1. Introduction

For a long time, forests in Japan have faced several risks. In particular, abiotic risks such as wind, snow, and fire have the potential to trigger enormous damage. Wind damage has become a critical issue in Japan, because the damage has been often found in semi-mature or mature forests. However, risk assessment and management of wind risk has not yet been well developed in Japan. This is partly because there are regional differences in the risk frequency and because only a few studies have until recently focused on the issue. This dissertation aimed to develop a decision-support system for risk assessment with a special focus on wind damage. In particular, typhoon damage was targeted. The system consisted of a mechanistic wind damage risk assessment model, an airflow model, a growth model, decision trees analysis, geographic information system (GIS), and labour efficiency analysis. In particular, a British empirical-mechanistic model (GALES) was seen as a key model for the decision-support system. Thus, this dissertation was composed of the following six chapters: Introduction; Literature review; Acquisition of tree parameter for wind risk model; Adaptation and validation of wind risk model; Construction of a decision-support system; and Discussion and Conclusions.

2. Literature review

Current methods of wind risk assessment are classified into three approaches: observational/empirical, statistical, and mechanistic. The former two methods are effective only when sufficient damage data are available; they can only be utilized for regions where wind damage is frequently observed and the damage data is adequately recorded. In addition, these methods are rarely used for estimating future damage due to the lack of mechanistic information on trees and stands. The mechanistic method is based on the mechanistic behaviour of trees and stands (a mechanistic wind risk assessment model) and wind climate (an airflow model). Thus, it is useful to simulate future risk as a function of thinning strategies. In particular, one of the mechanistic models, GALES, has been widely adapted for use in a number of different environments in order to simulate the critical wind speed causing wind damage at the centre of stand.

3. Acquisition of tree parameters for wind risk model

　To obtain the parameters for GALES, this chapter focused on tree-pulling experiment in which trees were artificially pulled down by using a wire cable. The tree-pulling experiment was carried out for sugi (Cryptomeria japonica (L.f.) D.Don) and hinoki (Chamaecyparis obtuse (Sieb. Et Zucc.) Endl.) in the experimental forest of the University of Tokyo in Chichibu, Saitama Prefecture, Japan. Ten sugi trees and nine hinoki trees were successfully pulled over. Six sugi trees and eight hinoki trees were overturned, and four sugi trees and one hinoki tree were broken. By calculating the maximum bending moment at the stem base (TM(max,total)), the relationship between TM(max,total) and stem weight (SW) was determined, so that the results could be used for the GALES parameters. To calculate the critical wind speed leading to overturning, the linear relationships were TM(max,total) =229×SW (R2=0.94; sugi) and TM(max,total) =244.2×SW (R2=0.96; hinoki). The average values of modulus of rupture (MOR) for stem breakage were calculated as 42.5 MPa (sugi) and 71.6 MPa (hinoki).

4. Adaptation and validation of wind risk model

　As a mechanistic approach to wind risk assessment in Japan, a modified version of the GALES model, namely ForestTYPHOON, was developed with an airflow model, WAsP8, in order to estimate the wind damage in sugi and Japanese larch (Larix kaempferi (Lamb.) Carriere) stands in Himi (Toyama prefecture) and Mt. Yotei (Hokkaido prefecture) regions. Because these models were originally constructed for European conditions, it was necessary to adjust the models for the Japanese environment and wind damage phenomena caused by typhoons. Before wind damage estimation, the model parameters were statistically tested to show whether they were appropriate for our study areas. Although limited information on tree-pulling experiments was available in Japan, significant agreement was obtained between the relationships for both species in different locations.

　Wind damage estimation was conducted by using ForestTYPHOON with field survey data, forest inventory data, and WAsP with wind climate data observed at AMeDAS (meteorological) stations. The estimated damage was compared to the actual typhoon damage in 2004. The initial damage estimates did not show good agreement with the actual damage. The desired accuracy level was set to 70% based on the validation of GALES in Britain. For sugi, the accuracy calibrated with the estimated local wind speed (EWS)+30% and the critical wind speed (CWS)±1 m/s was at this level. The Yotei region showed suitable accuracy with EWS+30% and CWS±1.5 m/s, although the accuracy level was not satisfied in the northern area, because severe terrain effects were caused by Mt. Yotei and because there were insufficient wind climate data. This study suggests that the methods to estimate wind damage might be limited to particular terrain conditions; thus, further study is necessary to determine the sensitivity of the CWS calculated from ForestTYPHOON and alternative methods of wind climate estimation under different conditions.

5. Construction of a decision-support system

This chapter focused on the construction of a decision-support system with different components: labour efficiency to choose a suitable thinning method, spatial and temporal analysis to prioritise the target area using GIS, and decision trees to choose a course of silvicultural actions. In the system, the probability of wind risk was determined by using a binary rule of risk for four wind directions. Consequently, the decision trees indicated the significant stand characteristics relating to wind damage. Therefore, this system could suggest appropriate management actions to deal with the uncertainty of typhoon wind damage. The system consisted of the following five steps (Figure 1):

STEP 1 Stand growth simulation and labour productivity for thinning: Stands conditions were simulated for a certain period of time using a growth model according to management scenarios (thinning and harvesting).

STEP 2 Estimating critical wind speed, local wind speed, and labour productivity: The mechanistic method (consisting of ForestTYPHOON and WAsP) was applied to estimate wind risk by comparing the CWS and the EWS. The outputs were the CWS values of two failure types (overturning and breakage) and the EWS values for four directions (north, south, east, and west).

STEP 3 Calculating the probability of windthrow damage: First, the existence of wind risk was defined by a binary rule, i.e. expected wind damage in a stand was 1, and no wind risk was 0. Second, two methods were applied to define the probability of wind damage risk for spatial and temporal analysis and for the decision trees. For the spatial and temporal analysis (by GIS), all values of probability were averaged and then multiplied by 100 to obtain a percentage: 0% indicates no probability, and 100% indicates the highest probability of wind damage. The highest risk suggested that the stand had a high possibility of wind damage of both failure types caused by wind from any direction. For the decision trees analysis, the probability of wind risk was calculated in terms of two failure types by averaging the binary values, which resulted in two kinds of risk values (overturning and breakage).

STEP 4 Spatial and temporal analysis: GIS raster layers were created based on the averaged probability calculated in Step 3. The number of layers was dependent on the procedure of the growth model.

STEP 5 Decision trees analysis: Classification and Regression Trees (CART: Decision Trees) were created in terms of overturning and breakage. The wind risk probabilities defined in Step 2 were used as dependent variables; independent variables were selected from geographical and tree characteristics. There were two types of independent variables: constant and temporal. The constant independent variables were based on geographic characteristics, which scarcely changed. The temporal independent variables were found in the aboveground characteristics, which change depending on tree growth.

　At the end of the system, management alternatives could be made by prioritising (GIS outputs) and choosing courses of action (labour productivity and decision trees). The decision-support system for typhoon wind risk is helpful for decision makers who have certain limitations of forestry activities, such as available labour and accessibility.

6. Discussion and conclusion

　The decision-support system was discussed from the perspective of forest management (Figure 2). In particular, it was important to classify management alternatives to sharing, accepting, and reducing risk so as to decide actions. For instance, the alternatives could suggest management strategies such as having forest insurance, recreating forest composition (e.g. roads and preserved stands), and prioritising locations to receive thinning and harvesting. However, several requirements were found to further develop forest management strategies for wind damage risk. For the decision-support system, it was necessary to improve the airflow model for complex terrain, to incorporate gap and stand-composition analysis, and to expand the model to other tree species. Risk management also required analysing post-damage phenomena, such as the economical impact on timber markets. These system and management should be also simplified for use in actual forest management. In addition, the scale of forest management should be clarified to allow flexible actions for a period of time. More importantly, the procedures needed to include other abiotic risks by using various modelling techniques of abiotic damage. Such an integrated management strategy would become a strong tool to provide suitable alternatives to balance the conflicts arising from managing different risks, and thus allow stable forest management in the long term.

Figure 1. Framework of the decision-support system

Figure 2. Framework of forest management

　我が国は台風の常襲地域に位置するため、森林もまた常に風害のリスクに曝されている。小規模な被害も含めればほぼ毎年のように各地で森林の風害が発生しており、時には甚大な被害を受けることも珍しくない。しかしながらこれまで我が国では森林の台風に対するリスク評価やその対策がほとんど行われてこなかった。

　本研究は、スギ、ヒノキを主な対象として、林分に被害が発生する限界風速と当該地域の風況モデルを用いてリスク評価を行いリスクに関わる要因を抽出し、森林管理における風害リスク軽減を目的とした意思決定支援システムに結びつけたものである。

　第一章においては、国内外の森林風害研究のレビューが行われた。欧州では風向が一定で広域に吹き渡る強い季節風による被害が多いのに比べ、我が国は風向や風速が変わりやすい台風による被害が多いため、我が国のこれまでの研究手法が被害発生地に対する観察的、帰納的方法が主であったことを指摘し、森林の構造と風の両者を解析的に扱う必要を説いている。

　第二章では、森林の風に対する耐性を物理的に推定する方法として、GALESモデルを取り上げた。これは立木に対して風によるモーメントの掛かる高さ(風心)や風の引っ張り抵抗、樹幹の強度、根と土壌の緊密さなどを変数として、立木が根返りまたは幹折れのいずれかを生じる限界風速を推定する汎用性の高いモデルである。また風況モデルとして、GIS上で稼働し地形等の要因によって風速分布を推定できるDAMSおよびWAsPモデルの我が国への適用可能性が検討され、WAsPモデルが選択された。

　第三章では、我が国の人工林にGALESモデルを適用するため立木の大規模な破壊実験を行ってパラメータを推定した。東京大学秩父演習林において、若齢から壮齢に至る様々なサイズのスギ10本、ヒノキ9本を実際にワイヤで引き倒してそのときの最大張力を計測した結果、個体により倒れと幹折れを生じたが、限界点における最大モーメントと立木の自重との間に決定係数で95％前後の強い相関関係が見いだされた。オリジナルのGALESモデルは幾つかの物理的変数から成るが、樹幹の強度や根の支持力等は樹木のサイズや重量と関連するため、今回の破壊実験で限界モーメントが自重のみにより精度よく推定できることが導かれたことの意義は大きい。またスギとヒノキの比較では、ヒノキの限界モーメントがスギのそれよりも68％も高いことが明らかとなった。

　第四章では、前章で得られたパラメータを用いて、最近大きな風害を受けた北海道羊蹄山麓と富山県氷見市を対象に、改編されたGALESモデルとWAsPモデルを組み合わせ(これをForestTYPHOONモデルと命名)、被害の実態とモデルの推定結果を比較した。AMeDASで観測された風速データは時間平均値であるため、風害を引き起こす台風の瞬間最大風速とは乖離がある。風害を受けた林分と受けなかった林分の現況をForestTYPHOONモデルに入力して限界風速を推定した結果は、観測された風速データよりも全般に高く、そのことが裏付けられた。瞬間最大風速を観測された風速データの30％増しとした時、モデルによる風害発生の推定精度は70％を超えて最大となり、十分に実用域に達した。

　第五章では、ForestTYPHOONモデルのほか林分成長予測モデルや森林GIS、要する労働量等を用いて、時間の経過に伴う各林分の風害リスク評価を行った。3通りの森林管理シナリオ、すなわち無間伐、頻繁な弱度間伐、強度間伐を想定して50年間の林分の推移をシミュレートし、期間中に最も風害リスクの高まる時期を推定した結果、林分の位置する立地環境とともに間伐の仕方が大きく影響することが分かった。この結果を二分岐の決定木に表示し、風害リスクを軽減させるための意思決定支援システムの根幹とした。このシステムによれば、最も風害リスクの小さい人工林施業は樹高が低いうちに比較的強度の間伐を少数回行って本数を減じ、樹高が高くなってからの間伐を極力回避することである。長伐期施業は樹高の高い高リスクの期間が長期化するため、風害リスク軽減の視点からは望ましくない。

　第六章においては以上を取りまとめ、風害リスクを軽減する人工林施業のあり方と本論文で開発した意思決定支援システムの有効性について考察を加えた。

　以上、本論文は我が国の代表的造林樹種であるスギ等を対象として、風害に対する限界モーメント推定に関する新たな知見を提示するとともに、森林現況と風況から風害リスクを高い精度で推定するシステムを開発し、さらに人工林管理のための意思決定支援システムの開発も行うなど、学術上応用上貢献するところが少なくない。よって審査員一同は本論文が博士(農学)の学位論文として価値あるものと認めた。