Both solid (snowfall) and liquid (rainfall) precipitation are critical components of the water and energy cycles. During precipitation formation huge amounts of heat are released into the atmosphere, becoming the driving force of weather systems. Rainfall provides fresh water, and especially for developing economies, it is very important in driving these economies as most of them are agriculture based, depending to a large extent on rain-fed farming. It is therefore necessary to have reliably accurate precipitation forecast

The potential of remote sensing to identify and inform on a wide variety of issues relating to the environment has been demonstrated over time. Remote sensing offers the opportunity to obtain information about systems, objects or things without necessarily being close to them. It thus is a powerful tool for informing decision makers and technocrats as they attempt to address issues relevant to their specialist fields or needs.

Within the microwave region of the electromagnetic spectrum, water exhibits interesting behavior depending on whether it exists in bound state (e.g. wet soil) or unbound state (free water). This behavior is captured in the form of the dielectric constant which defines the electrical properties of a material. To serve as an illustration, very dry soil has a dielectric constant of ~ 4, but the dielectric constant of free water is ~ 80 at L-band frequencies. This wide gap in the behavior the dielectric constant helps in understanding water quantities in substances using microwave frequencies.

Within the microwave region, there are two branches of remote sensing; namely active and passive microwave remote sensing. In the active part, a radar sensor emits radiation which interacts with the object. The sensor then receives backscattered signal and depending on its magnitude, various interpretations can be arrived at. On the other hand, a passive microwave radiometer does not emit radiation but rather senses radiation being emitted by the object under study. Although the theoretical basis of microwave remote sensing is clear, this field is relatively young (~ 40 years old) and many of its potentials have not yet been fully harnessed. This research effort aims at making use passive microwaves to improve predictability of precipitation over land, lending a hand to water resource management.

It is an accepted fact that it is difficult to have an adequate distribution of instrumentation to capture the water cycle. There are regions that are actually un-gauged due to their remoteness. On the other hand, many satellite sensor platforms have been launched in space and although their spatial resolutions may not be as would be desired, the trend is moving towards finer spatial resolution imagery. Microwave specific sensors already launched include but are not limited to, the Advanced Microwave Spectro-Radiometer on board the Earth Observing Satellite (AMSR-E), the Special Sensor Microwave Imager (SSM/I). In this research, AMSR-E imagery (brightness temperatures) is utilized.

Within passive microwave remote sensing, two distinct approaches have emerged, observation of atmosphere and observation of land surface. Due to response signature of microwaves, lower frequencies have been favored for observation of land surface condition, while higher frequencies are favored for atmosphere observation over ocean and sea surfaces. At lower frequencies (< 20GHz) the atmosphere is largely transparent and hence land surface condition can be inferred. Higher microwave frequencies are sensitive to atmospheric state and have therefore been typically exploited for its observation.

At all microwave frequencies, land surface exhibits heterogeneous emission due to its heterogeneous nature (soil moisture, roughness e.t.c.), and in an effort to overcome this, atmospheric research with microwaves has been devoted to observations over sea surfaces.

The sea surface exhibits largely homogeneous emission and is significantly darker (colder) than land surface emission, which means that atmospheric emissions can be easily detected and identified.

In the past empirical models have been developed and used to obtain estimates of soil moisture. These empirical models are region specific and lack strong physical basis.

They are simple and easy to implement and find application in grid based analyses.

Several physically based models have also been developed to address various parts of the challenge. The approach to incorporate surface roughness effects is by using statistical roughness parameters namely root mean square height and correlation length. In the past, the effects of correlation length were ignored in simulations yielding somewhat unreliable results.

To address precipitation estimation and prediction over land, it is necessary to unify the gains made in land surface condition remote sensing and atmosphere condition remote sensing. In this research, the first target is to improve surface emission modeling to address land surface heterogeneity, which would enable research of atmosphere over land. With surface emission modeling clarified, there is need to unify it with radiative transfer modeling in atmosphere.

The models developed in this quest though significantly improved still contain modeling assumptions and are thus not suited for direct retrieval of land and atmosphere conditions, it can best be used in forward modeling.

Data assimilation strategies are used to combine imperfect models and observations that include errors for forecast simulations. Data assimilation enables imperfect models and inadequate observations that contain observation errors give better predictions than if only either the models or the observations were used to obtain the same prediction.

Field experiments seeking to understand the effects of surface roughness on land surface emission were undertaken. The results of these experiments demonstrated that in addition to surface roughness, it is important to consider the effects of shadowing introduced by the roughness of the surface. By incorporating effects of shadowing on a rough surface, it was possible to obtain remarkably good agreement between observations and simulations. This research verified that the Advanced Integral Equation Model (AIEM) can be used to model emission from a rough, bare and wet soil surface if shadowing effects are considered.

A land data assimilation scheme (LDAS) developed using empirical QH model as its observation operator was modified to assess the capability of improved understanding of surface emission. Using this improved surface emission model as its observation operator,

LDAS was found to give reliable near surface soil moisture estimates. Near surface soil temperature was also found to improve significantly in comparison to no-assimilation cases.

After the verification of AIEM as being a reliable model for representing surface emission, it was coupled with Dense Media Radiative Transfer model (DMRT) coupled with 4 stream approximation for soil that accounts for volume scattering in dry soil, to address radiative transfer from soil. To address the effects from the atmosphere, 4-stream approximate model for atmosphere was used. To address radiative transfer in vegetation, the w - t model was used. The performance of the coupled Radiative Transfer Model referred to hereafter as the Land Atmosphere Radiative Transfer Model (LA-RTM) was tested using data from Tibet and Mongolia sites in the Coordinated Enhanced Observing Period (CEOP) reference sites.

For bare wet surfaces, LA-RTM and the improved surface emission model simulations agree perfectly at low frequencies, and hence it can be used for modeling radiative transfer at all microwave frequencies with an increased level of confidence.

An ice microphysics data assimilation scheme (IMDAS) had been developed previously for estimation of cloud properties over sea surfaces. To use it over land, it is necessary to have reliable estimation of land surface condition. By using LDAS, improved land surface condition estimate is obtained, which is then used as lower boundary condition in IMDAS. IMDAS as originally developed did not consider precipitation for direct retrieval as assimilation variables, since it was developed for non precipitating cloudy atmosphere. In this research it was extended to allow consideration of snow and rain for direct retrieval, since in Tibet, the target site, presence of precipitating clouds is reported as being significant.

Using this coupled data assimilation approach, improved prediction of surface conditions (mainly soil moisture and surface temperature) and integrated atmospheric variables (water vapor, cloud water content, cloud ice content, rain water and snow water) was realized.

By feeding back these improved land surface and atmosphere conditions to the mesoscale model, there was significant improvement in precipitation forecast skill. Comparisons of daily cumulative precipitation forecast by the model using assimilation results gives better agreement with reported daily cumulative precipitation. Time series prediction skill showed promise, though it has not been validated in this research as in-situ data was missing. To compensate for this, Infra-Red imagery is used to compare against forecast, giving an indication of the distribution pattern. 1-hour forecast (after assimilation) showed good agreement with observed cloud pattern. For subsequent 6 - hour, 12-hour forecasts, there was general agreement but the distribution did not match well. However, in the case of 24-hour forecast, though the distribution is somewhat different, there is agreement about the quantities simulated. It is suspected that wind fields are not modeled correctly, thereby interfering with precipitation and cloud transport.

This is attributed to the fact of the mesoscale model reading boundary conditions from Global Circulation Model (GCM) output. Thus if GCM output has incorrect wind fields, these are introduced in the mesoscale model upsetting the improved states.

全球規模、地域規模の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、数値気象予測モデル分野ではネスティングや、広域予測のための粗いグリッド規模のモデル出力の統計的性質と対象とする局所規模の観測データの統計的性質合わせる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極めて重要な初期値を、対象とする狭域規模で物理的整合性をもって得ることはできず、それがゆえに広域規模の予測情報を効果的に狭域規模に利用し、その予測精度を向上するには至っていない。

本研究では衛星マイクロ波放射計観測データを効果的に用い大気-陸面結合データ同化手法を中核とする、全球規模-地域規模―流域規模を一貫して記述できる陸域での降水予測精度向上のための物理的ダウンスケーリングシステムの開発を目指すものである。

まず、データ同化に用いる衛星マイクロ波放射伝達モデルを精緻化するために、東京大学田無農場に地上マイクロ波放射計を設置してさまざまな条件下での土壌のマイクロ波放射輝度温度観測を実施し、稠密媒体の放射伝達モデル(DMRT)に地表面でのshadowing効果を導入した新たな土壌マイクロ波伝達モデルを開発することにより、水平、垂直両偏波で定量的に妥当な値が得られるようになり、その有効性はモンゴルでの土壌水分の地上観測と衛星搭載マイクロ波放射計データにより実証された。

次にこの高度化された土層、地表面の放射伝達モデルを境界条件とする大気の放射伝達モデル(陸域-大気結合マイクロ波放射伝達モデル)を開発した。このモデルに、2004年にチベット高原で実施したゾンデによる大気観測から得られた気温、水上気圧のデータを用いて大気上端でのマイクロ波の各周波数での輝度温度を算定し、衛星による観測値と比較した結果、大気の影響を受けない低い周波数で計算値と観測値が一致している一方で、周波数が高くない、大気中の水蒸気量の違い、雲水や氷晶、降水粒子の存在などの影響を受ける高周波数側で観測値と計算値の誤差が広がっていることが示された。このずれは大気中の水分情報を含んでおり、陸面のマイクロ波放射を精度良く算定することによってマイクロ波放射計で陸域上の雲情報の抽出が可能であることを示す画期的な成果となった。

そこで、大気モデルとの結合に使われる鉛直一次元の陸面モデル(SiB2)、陸域-大気結合マイクロ波放射伝達モデル、数値気象予測モデル(ARPS)組み込まれている雲微物理スキーム、衛星搭載マイクロ波放射計データとを組みあわせた衛星による大気-陸面結合データ同化システムを開発し、チベット高原に適用し、高現上の雲水量を分布を算定した結果、衛星搭載可視・赤外データで得られる雲分布とよく一致する結論を得た。衛星搭載マイクロ波放射計で陸域上の大気の雲水量を算定できた事例はなく、国際的にも画期的な研究成果であると評価できる。

以上、本研究は、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ることにより、降水予測に大きく貢献している。この科学的成果は、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。