect is fundamental when considering CC adaptations. 74 Third, the marginal abatement costs of GHG for agriculture have been studied using 75 different modeling techniques. In a meta-analysis, [77] classify the different approaches 76 according to three groups: i) supply-side models specialized in agriculture [e.g. 29, 28, 38]; With the exception of general equilibrium models, the responses of farmers to GHG 85 taxation in terms of land use is ignored in previous work. Since land use feedback effects 86 have been shown to be important in the context of GHG mitigation policies such as incen-87 tives for using biofuels [74], in our simulations we account explicitly for LUC. Finally, [57] 88 estimate an econometric land use model for the USA and simulate landowner responses to 89 sequestration policies. They examine a two-part policy involving a subsidy for converting 90 land to forest, and a tax on converting land from forest. They then estimate the carbon 91 sequestration supply function of these policies by computing the corresponding flows of 92 carbon in terrestrial sinks. However, unlike our study, they do not simulate the impacts 93 of climate change on land use. 94 The present paper addresses three main questions: To investigate these questions we exploit the results from two mathematical program-100 ming models (AROPAj for agriculture and FFSM++ for forestry) to study the impacts 101 of CC on agricultural and forest rents. We use the supply model AROPAj to study 102 the impacts of a mitigation policy (tax on GHG) on agriculture, and we use a spatial The article is organized as follows. In section 2, we describe the models used to assess 112 GHG emissions from agriculture, and section 3 presents the data. Section 4 presents and 113 discusses the results of our simulations. 114

115
The study methodology is based on two mathematical programming models (AROPAj 116 for agriculture, and FFSM++ for forestry), coupled to bio-ecological models, and a spatial 117 econometric land use model that allows us to combine the results of the sector-specific 118 models. Figure 1 describes the modeling scheme adopted. The bio-ecological components   The type of production (type of farming) and the economic size are defined in the sense of FADN (http://ec.europa.eu/agriculture/rica/diffusion_en.cfm). For instance, farm type 35 in the Rhône-Alpes region is located at low altitude (< 300 m), the economic size of its composing farms is mostly superior to 25,000 e/year, and its activities are oriented mainly towards field crops. In the baseline case, its land is used mostly for maize (31%), wheat (30%), sunflower (14%) while only a small part of its area is devoted to pastures (5%). 6 Following the duality theorem, the shadow price provides an estimate of the marginal profitability of land, or in other words, its rent (under the economic equilibrium hypothesis).    In our simulations, we account only for CC and do not integrate any changes in produc-178 7 The possibility for conversion is partially limited by some technological constraints imposed during the calibration of the AROPAj model which avoids corner solutions to the model (mono-cropping). Also, the number of animals can vary within a ± 15% interval, otherwise, the model would be out of its calibration interval. However, the choice of animal feed (grazing or fodder) is free. tion technology (apart from adaptations such as changes to sowing dates, crop varieties, and fertilizer use). Some complementary information related to the CC scenarios' data 180 are provided in subsection 3.3 and in appendix A.

181
AROPAj models the farmer's choice between land uses in terms of crops and/or pas-182 ture. Farmers can choose also among different animal feedstuffs 8 which has an impact on 183 GHG emissions. We simulate GHG tax levels from 0 to 200 e/tCO 2 eq; these taxes reduce 184 the profitability of agriculture (ceteris paribus, no price feedback is considered).
where R g is a vector of land use rents, β R l is the associated vector of the parameters 220 to be estimated; P g is a vector of the physical parameters (soil characteristics and slope) 221 and β S l is the vector associated to the parameters to be estimated.

222
Linearizing the model in equation 1 allows us to estimate equation 2 with a reference 223 land use, L 224S gl = ln(S gl /S gL ) = R g β R l + P g β P l + u lg , ∀g = 1, ..., G, ∀l = 1, ..., L In the context of aggregated land use share models, spatial autocorrelation could result 225 from a structural spatial relationship among the values of the dependent variable, or a 226 spatial autocorrelation among the error terms. In the present study, we use an 8 km x 8 km 227 continuous grid which corresponds to the French climate data grid system, SAFRAN 9 .  lang=en . 10 See [49] which provide motivations for regression models that include spatial autoregressive processes.
W 1 is an n × n spatial weight matrix for grid cell neighbors, W 2 is a m × m spatial 254 weight matrix for regional neighbors, R g and P g are the fine scale explanatory variables 255 for neighboring cells, R j are regional scale variables for neighboring regions, β R l , β S l , and 256 β R l are the associated parameters, the parameter λ expresses the interaction between 257 residuals and ε is an iid 11 error term such that ε ∼ iid(0, σ 2 I).  Table 1.   figure 3 for the underlying hypothesis).

302
The agricultural sector simulations exploit two sets of climate data were used. For   -moderate economic growth -low economic growth -low demographic growth -average demographic growth -environmental sustainability -environmental sustainability -increase in temp. 1.1 -2.9 • C -increase in temp. 1.4 -3.8 • C Figure 3: Summary of the four major climate change scenarios as presented in [43] soil texture on four levels. Level 1, the lowest quality, is the reference. Land quality is an    Impacts of CC adaptation on land use.. Figure 6 shows that our land use model predicts 363 an increase in crop area under the two CC scenarios compared to current climate (CTL 364 scenario). Figure 6 shows also that the increase in the area to crops is more important 365 in B1 scenario, than in the A2 scenario. This increase is at the expenses of forest and    shows also that if we take account of the potential LUC due to a GHG tax, the reduction 392 in GHG is greater than if we consider the agricultural area as remaining constant. These 393 differences are more important for GHG tax levels higher than 50 e/tCO 2 eq. Compared      90 e/tCO 2 eq 120 e/tCO 2 eq B1 90 e/tCO 2 eq 130 e/tCO 2 eq Table 4: Abatement costs (in e/tCO 2 eq.) allowing 12% decrease in agricultural GHG emissions with or without accounting for LUC

Conclusion and perspectives 428
In the present study, we analyze the impacts of climate change adaptation and a 429 mitigation policy on land use changes in France. We used for this purpose two sector-

449
In addition, some potentially "win-win" measures (such as the "4 per 1000" program) could 450 increase abatement rates, and improve soil quality, and thus agricultural productivity.

451
Our results show that the targeted emissions cut for French agriculture is achievable 452 at a tax level close to the carbon price associated to energy CO 2 emissions (100 e/tCO 2 ).           Note: * p<0.1; * * p<0.05; * * * p<0.01 Table 12: Spatialized dual value, 4 LU, 3 rd order W 1 , 2 nd order W 2