A major research lesson over the past decades has been that sustainable environmental management cannot be achieved by focusing only on either the human or the environmental system. Understanding dynamic interactions between both systems is imperative. Emphasis has therefore progressively shifted to integrated approaches based on the concept of coupled human-environment (H-E) systems.

In desertified regions, land degradation and human livelihoods are closely interrelated; this relationship can be altered by policy intervention. However, most desertification estimates have been solely made either by ecological or by socioeconomic factors but rarely by both. In particular, since desertification is a global problem but is also strongly localized phenomena, local heterogeneity of both people and landscape is a critical factor to jointly address. Thus, accounting for complex H-E systems in drylands becomes necessary for accomplishing sustainable land management.

Researchers and policy-makers agree that the modeling approach is fundamental for understanding H-E systems. Recently, an agent-based model (ABM) has gained considerable attention. ABM is a computerized bottom-up simulation of entire patterns emerging from the interactions between numerous autonomous decision-makers and a dynamic environment. Although traditional models capturing combined ecological and socioeconomic systems were based on mathematical bioeconomics, they could not handle many important characteristics of H-E systems, e.g., space, interactions, and heterogeneity. ABM has the potential to overcome the shortcomings of traditional models and thus has become a major tool for understanding complex systems. However, ABM is still in its developing phase. In particular, multiple interaction mechanisms such as feedback, learning, and adaptation in a real heterogeneous system are poorly represented.

From an H-E system's perspective, this thesis aims to empirically elucidate ecological and socioeconomic characteristics in a desertified region and develop a spatial agent-based land-use model representing a heterogeneous dryland system with multiple interaction mechanisms. The model's operational capability as a decision-support tool for future land management was tested using scenario assessments. The study area was located in Naiman County, Inner Mongolia Autonomous Region of China. It is situated in the southern part of the Horqin Sandy Land, one of the most desertified and poorest regions in Inner Mongolia. I considered the H-E systems of this area as three interrelated systems: landscape, household, and policy. First, I conducted three empirical studies on the landscape system for demonstrating the dynamics of land-degradation and land-restoration and spatial heterogeneity (Chapter 2). Second, I conducted two empirical studies on the household system to elucidate households' heterogeneity and land-use decision-making (Chapter 3). Third, I conducted an empirical study on the policy system to clarify the local mechanisms of a key environmental policy, called Sloping Land Conversion Program (SLCP), and designed hypothetical scenarios that could improve the current SLCP (Chapter 4). Finally, I integrated the results of these empirical studies using agent-based modeling and constructed the spatial land-use model, named Inner Mongolia Land-Use Dynamic Simulator (IM-LUDAS) (Chapter 5). Further, using IM-LUDAS and the designed scenarios, I assessed the effects of current and hypothetical SLCP on the land-use change and dynamics of households' livelihoods, land degradation, and land restoration.

Chapter 2 Landscape system

Land degradation in this area is mainly due to cultivation and grazing. Since pasture degradation has been intensively studied by earlier research, I conducted empirical research on cropland degradation.

The landscape of this area is characterized by sand-dune topography, a factor determining land conditions and use. I examined the changes in crop, weed, and soil properties in three typical cropland types having cultivation periods of up to 20 years: maize cropland on non-irrigated lowlands, maize cropland on irrigated flat sandy lands, and bean-centered cropland on non-irrigated sand dunes. The crop biomass and soil properties were more degraded in non-irrigated lowlands and non-irrigated dunes than in irrigated flatland. The weed communities in non-irrigated croplands became established in drier conditions, whereas wetland weeds were more abundant on irrigated flatland. The changes in property did not always occur simultaneously and differed statistically for different croplands.

I conducted vegetation and soil surveys along a topographic gradient (interdune lowland, lower, middle, and upper part of sand dune) in the following four types of restoration site, having restoration periods of up to 35 years, and in two control sites: grazing-exclusion sites; shrub, pine, and poplar plantations; fixed and shifting sand dunes. Vegetation and soil conditions were restored simultaneously. Tree plantations facilitated land restoration more rapidly than grazing exclusions and shrub plantations until the first 25 years; however, both tree plantations and grazing exclusion sites reached a stable state at the end of 35 years. Land restoration progressed more at lower positions than at higher positions; this indicated that the establishment and dispersal of invader plant species beginning at interdune lowlands would promote restoration.

I developed a method to spatially extract topographic and land-use types, survey units of ecological studies, for demonstrating spatial landscape heterogeneity and for enabling field-scale ecological models to function in IM-LUDAS. The method used a combination of object-oriented analysis, digital photogrammetry, and GIS analyses using images from Advanced Land Observation Satellite. First, object-oriented classification was performed to discriminate land-use types that were hardly differentiated from others using only spectral information. Second, a digital surface model (DSM) was extracted by digital photogrammetry; the landform types were classified on basis of DSM using GIS analyses. The land-use types were then extracted by overlaying two classification maps. I applied the method to a test site in the study area, and the overall accuracy of object-oriented and landform classification was high, i.e., 87% and 88%, respectively.

The clarifications of unique degradation and restoration patterns corresponding to spatially-explicit landscape heterogeneity enables the formation of uneven policies such as detailed zoning and also makes IM-LUDAS sensitive to uneven-policy scenarios.

Chapter 3 Household system

First, I examined regional socioeconomic heterogeneity to find representative villages where household surveys would be carried out. I interviewed village representatives about the socioeconomic conditions and classified all villages into three types: livestock-farming-oriented, crop-farming-oriented, and off-farming-oriented. Because off-farming-oriented villages did not largely depend on land resources or undergo land degradation, other types seemed to need prioritized support. Next, I conducted household surveys in five villages selected as representatives of the prioritized village types. From the collected households' socioeconomic data, their livelihoods were classified into three types: livestock-farming-oriented, crop-farming-oriented, and off-farming-oriented. These were similar to the village classification types; this implied that the main economic activities would be the primary indicator to characterize socioeconomic heterogeneity, regardless of scales, in this area.

From spatially-explicit landholding data of households, collected in household surveys, I built the land-use decision-making model for each livelihood type using multinomial logistic regression. The result showed that the topography and proximity to used land significantly affected the land-use patterns of each livelihood type. However, these effects on land uses were positive or negative, depending on the livelihood types.

The identification of different livelihood types with distinct land-use patterns clarified the socioeconomic heterogeneity of households in this region. This identification also facilitated adaptive decision making to be embedded in IM-LUDAS; this decision making mechanism enabled households to change their livelihood types by changing their socioeconomic conditions.

Chapter 4 Policy system

Three hierarchical actor levels were involved in SLCP: township government, village committees (self-governing bodies), and households. I interviewed the senior officials of the township government, village, and household representatives to understand the SLCP local mechanisms. The results showed that the target villages of SLCP were determined by voluntary village applications and government screening. This process spontaneously considered regional heterogeneity; i.e., SLCP was not uniformly executed across the township. Households hardly participated in the implementation process as autonomous actors. In researched villages, although the SLCP participation was not compulsory, all the villagers were given uniform land quotas for SLCP and they actually attended it. Moreover, even though households were given the option to choose implementation plots, they were encouraged to choose flatland because flatland was suitable for cultivation as it could be irrigated; thus, the planted trees could be easily preserved. The government aimed to increase the plantation area, and the households were required to preserve the planted trees in order to receive the SLCP subsidy. Therefore, a major shortcoming of the present SLCP was the lack of proper prioritization in identifying target plots and beneficiaries, which raises doubts in the program's cost-effectiveness. Further, even if target villages, households, and plots were appropriately selected, certain proactive measures would be essential for the prevention of desertification because the planting of trees was a reactive measure. The Chinese government expects the SLCP to facilitate the households' labor reallocation, particularly from agriculture to non-agriculture, which would contribute to stable livelihoods after the implementation of SLCP. This SLCP effect could be considered as a proactive measure against desertification; however, it is still not clear whether SLCP has the potential to achieve this effect.

In addition to a baseline scenario designed on the basis of the current SLCP, I designed two hypothetical SLCP scenarios with reference to the two abovementioned challenges on the current SLCP: rational SLCP scenario, introducing rational criteria for selecting implementation plots, i.e., only targeting the cropland abandoned by households below average income in this region because of unacceptably low yield; intensive SLCP scenario, introducing intensive criteria for selecting implementation plots, i.e., targeting all croplands used by households below average income in this region. The former scenario aimed to assess whether the cost-effectiveness of the SLCP could be improved in terms of both ecological and economic benefits, whereas the latter aimed to assess whether households' labor reallocation could be induced by SLCP.

Chapter 5 Agent-based modeling of a coupled human-environment system

I constructed IM-LUDAS using empirical studies results. The model's spatial extent was confined to the boundaries of the five selected villages (Chapter 3); the landscape data of these villages were created by the developed method (Chapter 2). Landscape agents (congruent land pixels), having ecological attributes and sub-models of crop yield, land degradation, and land restoration (Chapter 2), and household agents, having socioeconomic attributes and land-use decision sub-model (Chapter 3), constitute the IM-LUDAS. Default settings of IM-LUDAS are based on the current SLCP (Chapter 4). The agent's attributes, functions, and interaction mechanisms are related to external policy settings, so that the mutual ecological and socioeconomic impacts of SLCP can be assessed. Because IM-LUDAS represents landscape heterogeneity and household population, the effects of various implementation mechanisms, e.g., designed scenarios (Chapter 4), on the program's cost-effectiveness can be quantitatively tested. Moreover, the effectiveness of the SLCP in changing the households' labor allocation can be assessed using the embedded adaptive decision-making mechanism.

Using IM-LUDAS, I assessed the current and hypothetical SLCP impacts according to three designed scenarios. As expected, the ecological benefit was higher and expected budget was lower in the rational SLCP scenario than in the baseline scenario. However, poverty alleviation effects were not observed probably because the resultant total implementation area was very small; this means that the paid subsidy was too low to contribute to poverty alleviation. The intensive SLCP scenario did not induce the reallocation of the households' labor, whereas their land-use structure and income level largely changed. This is probably owing to other unchanged critical attributes such as ethnicity and education levels. The livelihood types are determined by their multiple attributes including those mentioned above. This result implies that the use of only SLCP is not enough to induce livelihood change; other policies, e.g., educational programs, should be implemented.

This thesis empirically elucidated the heterogeneous landscape and its dynamics as well as heterogeneous households and their decision-making in a desertified region. Moreover, their complex interactions, including adaptive decision-making, were represented by agent-based land-use modeling. IM-LUDAS is one of the first models that can assess the mutual impacts of external factors on the change in land-use along with the dynamics of households' livelihoods, land degradation, and land restoration with reference to landscape heterogeneity and human population. The scenario assessments generated rational and new insights for future policies, indicating its usefulness as a decision-support tool. Although many features of relevant H-E systems in this region are not included in the model yet, its agent-based structure has a built-in flexibility for upgrading and modification. As a virtual computational laboratory, IM-LUDAS can contribute to scientific experiments on various complex phenomena in desertified regions and can support negotiation among multiple stakeholders for future land management in this area.

本研究は、人間―環境相互作用系の観点から、砂漠化地域における異質な生態・社会経済特性を実証的に明らかにし、それらの相互メカニズムを再現できる空間明示的な意思決定支援システムを開発することを目的とした。研究対象地は、中国内蒙古自治区のホルチン砂地と呼ばれる半農半牧地域であり、内蒙古自治区の中で最も砂漠化と貧困が深刻な地域の一つとされている。この地域の人間―環境系をランドスケープ、農家、政策という三つのサブシステムの相互作用系としてとらえた上で、以下の手順によって研究を実施した。(1)ランドスケープシステムに関する実証研究を実施し、土地の荒廃と回復のメカニズムおよび土地の空間的異質性を明らかにした。(2)農家システムに関する実証研究を実施し、農家の社会経済状態および土地利用の意思決定における異質性を明らかにした。(3)政策システムに関する実証研究を行い、環境改善と貧困削減を目的とした、現在最も影響力の強い環境政策である退耕還林の実施プロセスを明らかにした上で、現行の退耕還林を改善するための代替政策シナリオを策定した。(4)以上の実証研究の成果をエージェントモデルによって統合し、空間明示的な意思決定支援システム(IM-LUDAS)、を開発した。さらに、代替政策シナリオを用いたシナリオ分析から、土地の荒廃と回復、農家の社会経済状態、そして政策実施の相互作用を明らかにすることで、IM-LUDASの有用性を示した。

まず、対象地域特有の砂丘地形タイプと土地利用タイプの組み合わせごとに植生・土壌・作物生産性の劣化、または植生・土壌の回復の動態を、フィールド調査と統計解析により調べた。その結果、植生・土壌・作物生産性の劣化パターンは必ずしも一致せず、さらに地形タイプと土地利用タイプの組み合わせによっても異なることが明らかになった。一方、植生と土壌の回復パターンは類似しており、砂丘下部の平砂地から砂丘上部に向かう地形傾度に沿って回復が進行することが明らかになった。

空間明示的な農家調査から、まず農家における社会経済的異質性を主成分分析とクラスター分析により調べ、農家タイプの類型化を行った。続いて、上記ランドスケープデータと空間明示的な農家属性データを結合して多項ロジスティック回帰を行い、異なる農家タイプごとの土地利用選択について調べた。その結果、主要経済活動の違いが強く反映された三つの農家タイプが抽出され、全ての農家タイプの土地利用選択において、使用地への近接性と使用地の地形タイプが重要な因子として選択された。しかし、各土地利用タイプに対するそれらの因子の係数の符号は農家タイプごとに異なっており、意思決定に違いがあることが明らかになった。

退耕還林の関連主体、その管轄化にある村落の各自治組織、および実際に緑化を行う農家への階層的な聞き取り調査を実施し、現行の政策実施プロセスを調べた。その結果、助成金の対象、植林地化の対象ともに効果的に選ばれていない現状が明らかになった。そこで、現状維持をベースラインとして、(1)環境および経済効果が改善されるか評価するための効果改善シナリオ、(2)経済構造の転換を誘発しうるかどうか評価するための構造変化シナリオ、という二つの代替シナリオを策定した。

以上の実証研究の成果をエージェントモデルにより統合し、IM-LUDASを構築し、退耕還林の相互的な生態・社会経済効果を評価するため、エージェント間の相互メカニズムを外部要因である政策パラメータ(政策シナリオ)と関連付けた。退耕還林による経済構造への影響を評価するため、各農家の主要経済活動を示す農家タイプがシミュレーション中に変化する適応メカニズムを組み込んだ。シナリオ分析の結果、効果改善シナリオでは、植林地の総面積と生態環境および経済環境は変化しなかった。これは、対象の選択基準が厳しく、実施規模が限定的だったためと考えられた。一方、構造変化シナリオでは、経済構造の変化が十分に誘発されず、長期的には生態・経済環境共に悪化する結果となった。これは、退耕還林によって収入構造のみ大きく変化させたとしても、経済構造までは転換し得ないことを示唆していると考えられた。さらにその場合、代替収入源をもたないまま助成金が終了することで収入が落ち込み、さらに植林地の拡大に伴い周辺放牧地が縮小し家畜密度が高まることで徐々に土地の荒廃も進む、という可能性が示唆された。

本研究は、半乾燥地域におけるランドスケープとその動態および農家とその意思決定の異質性を実証的に明らかにし、それらの複雑な相互メカニズムを表現することのできる空間明示的な意思決定支援システム(IM-LUDAS)を開発したものであり、砂漠化研究分野における先駆的な成果として高く評価できる。現状のシステムでは、水環境や気候といったいくつかの重要な要素を考慮できていないものの、エージェントベースのシステム構造により、追加・更新・修正を柔軟に行うことができる。今後、さらに個別要素の実証研究を進めて行くことで、より包括的なシステムに発展させることが可能であり、本研究は今後の実証研究のための統合プラットフォームを提供したということができる。よって審査委員一同は、博士(農学)の学位を与えるに十分値する論文であると判断した。