This thesis examines the predictability of precipitation variability and the role of land-atmosphere interactions in atmospheric general circulation models (AGCMs). In particular, it considers how and to what degree the contribution of land surface conditions to weather evolution and seasonal prediction can be improved by the prescription of land surface conditions.

　In the first chapter, the author described the background and purpose of this study. The inherent behavior of the atmosphere and its internal variability from its nonlinear and unstable characteristics are chaotic. Therefore, prediction accuracy is limited even under perfect initial conditions. The first person to estimate the limitations of prediction was Thompson [1957]. Lorenz [1963] suggested that a non-periodic evolution occurs in three types of simultaneous, and ordinary differential equations, even if initial conditions have subtle differences. Because of the chaotic behavior of the atmosphere, the use of individual forecasts of instantaneous weather patterns is constrained to about 10 days or less [Lorenz 1982]. Thus, a deterministic numerical forecast with a single atmospheric initial condition may have limited value for prediction. An ensemble forecast consisting of a number of individual forecasts based on a set of simulations, in which each simulation is slightly different, includes several initial conditions, and uses values that have been slightly perturbed, can gauge and reduce the numerical prediction errors that arise from chaotic behavior. Lorenz's famous statement, "Does the flap of a butterfly's wings in Brazil set off a tornado in Texas?", made at a conference of the Global Atmospheric Research Program (GARP), reflects the problem of atmospheric chaotic behavior. Predictability has been estimated by some distinguished researchers, e.g., Charney [1966], Smagorinsky [1969], and Epstein [1969] in different ways. All of their estimations suggested that predictability is only reliable within one to two weeks. Contrary to these pessimistic results concerning predictability (predictability in the first kind), Lorenz [1982] showed the existence of another aspect of predictability and demonstrated that seasonal-scale predictability could be improved if atmospheric boundary conditions (e.g., sea surface temperature, snow cover, and soil moisture) are predicted. The potential predictability (or theoretical limit of predictability) was investigated at the seasonal scale using statistical approaches based on the assumption that the model is perfect and the lower boundary conditions are predicted correctly [e.g., Charney and Shukla, 1981].

　In the second chapter, the author derived the mathematical structure of a similarity index known as Ω diagnostics, which can be calculated by a simple expression. The author also revealed its statistical characteristics; the Ω diagnostics quantify the similarity among several time series of ensemble members from two aspects of "phase (correlation)"and "shape (mean value and amplitude)". Furthermore, the author introduced two new statistical indices, which quantity the similarity in phase and shape, as singular solutions of Ω diagnostics. One is the average value of the anomaly cross correlation coefficient (ACCC) and the other is the average value of the variance ratio (AVR) against ensemble members. Present medium-range and seasonal weather forecasting are performed by ensemble simulations, which have subtle differences in initial conditions. Skill in ensemble weather forecasts is typically estimated with anomaly correlations or root mean square (RMS) differences. Even if high predictability of shape is estimated with the RMS difference, such accuracy may not be practical or reliable in spite of small predictability for phase similarity. The converse is also true. In the predictability studies originated by Charney and Shukla [1982], ensemble simulation results considered the influence of lower boundary conditions (e.g., sea surface temperature, soil moisture) on atmospheric variability by statistical approaches. When all three types of similarity indices are used, we can categorize the predictability of phase and shape taken individually or in combination.

　In the third chapter, the author considered the global distribution of land-atmosphere coupling strength (CS) for precipitation variability in the boreal summer. Many sections of this chapter are related to the Global Energy and Water Cycle Experiment (GEWEX) Global Land-Atmosphere Coupling Experiment (GLACE). GLACE involves highly controlled numerical simulations of the boreal summer using 12 climate models to show the global distribution of land-atmosphere CS. The author participated in GLACE using AGCM version 5.6 of the Center for Climate System Research/National Institute for Environmental Studies (CCSR/NIES). GLACE results showed that large land areas of subsurface soil moisture (called "hot spots") impact precipitation variability; the large Ω similarity index was found in the Great Plains of North America, the Sahel, equatorial Africa, and India. The CCSR/NIES AGCM also revealed strong CS over the Great Plains of North America, central Eurasia, and eastern India. CS exhibited large model dependency. However, each model revealed hot spots in semiarid land areas. Including such hot spots in global initializations of soil moisture may enhance precipitation prediction over semiarid regions in boreal summer. The CS, however, cannot reveal locations with long soil moisture memory. The author investigated the global field of soil moisture memory in the boreal summer using three types of similarity indices; Ω, ACCC, and AVR. Analyses with Ω diagnostics revealed long soil moisture memory in some tropical regions, deserts, and semiarid regions, including the Great Plains of North America and parts of central Eurasia. Both of these semiarid regions have strong land-atmosphere CS and long soil moisture memory; thus, precipitation variability could be improved over these regions in the boreal summer by monitoring subsurface soil moisture.

　In the fourth chapter, the land-atmosphere CS for precipitation variability was associated with the prescription of all land surface prognostic variables. In this chapter, CS shows a potential limit of predictability associated with land surface boundary conditions. Strong CS for precipitation variability appeared in many regions of North America, especially in the central area, the central region of Eurasia, and Southeast Asia. This chapter discusses constraint factors to define the geographical distribution of the land-atmosphere CS for precipitation variability. In addition, hot spots are found in semiarid regions, where the 6-day variability in latent heat (LH) flux is large. The diagnostic, CS, shows land surface impacts from the viewpoint of sensitivity. LH flux variability, in contrast, indicates land surface impact as the signal. Thus, both large CS for LH flux and large LH flux variability are required for hot spots to occur. Furthermore, large sensible heat (SH) flux is found over the hot spots. The vertical structure of land-atmosphere CS was also focused in this chapter. CS largely decreases within the atmospheric boundary layer (ABL). Large SH flux induces atmospheric instability and creates conditions suitable for conveying evaporated water vapor into the upper level of the ABL. Thus, the author suggests that both large SH flux and large variability in LH flux are necessary for high CS for precipitation. This could be the mechanism determining the locality of land-atmosphere CS. Finally, the locality of land-atmosphere CS (local CS) for precipitation was quantified using another type of land surface prescription experiment. In this experiment, all land surface prognostic variables were prescribed in the hot spots. The results showed that the spatial correlation coefficient between the CS for precipitation and the local CS was about 0.7. In the hot spots, about 60% of the land surface impact was determined locally.

　In the fifth chapter, the author discussed the seasonal variation in land-atmosphere CS for precipitation variability associated with land surface conditions. In this chapter, the author estimated the seasonal variation in CS throughout the year and discussed the relationship between local evaporation characteristics and CS. Large seasonal variation in CS decreased from fall to winter, despite strong CS in summer. However, land-atmosphere CS for precipitation was correlated with local evaporation characteristics throughout the year over central North America (CNA). For Indochina, large CS was estimated for the transition periods between the rainy and dry seasons. Over Indochina and southeastern China (SEC), seasonal variation in CS was not correlated with local evaporation characteristics, but it was relatively well correlated with the CS for 850 hPa geopotential height at the annual scale. Thus, land-atmosphere CS for precipitation is affected not only by direct evaporation characteristics, but also by the atmospheric field, which was already impacted by the land surface and could indirectly create high CS from remote impacts to the land surface.

　The sixth chapter examined the predictability of the Asian summer monsoon using several sets of 16 simulations with prescribed sea surface temperatures (SSTs) and land surface prognostic variables. Predictability was estimated with the Ω statistical index, which quantified similarity among ensemble members. The model simulations show that the predictabilities of precipitation and low-level moisture flux during May and June are much larger than those in July and August. The role of land-atmosphere interactions was also examined. During May and June, the predictabilities of soil moisture and near surface temperature increased with higher predictability of precipitation. At the same time, the predictability of cumulus-cloud-type precipitation could be improved or maintained by positive feedback from the high predictability of soil moisture and near surface temperature, which was associated with atmospheric instability in the ABL. The land-atmosphere CS and the role of the Tibetan Plateau were also investigated. The results suggest that land surface conditions of the Tibetan Plateau may contribute to the predictability of air temperature at the 850 hPa height in June from the Tibetan Plateau to Japan, a region of East Asia including large areas of China and Korea. Increased predictability of air temperature at this height relates to the accuracy of atmospheric condensation temperature predictability and leads indirectly to enhanced precipitation predictability. This result suggests the possibility of well-connected predictability associated with the prescription of land surface conditions.

　In seventh chapter, three sets of ensemble experiments were carried out to investigate the predictability and the reproducibility of precipitation and near surface variables variability associated with the semi-observed subsurface soil moisture. One of the target regions is East Asia where the Meiyu (Baiu) frontal activity produces much rainfall in the boreal summer. The author conducted three sets of ensemble experiments with both T239 (50km grid cell) and T39 (300km grid cell) horizontal scales, and discussed the horizontal resolution dependency concerning the predictability associated with the subsurface soil moisture. The author also examined the relationship between the predictability and the reproducibility of near surface variables. As the results of analyses, a model simulation using the semi-observed subsurface soil moisture improved the reproducibility of the surface soil wetness, 2m height air temperature, and the sea level pressure field. In particular, large predictability and reproducibility were shown over China, and these results suggest that the initialization of subsurface soil moisture could be one of the contributors to improve the seasonal prediction of at least near surface variables.

　陸面水循環過程は気候システムにおける下部境界条件を与え、海洋ほどではないにせよ、大気乱流よりも変動の時間スケールが長いため、陸面水循環過程の予測精度の向上は季節スケールの降水量変動の予測可能性の精度を向上させる可能性があるとして、過去20年以上にわたって大気陸面相互作用の物理的な研究、野外における陸面と大気との間の運動量やエネルギー、水フラックスのやりとりなどが観測されたりしてきた。

　本論文は、大気大循環モデル(AGCM)を用いて、季節スケールの降水量変動の予測精度向上に対して陸面境界条件がどういう場所でどういう季節にどの程度寄与しうるか、その潜在的可能性を明らかにしたものである。

　第1章では背景と本論文の目的が述べられ、大気の非線形性と不安定性によりその挙動はいわゆるカオスであり、初期値のわずかな差異が時間発展によって大きな差に発展するため、通常の天気予報は10日から14日程度が限界であるとされていることなどが紹介されている。これに対し、わずかに異なった初期値を用いた複数の予報結果を統計的に利用するのがアンサンブル予報であり、本論文もこれに準じた手法を用いて、陸面境界条件が大気のカオス的な挙動のばらつきをどの程度抑えるかを明らかにしている。

　第2章では、Ωとして知られるアンサンブル予報間の相似度を示す指標の数学的な意味づけに関して詳細な解析がなされている。その結果、従来はアンサンブル間の時系列の相似度を0〜1で示す簡便な指標としてテータ処理的にしか捉えられていなかったΩが、時系列同士の位相の類似度と振幅や平均値といった強度の類似度からなるという数学的な位置づけが明らかにされている。さらに、アンサンブル平均値からのアノマリー同士の相互相関係数を平均した値であるACCCと、他のアンサンブルメンバーとの分散の比の平均であるAVRがΩの数学的解析に基づいて提案され、現時点では通常アノマリー同士の平方平均分散誤差(RMS)で表現されているアンサンブルメンバー間の類似度を、より適切に指標化することが可能であることが述べられている。

　第3章では、GLACEと呼ばれるグローバルな大気陸面相互作用強度(CS)を様々なAGCMによって推定する国際研究プロジェクトに参加した結果がまとめられている。GLACEでは、北半球夏に関して、16アンサンブルメンバーによる海面水温を与えた数値実験(コントロール実験)と、コントロール実験のうちのひとつのアンサンブルメンバーを真値だとみなし、その土壌水分条件を海面水温と同様境界条件として毎時間ステップごとに与えた数値実験(陸面固定実験)とにおいて、Ω指標がどの程度異なるかにより、CSの強さを推定する、ということが12のAGCMによって行われた。その結果、北アメリカのグレートプレーンズ、アフリカのサヘル、赤道アフリカ、そしてインドなどでCSが大きいhot spotと呼ばれる地域があることが明らかになり、特に半乾燥条件の領域でCSが大きい傾向が示された。

　第4章では、土壌水分のみならず、全陸面境界条件を固定した数値実験により、どういう陸面パラメータがどの程度降水量変動の予測可能性に関係しているかが調べられた。その結果、6日平均降水量の季節スケールの変動に対しては、潜熱フラックスの変動が大きい地域でCSが大きくなることが明らかとなっている。さらにその鉛直一次元的な構造に関しても地表面から大気中層までの気圧や風速等に関するCSを算定することによって解析され、特に雲底より上の大気下層の非断熱加熱が大きい高度の気温場に陸面条件を固定する影響が伝わっていること、しかしながら、水平スケールが小さい対流性降雨によって主にAGCM中では降水量変動に陸面条件が影響を及ぼしているので、風速場への影響は限定的であることなどが示されている。対流性降雨の変動を通じて陸面が鉛直1次元的に大気循環の変動に影響を与えていることは、hot spotとして特定された領域の条件のみを固定した数値実験によっても示されている。

　第5章ではCSの季節性が調べられ、北アメリカとインドシナではCSが強まるメカニズムがやや異なり、1次元だけでない陸面の大気への影響が示唆されている。

　第6章では、アジアモンスーンによる降水量変動の予測可能性に焦点をあてた数値実験の結果が示され、陸面状態が適切に与えられることにより予測精度が向上する可能性があることが示されている。

　第7章では、アンサンブルメンバーではなく観測に近いとみなされる土壌水分条件を与えた場合に地上気温や気圧の予測精度が向上することが示されている。

　この様に、大気大循環モデル(AGCM)をまさに数値実験の道具として用い、陸面条件を所与とすることが降水量変動の予測精度向上にいかに資する可能性があるかの研究や、その評価指標の数学的位置づけと新たな指標の提案、AGCMの中での陸面が大気循環と降水量変動に及ぼしているメカニズムを明らかにした点などは大気陸面相互作用と降水量予測可能性の研究に新たな境地を切り開いており、グローバルな水循環科学としての水文学、その応用としての水資源工学への貢献は極めて大きい。

　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。