CONSIDERATION OF ANTECEDENT SOIL MOISTURE FOR PREDICTING FLOOD CHARACTERISTICS

Transcription

1 Annual Journal of Hydraulic Engineering, JSCE, Vol.59, 215, February CONSIDERATION OF ANTECEDENT SOIL MOISTURE FOR PREDICTING FLOOD CHARACTERISTICS Rie SAKAZUME 1 Masahiro RYO 2 Oliver SAAVEDRA 3 1 Student Member of JSCE, Graduate student, Dept. of Civil Engineering, Tokyo Institute of Technology ( M1-4 Ookayama, Meguro-ku, Tokyo, , Japan) 2 Student Member of JSCE, Doctoral student, Dept. of Civil Engineering, Tokyo Institute of Technology ( M1-4 Ookayama, Meguro-ku, Tokyo, , Japan) 3 Member of JSCE, Associate Professor, Dept. of Civil Engineering, Tokyo Institute of Technology ( M1-4 Ookayama, Meguro-ku, Tokyo, , Japan) Antecedent soil moisture is a key factor to determine flood characteristics. Understanding the role of antecedent soil moisture for flood is needed for conservation of ecosystem and for damage mitigation as well. The aim of this study is to quantify the relationships of the antecedent soil moisture to flood volume, peak discharge and its duration. Discharge and soil moisture were simulated by a hydrological model. Then, multiple linear models quantified the relationships. Thirty-five floods were analyzed from 25 to 212 in Huong River, Vietnam. As a result, not only total precipitation volume but also soil moisture before flooding explained flood volume and its duration. Similarly, antecedent soil moisture increased the predictability of peak discharge with peak rainfall amount. The flood characteristics changed by reflecting wet/dry condition at top soil. We expect that consideration of antecedent soil moisture to identify the flood characteristics accurately leads to practical flood countermeasure. Key Words : flood characteristics, antecedent soil moisture, multiple linear regression, hydrological model 1. INTRODUCTION Unpredictable heavy precipitation due to the climate change would increase the risk of flood disaster particularly in wet regions in Asia and Africa 1). Floods directly trigger damage to populated regions and indirectly increase poverty, which provides the vulnerability of human systems 2). Floods devastate not only infrastructures such as roads, buildings, power and telecommunication lines, but also enormous livelihoods 3). Therefore, appropriate countermeasures should be taken with proper understandings of the physical processes (i.e., generation process) of flood caused by heavy precipitation. Whereas mitigating flood disaster is critically important for development of our society, the ecological roll as flow regime must be considered to preserve biodiversity. Flood is regarded as a key driver of riverfloodplain ecosystems 4),5). The magnitude, frequency, duration, and timing of inundation influence multi-facets in the fluvial systems: nutrient cycling, primary production and biomass transport, and community dynamics of aquatic plant and fish. By aiming at conservation of the ecosystem in river, positive effects of flood disturbance have been surveyed 6). It has been revealed that antecedent soil moisture is one of the factors to understand flood characteristics as well as precipitation. For example, the role of soil moisture on the threshold runoff response was investigated by the observation 7) and simulation 8). Resulting from the evidence, the strong relationship between antecedent soil moisture and runoff coefficient was found. However, no studies have analyzed the relationship between antecedent soil moisture and flood from the viewpoint of flow regime facets (e.g., duration and frequency). Previous studies paid attentions not to the flow regime aspects but to runoff coefficient to reveal the relationship, although antecedent soil moisture influencing on flood should be significant for ecological system. It is inevitable to properly understand the role of antecedent water content for flow regime in order to have potential strategies for the conservation of ecosystem as well as for increasing chances for evacuation and mitigation against flood disaster. Thus, our study initially attempted to answer how strongly

2 antecedent soil moisture affects flood properties in the context of flow regime. The aim of this study is to quantify the relationship of the antecedent soil moisture in topsoil to the flow regime aspects of flood: total volume, peak discharge, and its duration. 2. STUDY AREA We targeted a watershed where floods occur frequently, and the characteristics are not altered strongly by water control facilities. Huong River basin was selected, which is located in Thua Thien Hue Province of the central Vietnam, between 16 to of the north latitude and from 17 to of the east longitude (Fig. 1). The total catchment area is 1,5 km 2, the topography changes from 1 m to 1,78 m above sea level. The terrain is classified as hilly mountainous region. The basin captures the heaviest precipitation during the monsoon season. Flood characteristics are not controlled by two dams existing in the region (Fig. 1). Binh Dien dam purposes to generate hydropower, whereas Ta Trach dam has not been operated yet. In total 35 single-pulse flood events during the target period from September to November in were selected for this study. The characteristics of total volume, peak discharge, and its duration are summarized in Table 1 (for the definition, see method 3.(2)). 3. METHODS (1) Discharge and soil moisture simulation A physically-based semi-distributed hydrological model was applied to simulate soil moisture and streamflow. In this study, the Geomorphology-Based Hydrological Model (GBHM) was employed 9). The [m] Kim Long Kim Long Binh Dien Binh Dien Binh Dien BinhDam Dien Dam Huu Huu Trach Trach river river Thuong Thuong Nhat Nhat Raingauge Raingauge location location Dam location Dam location Ta Trach Ta Dam Trach Dam Ta Trach Ta Trach river river Nam Dong Nam Dong Fig. 1 Huong River basin with the locations of rain gauges and dams. Table 1 Summary of flood characteristics: simulated flood volume, peak discharge and duration (n=35). unit average ± standard deviation volume 1 8 m ±3.56 peak discharge 1 3 m 3 s ±1.71 duration hrs 87±69 main modules are for hillslope and river routing. The spatial and temporal resolutions were set to 5 m and hourly, respectively. The river network was delineated from a 9 m digital elevation model (the Shuttle Radar Topography Mission: The whole basin was divided into 9 sub-basins based on the Pfafstetter s scheme, and sub-basins are further subdivided into a set of computational units namely flow interval with same distance bands (1 km width in this study) to the outlet of each sub-basin by using the contours. The hydrological processes in each flow interval were simulated sequentially from the uppermost section to downward. The hillslope module simulates the hydrological processes such as evapo-transpiration, the movement of groundwater, and overland flow at each grid within a flow interval. Then, the accumulated lateral inflow from flow intervals becomes later inflow to the main stream in a sub-basin. At last, outflow from each sub-basin is routed downstream sequentially. The topsoil with average depth of 1.5 m was horizontally subdivided into 12 thin layers. For each layer, the amount of soil moisture was estimated and taken their vertical mean per grid according to each soil type (Fig. 2 (a)). Soil moisture was simulated by onedimensional Richard s Equation (1). θ(z, t) t = z (Ψ(θ) + z) [k(θ, z) ] E(z, t) z θ rsd θ θ sat (1) Where t is time, z is vertical coordinate, θ(z, t), θ rsd, and θ sat are volumetric water content, residual water content, and saturated water content, respectively (θ rsd =.6 and θ sat =.46 in basin mean). k(θ, z) is hydraulic conductivity described by Van Genuchten s Equation, Ψ(θ) is capillary suction and E(z, t) is evapo-transpiration. We used the model calibrated by our previous study for high reproducibility of flood. The detailed description is available in Ryo et al 1). To consider the spatial heterogeneity of hydrological processes in the basin, 3 soil types and 6 land-use classes were applied (Fig. 2 (a), (b)). The soil type map was collected from the global soil type map of the Food and Agriculture Organization (FAO) Harmonized World Soil Database, version 12, and land-use map was provided by the National Hydro- Silty-clay Sandy-silt Clay-silt Cropland Deciduous broadtree Evergreen broadtree Tall grass Evergreen shrub Bare soil with shrub (a) (b) Fig. 2 (a) Soil map and (b) land use classification of this study area.

3 discharge [m 3 s -1 ] soil moisture [-] precipitation [mm] Discharge (Q) Soil moisture (θ) Precipitation (P) Meteorological Forecasting Institute of Vietnam. The spatial resolution of both parameter maps was 1, m. Discharge simulation to Kim Long station, the outlet of the watershed, was conducted during the target period. The results showed agreement with the observation by evaluating the performance (For example, simulation in 29 as Fig. 3, Nash-Sutcliffe Efficiency =.88). (2) Flood characteristics: volume, peak discharge and duration Flood characteristics such as flood volume (V total ), maximum peak discharge (Q peak ) and duration (D) were considered as objective variables in this study (see the conceptual framework as shown in Fig. 4). The threshold discharge for flood was defined as exceeding 75 th percentile of discharge (Q 75 : 37 m 3 s -1 ) at Kim Long st. during the entire period (median: 175 m 3 s -1 ). V total was defined as total water volume (m 3 ) integrated from hourly stream flow Q(t) (m 3 s -1 ) during flood occurrence: T e V total = Q(t) 36 (s) (2) T s Correspondingly, the beginning and end of times of one flood event (T s, T e ) were defined as timings when simulated discharge exceeds and falls down Q 75, respectively. Q peak (m 3 s -1 ) was selected as maximum hourly discharge of each flood event. D (hrs) is defined as duration of flood: D = T e T s (3) (3) Explanatory variables: precipitation and antecedent soil moisture To explain these flood characteristics, conditions of Discharge and soil moisture simulation Soil moisture Simulation Observation September October November time [hrs] Fig. 3 Discharge and soil moisture simulation in Q 75 Discharge Soil moisture Precipitation θ ini T ini Time scale P total P peak Q peak V total.5mm precipitation and initial soil moisture (i.e., soil moisture before flooding) were considered: total precipitation volume (P total ), peak precipitation (P peak ), and initial soil moisture (θ ini ) (Fig. 4). P total (mm) was calculated by summing up hourly rainfall amount (P) since the beginning of rainfall (T ini ) until T e. T ini was determined a time to be dated back from a timing of peak discharge to the moment which a corresponding rainfall event had started (threshold :.5 mm h -1 ). P peak (mm h -1 ) was defined as maximum rainfall amount between T ini and T e. θ ini ( ) was basin average water content at the top soil at time T ini. (4) Single / Multiple linear regression Firstly, single linear regression carried out for comprehension of the direct relationships of initial soil moisture to the flood characteristics (V total, Q peak, and D). Secondly, multiple linear regression was conducted to explain these characteristics by the explanatory variables (P total, P peak, and θ ini ). For explanatory variables, significance level (α) was set to.5. Among the explanatory variables, P total and P peak had nonsignificant relationships with θ ini as p-values indicated (.46 and.96, respectively). The performances of the statistical models with/without inclusion of θ ini were compared based on adjusted R-squared (R 2 ), which is a modified version of R-squared that has been adjusted for the number of predictors in the model as R 2 1 i (y i f i ) 2 /(n p 1) (4) i (y i y ) 2 / (n 1) Where y i is sample value, f i is estimate value, y is mean value of samples, n is the number of samples and p is the number of explanatory T s D T e Fig. 4 Conceptual diagram of each parameter.

4 Volume prediction [m 3 ] Peak discharge prediction [m 3 s -1 ] Flood duration prediction [hrs] Flood volume[m 3 ] Peak discharge [m 3 s -1 ] Flood duration [hrs] variables. Analysis was conducted using R software, version ). 4. RESULTS AND DISCUSSION It was found θ ini in relationships to V total, Q peak, and D as shown in Fig. 5 (R 2 for flood volume, peak discharge, and duration were.6,.4, and.8, respectively. p-values were.9,.13, and.5, respectively). These p-values may represent that the relationships of initial soil moisture to the flood characteristics were non-significant. Two possible reasons are argued as following. One possibility is that there were too few events under dry condition (θ ini : , Fig. 5), which induced the wide range of 95% confidence interval. The significance may be confirmed by the number of samples at the low initial soil moisture. Another possibility is nonlinear relationships. Three panels showed that flood volume, peak discharge and duration were low values during dry conditions and sharp increases occurred when θ ini exceeded.325 (i.e., wet condition, Fig. 5). Similarly, non-linearity of antecedent soil moisture to runoff generation was reported in Penna D. et al (211) 7). P-values would increase if linear regression is applied for non-linear relationships. Other statistical approaches, which volume peak discharge duration 8e+8 8e+8 R 2 =.6 (p-value=.9) 5 R 2 =.4 (p-value=.13) 15 R 2 =.8 (p-value=.5) 1 4e+8 4e e+ e Initial soil moisture [-] Initial soil moisture [-] Initial soil moisture [-] Fig. 5 Relationship of initial soil moisture to each flood characteristics: flood volume, peak discharge, and duration. Black lines show the results of linear regression. Gray dots represent the simulation results. Shaded area for each panel corresponds to the 95% confidence interval. Table 2 Summary of estimated models using multiple linear regression for the flood characteristics. Equations with adjusted R-squared for each flood characteristics are shown. unit volume m 3 peak discharge m 3 s -1 duration hrs equation V total = 1. 1 P total θ ini. 1 V total = 1. 1 P total. 1 7 Q peak = P peak θ ini. 1 Q peak = P peak 1. 1 D =. 1 P total θ ini D =. 1 P total adjusted R-squared (5).996 (6).968 (7).97 (8).834 (9).874 (1).83 volume peak discharge duration 9e+8 6e+8 R= R= R= e e+ e+ 3e+8 6e+8 9e+8 Volume observation [m 3 ] Peak discharge observation [m 3 s -1 ] Flood duration observation [hrs] Fig. 6 Comparisons between observed and calculated flood characteristics using Eq. (5), (7), (9) with respective correlations. Black line corresponds to 1:1.

5 discharge [m 3 s -1 ] precipitation [mm] demonstrate non-linearity like Random Forest, and Artificial Neural Network, may be employed to describe these relationships adequately. In the best model for V total, the two variables with positive coefficients were P total and θ ini. In the case of Q peak, P peak and θ ini were selected for the prediction model. For D, P total and θ ini were included as explanatory variables in the model. For all the characteristics by adding initial soil moisture as an explanatory variable, adjusted R- squared values increased in comparison to the models without consideration of initial soil moisture as shown in Table 2 (V total :.968 to.996, Q peak :.834 to.97, D :.83 to.874). The comparisons between observed and calculated flood characteristics using the best models (i.e., Eq. (5), (7), and (9) in Table 2) are illustrated in Fig. 6. Correlations for the relationships were.998,.955, and.939, respectively. Antecedent soil moisture condition was selected as a key predictor for the flow regime aspects (p<.1) in the best models as follows. (1) Not only total precipitation volume but also initial soil moisture explained the flood volume by comparing the adjusted R-squared values (increment from.968 to.996). Thus, it can be interpreted that the flood volume tends to be high due to heavy precipitation especially in the case of high initial soil moisture. (2) Antecedent soil moisture increased the predictability of peak discharge with peak rainfall amount (R 2 : from.834 to.97). High peak discharge during flood period would be generated by high rainfall amount at the moment of high initial soil moisture. (3) For flood duration, initial soil moisture was an explanatory factor as well as total precipitation volume (R 2 : from.83 to.874). Flood duration may be prolonged when precipitation volume is large in the case of high initial soil saturation. Especially for peak discharge, initial soil moisture enhanced the predictive power with 7% increase. For flood volume and duration, the predictive powers were increased similarly by including initial soil moisture (3% and 4%, respectively). In hydrological process, antecedent soil moisture relates to surface runoff and base flow. Wet ground surface has less capacity to store precipitated water. Therefore surface runoff flowing to a river is generated promptly. At the same time, the wetted ground has a propensity to push out the stored water continuously and increases base flow. These physical processes force to increase the flood volume, peak discharge and duration. Experimentally, we simulated a flood event under the different initial soil moisture conditions as θ ini =.2,.3, and.4 (Fig. 7). The higher water θini=.2 θini=.3 θini=.4 Q content, the larger the flood sizes were as our hypothesis (Table 3). In comparison with dry condition (θ ini =.2) to wet condition (θ ini =.4), the differences for flood volume, peak discharge, and duration were m 3, m 3 s -1, and 59 hrs, respectively. Interestingly, the effect of initial soil moisture was more impressive than the results of statistical analysis. Thus, ground wetness should be considered carefully for flood prediction. It should be noted that the results of this research can be model and site-specific (the characteristic of target basin is hilly mountainous region in high humidity during monsoon season). Thus, it must be interesting to demonstrate this analysis under various conditions (e.g., comparison with observation or different models and regions). The realization of this concept would be of value to manage flood considering both ecosystem and our society. 5. CONCLUSION Discharge simulation with different ground situations 1th Nov. to 2th Nov in 21 time [hrs] Fig. 7 Discharge simulation under different θini conditions. Table 3 Flood characteristics under differentθini conditions. unit θini=.2 θini=.3 θini=.4 volume 1 8 m peak discharge 1 3 m 3 s duration hrs We attempted to get reader s attention in order to consider the role of antecedent soil moisture which corroborates understandings of the ecologically-relevant hydrological aspects of runoff: volume, peak discharge, and duration. The realization of this concept has a potential to provide some benefits as follows. Comprehension of these relationships should be beneficial to predict the ecological responses. Similarly, accurate estimation of flood characteristics helps to reduce the risk of

6 damage. For stepping forward, we propose three directions: verification, further evaluation, and application. (1) Verification of the hypothesis: it should be conducted at an experimental basin or real watershed where streamflow, precipitation, and soil moisture are densely measured. Then, the relationships could be demonstrated based on field measurements. In fact, actual soil moisture is going to be measured from the next year in this target basin aiming to verify our findings and to increase the accuracy of flood prediction. Verification of this relationship with observation data will become available in the near future, and the results should lead to practical flood countermeasure. (2) Further evaluation: We demonstrated the relationships of antecedent soil moisture to both magnitude and duration of high flow pulses (i.e., floods). The other aspects, which are frequency, timing, and rate of change, could be further investigated with similar approach. Detecting such influences of hydrological processes to ecologicallyrelevant flow facets newly provides basic information for eco-hydrological studies. (3) Application of the theory for flood management: in recent, antecedent soil moisture condition has been taken into account to hydrological models explicitly as an input data for forecasting floods with high degree of accuracy 12),13). As input data to the models, remote sensing instruments including satellite have been employed because of estimating soil moisture at topsoil surface on large coverage and oppositely due to the limitation of data availability in a field 14). These products should lead to assist the management of flood thanks to high accuracy of flood prediction. ACKNOWLEDGEMENT: This work was supported by JSPS Asian CORE Program and JSPS Core-to-Core Program B. Asia-Africa Science Platforms. We appreciate the National Hydro- Meteorological Forecasting Institute of Vietnam for guiding us in field and sharing observation data. REFERENCES 1) Hirabayashi, Y., Mahendran, R., Koirala, S., Konoshima, L., Yamazaki, D., Watanabe, S., Kim, H. and Kanae, S.: Global flood risk under climate change, Nature Climate Change, Vol. 3, pp , ) Tran, P., Marincioni, F. and Shaw, R.: Catastrophic flood and forest cover change in the Huong river basin, central Viet Nam, Journal of Environmental Management, Vol. 91, pp , 21. 3) Adikari, Y. and Yoshitani, J.: Global Trends in Water- Related Disasters, The United Nations World Water Development Report 3, 29. 4) Poff, N. L., Allan, J. D., Bain, M. B., Karr, J. R., Prestegaard, K. L., Richter, B. D., Sparks, R. E. and Stromberg, J. C.: The natural flow regime, Bioscience, Vol. 47, pp , ) Junk, W. J., Bayley, P. B. and Sparks, R. E.: The Flood Pulse Concept in River-Floodplain Systems, Proceedings of the International Large River Symposium, pp , ) Robinson, C. T., Uehlinger, U. and Monaghan, M. T.: Stream ecosystem response to multiple experimental floods from a reservoir, River Research and Applications, Vol. 2, pp , 24. 7) Penna, D., Meerveld, H. J. T., Gobbi, A., Borga, M. and Fontana, G. D.: The influence of soil moisture on threshold runoff generation processes in an alpine headwater catchment, Hydrology and Earth System Sciences, Vol. 15, pp , ) Camporese, M., Penna, D., Borga, M. and Paniconi, C.: A field and modeling study of nonlinear storage discharge dynamics for an Alpine headwater catchment, Water Resources Research, pp , ) Yang, D., Herath, S. and Musiake, K.: A hillslope-based hydrological model using catchment area and width functions, Hydrological Science Journal, Vol. 47, pp , 22. 1) Ryo, M., Valeriano, O. C. S., Kanae, S. and Ngoc, T. D.: Temporal Downscaling of Daily Gauged Precipitation by Application of a Satellite Product for Flood Simulation in a Poorly Gauged Basin and Its Evaluation with Multiple Regression Analysis, Journal of Hydrometeorology, Vol. 15, pp , ) R Core Team: R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria, ) Merz, R. and Blöschl, G.: A regional analysis of event runoff coefficients with respect to climate and catchment characteristics in Austria, Water Resources Research, Vol. 45, ) Tramblay, Y., Bouvier, C., Ayral, P.-A. and Marchandise, A.: Impact of rainfall spatial distribution on rainfall-runoff modelling efficiency and initial soil moisture conditions estimation, Natural Hazards and Earth System Science, Vol. 11, pp , ) Toyra, J., Pietroniro, A., Martz, L. W. and Prowse, T. D.: A multi-sensor approach to wetland flood monitoring, Hydrological Processes, Vol. 16, pp , 22. (Received September 3, 214)