The skill of operational weather forecasts has increased substantially over the last five decades. The improvement has taken place gradually and relatively steadily, driven by advances in scientific understanding of physical processes and rapidly increasing computational resource developments. Numerical Weather Prediction (NWP) is an initial-boundary value problem: given an estimate of the present state of the atmosphere and land surface, the model simulates (forecasts) the evolutions of land and atmosphere. Due to the chaotic and nonlinear nature of the atmospheric processes, numerical weather forecasts are more sensitive to the conditions that represent the present reality of land and atmosphere in the model. More and better observations that represent the complete nature of the current atmosphere and land surface will greatly improve our understanding and will enhance the forecasting capabilities of NWP models. Therefore, the first step in numerical weather forecasting is to collect the adequate information about land surface and atmosphere in real or near-real time.

The potential of remote sensing to monitor the Earth weather and climate system has been demonstrated over the years. The last decade has seen remarkable progress in exploring satellite observations, especially microwave measurements and the launching of new platforms (e.g., Terra, Aqua, and Advanced Land Observing Satellite (ALOS)). In microwave frequencies, many earth materials and atmospheric constituents exhibit a distinctive character for their electrical property called dielectric constant, which is more sensitive to the amount of water (e.g., at lower frequencies, dielectric constant for dry sand and for free water). This dielectric property allows for the quantitative estimation of moisture quantities such as soil moisture, vegetation water and snow water contents as well as atmospheric water vapor and cloud condensate. Low-frequency passive microwave sensors are uniquely suited for soil moisture measurements owing to their penetration capability through atmosphere, whereas higher frequencies contain both land and atmospheric information.

Land surface processes play an essential role in understanding and predicting both global water and energy budgets. Soil moisture has been the central focus for accurate land surface and atmospheric modeling because it controls surface water and energy fluxes and it consequently influences land-atmosphere interactions. In addition, it possesses a long and persistent memory in forcing atmosphere over land surface. Numerous sensitivity studies on NWP models have shown that the consideration of accurate soil moisture content can influence short- and medium- range forecasts, has strong coupling with precipitation in mid-continental region, and can improve the cloud convection processes.

However, the use of in-situ soil moisture information in NWP models is not practical owing to limitations in such datasets. Current practices to incorporate actual soil moisture conditions are based on proxy observations (2 m air temperature and humidity), but this method has several limitations and does not directly linked to the actual moisture conditions. Use of space--borne microwave observations is most promising owing to their frequent overpasses and wider coverage. In addition, the spatial resolutions of microwave sensors are compatible with NWP models requirements. On the other hand, the current satellites microwave measurements cannot provide complete space-time coverage and their observations are limited to a few centimeters of soil depth. To overcome these limitations, data assimilation methods are developed to merge the satellite surface information with Land Surface Model (LSM) outputs to produce spatially and temporally complete information of superior products such as profile information of soil moisture and temperature.

Although global or regional satellite-derived surface soil moisture datasets are readily available from this approach, knowledge about assimilating these datasets into NWP models is very limited and to date only very few studies have been reported. In addition, the proposed methods have several limitations and cannot be applied in operational near-real-time practices. Therefore, at present, land surface analysis is not considered in operational forecasting. With increasing satellite observations, there is an urgent need to enhance research activities on exploring new methods and techniques that can be feasibly adopted in real or near-real time applications. As a result, this research initially focused on introducing the satellite observed land surface heterogeneities in a NWP model using physically based model and data assimilation methods and to investigate the soil moisture influences on simulated land-atmosphere interactions and atmospheric structures.

Consequently, a system (LDAS-A) that couples a satellite based land data assimilation with an atmospheric model was developed to physically introduce land surface heterogeneities into a land-atmosphere coupled model. LDAS-A consists of a mesoscale atmospheric model (Advanced Regional Prediction System-ARPS) as an atmospheric driver, a land surface model (SiB2) that acts as a land surface driver for the atmospheric model and the model operator for land data assimilation system, a physically based and well described land surface microwave radiative transfer model as an observation operator and an ensemble Kalman Filter (EnKF) as a sequential assimilation algorithm. LDAS-A has adopted the concepts of sequential and on-line data assimilation, which directly integrates the satellite raw (level 1B) data and reinitializes the model with observed land surface conditions whenever the observations are available. Sequential and on-line assimilation strategy removes many restrictions reported in the previous studies. LDAS-A was implemented on a standardized interface (Coupler) that consists of a superstructure to effectively handle the coupling and exchanges of data between individual components of the system (i.e., atmospheric model, model operators and assimilation algorithms). To meet the computational requirements, the Coupler was designed to run on a parallel computing platform. All these features make the system feasible for operational near-real-time NWP applications.

LDAS-A was validated on a mesoscale domain in the western Tibetan Plateau using surface measurements, atmospheric sounding and satellite observations. LDAS-A effectively improved the land surface variables (i.e., soil moisture and temperature) and has the potential to correct the uncertainties resulting from model-specific parameters and model atmospheric forcing (i.e., precipitation and radiation). The improved land surface conditions resulted to an improvement in the land-atmosphere feedback mechanism and the assimilated results showed better prediction of atmospheric profiles (i.e., potential temperature and specific humidity) when compared with radiosonde soundings. In addition, cloud-top temperatures predicted by LDAS-A showed significantly better spatial distributions and diurnal trends of cloud activity over the model domain, as confirmed by satellite observations of the infrared brightness temperature (MTSAT/1R1).

However, during or immediately after the assimilation, the reinitialized land surface conditions often suffered from substantial errors and drifts owing to predicted atmospheric forcing especially precipitation that destroyed the improved land surface conditions at very short time and in wider scales, misguides the land-atmosphere interactions thus severely affecting the model forecasts. Due to very strong influences of model predicted rainfalls on the assimilated land surface conditions, land surface assimilation processes dropped their merits and become unproductive during model forecasts. This problem is very severe and cannot be corrected in operational weather forecasting owing to the unavailability of future observations during the model execution. Therefore, the operational pitfall that arises between model atmospheric conditions and nature is one of the most challenging and unresolved problems encountered by both LSM and NWP communities.

To overcome the issues related with model inaccurate atmospheric forcing, atmospheric model physics should be in conjunction with land data assimilation. As an initial step to improve the model atmospheric conditions, LDAS-A modeling framework was extended by coupling the available Cloud Microphysics Data Assimilation System (CMDAS). CMDAS was basically developed over ocean (due to weak and homogeneous ocean surface emission) to improve the atmospheric moisture variables by assimilating AMSR-E higher frequency observations. It has been widely recognized that the use of AMSR-E higher frequency observations over land is not practical owing to the strong and variable land surface emission. However, LDAS-A has shown its merits in improving the land surface emission by assimilating lower frequency of AMSR-E observations and therefore has the potential to facilitate the microwave higher frequency (atmospheric) observations over land surface.

Consequently, a new, extended system that is referred to as Coupled Atmosphere and Land Data Assimilation System (CALDAS) was developed. CALDAS merges the land surface information obtained from lower frequencies (6.9 and 10.6 GHz) of AMSR-E channels with that of higher frequencies (23.8 and 89 GHz) to obtain the atmospheric moisture information over land surface, whereas the model operators maintain the consistency between model variables and assimilated variables. In this way CALDAS performs a synchronized improvement of land and atmospheric initialization in a physically consistent manner in a land--atmosphere coupled model.

Though it has been cited that higher frequency data have been contaminated by land surface emission, CALDAS has better exploited the multi frequency observation of AMSR-E and thus the assimilated cloud activities showed high correlation and compared well with MTSAT satellite observations. Particularly, CALDAS removed the model inaccurate rainfall events that contaminated the reinitialization of land surface conditions, and maintained the assimilated surface conditions during model forecasts. The elimination of model predicted rainfall events and cloud coverage improved the atmospheric forcing (e.g., solar radiation and rainfall) to the LSM. The improved atmospheric forcing combined with the assimilated soil moisture content guided the LSM to accurately represent the surface processes and land--atmosphere interaction in the land-atmosphere coupled model as confirmed by surface and radiosonde observations.

CALDAS also introduced cloud distribution that was not simulated by model but observed by AMSR-E channels. However, the significant reduction in the assimilated cloud condensate was observed after few hours from model reinitialization. This could be related to the model dynamics that were not adjusted in accordance with the assimilated atmospheric parameters. Improving the model dynamics in a physically consistent manner has been proposed as one of the future directions of this research.

全球規模、地域規模の数値気象予測情報を局所規模の情報にダウンスケーリングするためには、細かな計算格子を有するモデルの初期値や側方境界条件として、広域予測のための粗い計算格子モデルの出力をネスティングする力学的ダウンスケーリング手法や、局所規模の観測データの空間分布に関する統計的性質を用いて広域モデルの出力を補正する統計的ダウンスケーリング手法が用いられてきた。しかし、これらの手法では短期の数値予測において重要な大気状態量の初期値の局所スケールでの空間分布特性を直接導入することはできない。とりわけ、雲・降水過程は10kmスケールの積雲対流現象であり、河川・水資源管理に必要な流域スケールの降水分布の予測情報を高精度で得ることは困難であった。

これらの問題を解決に導くために、近年衛星情報と数値モデルを組み合わせたデータ同化手法が開発されてきた。まず、土壌層と陸面での電磁波伝播を表す高度なマイクロ波放射伝達モデル、鉛直一次元の陸面での水・エネルギーフローを表す陸面スキーム(SiB2)、および衛星観測と地上観測による大気からの強制力データ組み合わせた陸面データ同化手法(LDAS)が開発された。次に積雲対流を表現できるメソスケールの非静力大気モデル(NHM)とLDASを組み合わせたシステムも開発され(LDAS-A)、陸面の不均一性の影響を考慮したメソスケールの雲・降水予測システムが開発された。一方、大気中でのマイクロ波放射伝達モデル、NHMに組み込まれている雲微物理スキームと衛星観測データを組み合わせた雲微物理データ同化システム(CMDAS)が開発された。さらにLDASから得られる地表面のマイクロ波放射を境界条件とする大気の放射伝達モデルとCMDASを組み合わせた陸域-大気結合データ同化システムのプロトタイプも開発された。

本研究の目的は、これらを適切に組み合わせるシステムを開発し、その性能を評価して、降水予測のダウンスケーリング手法の基盤を構築することにある。

論文提出者は、まずLDAS、LDAS-A、CMDAS、NHM間のデータの受け渡し方法とそのタイミングを精査し、同化ウィンドウが比較的長いLDASにはEnsemble Kalman Filter (EnKF)を、一方同化ウィンドウが数10分と短いCMDASにはShuffled Complex Evolution(SCE-UA)をよる同化手法を用いることとした。また、CMDASによる同化結果を改善されたNHMの初期値として引き渡す手法を吟味し、初期に開発されたCMDASプロトタイプの欠陥を改善するとともに、NHMにおける雲微物理関係の変数だけでなく、力学場においてもデータの引渡しが確実に実施できる手法(dumping)を開発した

以上を踏まえ、論文提出者は、計算時間ステップの異なるNHM、雲微物理スキーム、陸面スキーム、放射伝達モデル間でモデル出力や観測データを適切に引き渡すことができ、かつ様々なスキームを結合、切り替えを可能とするカップラーの基本構造を開発した。汎用性に関して具体的には、陸面スキーム(SiB2、JMA新SiB)、マイクロ波放射伝達モデル(土壌層-地表面系、地表面-大気系)、同化スキーム(EnKF、SCE-UA)をそれぞれ2種類づつ実装した。

論文提出者は、さらにこのカップラーによる大気-陸面結合データ同化システム(CALDAS)をチベット高原の東西の地域に適用し、南北を東西に走る山脈で囲まれた高原平地部での陸面、大気状態の同化実験を行った。大気大循環モデル(GCM)は米国大気海洋庁の国立環境予測センター(NCEP)の現業予測モデルの同化-予測サイクルの出力を用い、NHMは改良型領域予測システム(ARPS)、衛星観測は改良型高性能マイクロ波放射計(AMSR-E)データを用いた。また、雲微物理スキームはLinの手法、陸面スキームはSiB2を用い、LDASにはEnKF、CMDASにはSCE-UAを選択している。

適用結果を、2008年に実施されたJICA技術協力プロジェクト「日中気象災害協力研究センター」で実施された強化観測データを用いて検証したところ、土壌水分、地表面温度、大気鉛直プロファイル、雲分布、日射量に関して、大幅な改善が見られた。土壌水分、地温に関しては、NHM単体、LDAS-Aで見られた大気モデルによる誤った降雨出力の影響を取り除くことができ、安定した高精度の予測値が得られた。ゾンデ観測値で検証した大気プロファイルの予測値は、NHM単体、LDAS-Aで見られた境界層での低温化傾向が大幅に改善され、混合層の発達がはっきり予測された。積算雲水量の予測値の空間分布は静止気象衛星の熱赤外画像で観測された雲の分布と良く一致しており、その結果地上での日射量も観測値と良く整合していた。論文提出者はこのように、新たに開発されたシステムの有効性を明確に示している。

以上、論文提出者は、電磁波伝播の基礎検討をもとに衛星による土壌水分観測精度の向上を通して、その成果をデータ同化システムに組み込むことにより、陸域上の大気中の水分量の算定精度を向上させ、降水予測の精度向上に大きく貢献している。また開発されたカップラーは、汎用性を有しており、今後様々なシステム要素(雲微物理スキーム、陸面スキーム、放射伝達モデル、同化スキーム、観測データ)の改善、交換に柔軟に適応でき、拡張性によるメリットが大いに期待できる。この科学的、工学的成果は、水資源、農業、生態系などの社会的利益分野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。