Floods and droughts have been affecting human beings since ancient times. In recent decades flood events have occurred at higher frequency and magnitude, particularly in humid regions due mainly to a changing climate and human activities.

　Recent extreme flood events have been causing enormous disasters; therefore, the need to reduce them is latent and growing. More than ever the ability to efficiently forecast and manage these events plays a key role in protecting human lives and avoiding other material damages.

　An extreme event can be substantially reduced by an optimal dam operation assuming that its capacity is sufficient to attenuate floods and store water for future usage. However, one should take into account that complex river systems need more than one operating reservoir, not only used for flood control purposes, but also for irrigation, water supply, power generation and so on. Then, water release from dams should consider reducing flood damage and storing water at the reservoirs for future water use. In this way, an efficient water resources management can be achieved by an optimal dam operation.

　In order to make appropriate dam water release decisions, the potential heavy rainfall needs to be predicted and the river discharge within the whole basin and dams' situation needs to be forecasted. Then, a reference system which is able to take maximum advantage of available forecast information and optimize water release from dams is highly expected by dam operators and community in general. The present study attempts to contribute in this last urgent requirement by coupling a distributed hydrological model to an optimization scheme in order to suggest the dam release schedule.

　Conventional dam operation employs lumped models relying partially in the physics with uniform rainfall input and the water release schedule is performed independently for each dam. However, distributed hydrological models can capture rainfall patterns using spatially distributed rainfall. This study couples a physically based distributed hydrological model namely Geo-morphological Based Hydrological Model (GBHM) with embedded dam network operation to the Shuffled Complex Evolution (SCE) optimization algorithm. The optimization variables are the dam water releases. The simulated inflow to dams, updated reservoir status, and river discharge downstream are calculated by the GBHM with embedded dam operation and introduced to the SCE scheme. Normally, manual dam release is performed by dam operators according to expertise to reduce extreme floods downstream. This proposed scheme takes advantage of the SCE in order to evaluate different dam release sets automatically based on stochastic seeding considering the dam constraints and objective function. Therefore, two objective functions are proposed. The first one to minimize the overflow downstream where flood prevention is desired by releasing water before flood develops during heavy rainfall. The second one besides flood reduction also aims future water use by releasing only the amount of water which will be replenished by forecasted flood volume. Moreover, the coupled GBHM-SCE is able to run sequentially using spatial distributed rainfall forecast to optimize dam release while lead-time lasts and define next time step's initial condition using observed data.

　The system was applied to the upper Tone river basin which covers 3,300 km2 targeting the flood 22-24 August 2001 using observed data. First, it was simulated water release of a small dam in a sub-basin of 1200 km2. It was found that the simulated inflow to dam and over-all river discharge in closer agreement to observed discharge using radar rainfall products rather than point rain gauge data. By using radar rainfall, the operation of the small dam was successfully carried out by setting the release proportional to the water height. Then, in the same sub-basin, the coupled GBHM-SCE with virtual Yamba dam operation, nowadays under construction, was carried out for flood reduction downstream. Next, in the whole upper Tone river basin the coupled GBHM-SCE considered two key dams operation in the objective function for flood reduction downstream. The results showed the feasibility in computation time and efficiency of the objective function to reduce flood downstream when compared simulation using optimized release against using observed release. In addition, the effect of Yamba dam as a case scenario with optimized dam release was simulated at the outlet of the basin.

　The application in the whole upper Tone river using 18 hours lead-time rainfall forecast targeted the flood 9-12 July 2002. First, the optimal water release from the two key operating dams was obtained using three different series: 1-6, 7-12 and 13-18 since forecast was issued every 6 hours. The results showed that 1-6 series, opposite to expected, in lower agreement with observed rainfall values than the other two series. Then, three dams participated in the optimization process using the complete 18 hours lead-time with the integrated objective function. This was conceived in order to reduce floods downstream by releasing water as soon as a flood was forecasted downstream; however, the maximum water release was set to be comparable to the total forecasted flood volume downstream. Moreover, it was found that the water level in the reservoirs after the flood event remained close to initial ones suggesting the replenishment of water. Those dams with high initial water level showed the best performance since they were able to release water and store again; on the other hand, the third small dam with low initial water level ended up with higher water level. This process was carried out every 6 hours according to issuing interval forecast. Within this period the optimization procedure was completed providing the dam release schedule and updating initial condition using observed data for the next time step. Then, the forecasted dam release was tested combining with observed rainfall data in order to get the stream flow downstream. The latter was compared against the stream flow obtained using observed release and observed rainfall. In summary, optimized dam release schedules were suggested relying on weather forecast not only reducing flood damage downstream but replenishing water in reservoirs for future water uses such irrigation, power generation, etc. This is an important point in regions affected by typhoons where floods and droughts can be alternated from season to season.

　The system has demonstrated high efficiency which can be used as a reference tool in real-time operation. In this way, the present study shows feasibility of taking advantage of forecast information for social benefit considering not only for flood damage reduction, but also for future water use.

　近年，集中的な豪雨による被害が国内外で顕著となっており，また温暖化などによる気候変動下で豪雨派生頻度の増加が懸念されている．一方，環境への配慮や地域住民との合意形成の難しさから，洪水制御のためのダムなどの貯留施設の新設が滞っている．本論文は，このような自然科学的課題や社会からの要請に応えるため，現在利用可能な降雨予測情報を効果的に用いて，既存のダムの最適管理することによって，洪水軽減，水資源の効率的利用を図る手法を提案するものである．

　本論文では，まず河川流域の標高，地質，土地利用などの空間分布情報や降雨の時空間分布を効率的に利用できる分布型流出モデルを基本としてシステム開発を行なった．本研究で用いた分布型流出モデル(GBHM)は，降雨の表面流出，土壌中の鉛直浸透，地下水流出を表現できる物理的モデルである．本研究ではモデル中の多くのパラメータを同定するために，Shuffled Complex Evolution(SCE)法を組み合わせたパラメータの最適化手法を導入した．さらに，河道流出部分にダムによる流出制御システムを導入して，SCE法を組み合わせたダム操作パラメータ推定法を提案している．これらのシステムを利根川上流域の単一ダムを含む小河川流域，複数ダムを含む中河川流域に適用し，地上観測雨量でキャリブレーションされたレーダ観測雨量を入力として河川流出シミュレーションを行ったところ，良好な再現性が確認された．

　そこで，このダム操作機能を組み入れた分布型流出モデルに，現業の数値気象予報値を入力して，ダム下流の洪水流量を一定値以下に抑える条件，ダムの現在の貯留容量と最大貯水能力，洪水後の効果的な水利用などの目的を達成するための目的関数を設定して，その効用を最大化するためのダム操作の最適化を行なった．入力とする数値気象予報値は，気象庁から提供されているグリッドポイント値(GPV)であり，6時間ごとに更新される18時間先までの毎時降水量の予報値が0.125度グリッドで利用できる．そこで1-6時間予報値，7-12時間予報値，13-18時間予報値の3種類について，降水量の観測値と予測値との誤差の逆数を最適化の重みとして用いることによって，それぞれの予測値を入力として用いてダム操作の最適化を行なったところ，7-12時間および13-18時間の予報値を用いると，現行のダム操作に比較して5-6%の洪水ピーク流量の低減が図れることが明らかとなった．さらに，ダム貯水量を洪水イベント前の状況に戻せるようなダム操作の最適化により，本システムが洪水だけでなく水資源利用の効率化においても有用な手法であることを示した．

　以上，本研究は，河川流出の物理特性を踏まえた分布型流出モデルに予測降雨を効果的に用いることによってダム操作を最適化し，既存の施設によって洪水危険度を低減するものであり，水循環研究の科学的側面だけでなく，水災害軽減と社会的重要課題にも貢献するところが大きく，科学的，社会的有用性に富む独創的な研究成果と評価できる．よって本論文は博士(工学)の学位請求論文として合格と認められる．