A new process-based distributed model, the National Institute for Environmental Studies (NIES) integrated catchment-based ecohydrology (NICE) model, was developed. The model includes surface-unsaturated-saturated water processes and assimilates land-surface processes describing the variation in phenology with Moderate Resolution Imaging Spectroradiometer (MODIS) satellite data. The model was applied to the Kushiro River catchment (northern Japan, area of 2204.7 km2) with a resolution of 500 m and 8 days averaged vegetation changes. Excellent agreement between simulated and measured values was obtained for soil temperature, soil moisture, groundwater level, and river flow discharge during the 6 month snow-free period, achieved by taking into account vegetation phenology, soil properties, and geological structure. The model explains water cycle change and drying phenomena in the Kushiro Mire associated with vegetation change caused by the increased sediment load due to river channelization and consequently the invasion of alder (Alnus japonica) into the mire.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The Kushiro Mire (the largest mire in Japan) and the Kushiro River catchment (area of 2204.7 km2 located in northern Japan), shown in Figure 1 (the Digital National Land Information GIS data of Japan, 1993), have been changed by conversion to urban or agricultural uses since 1884. The rivers flowing through the mire have been altered by water works. Channelization of meandering rivers was introduced in the 1980s in the northern part of Kushiro Mire in order to smoothly drain runoff and protect farmlands from floods. Owing to channelization (Figure 2a), runoff containing nutrients from farmland and sediments from short-cut channels (arrow) have flowed directly into the mire and deposited flood-borne sediment.
 Sediment deposition in the mire had caused topographical changes, lowering the groundwater level and causing some of the soil to dry out [Environment Agency of Japan, 1984, 1993]. Consequently, alder (Alnus japonica (Thunb.) Steud.), a deciduous tree forming swamp forests with a height of ∼15 m, has invaded the mire in the downstream areas of the Kucyoro River and has increased its distribution (red area in Figure 2b). Alder has propagated widely around the Kushiro Mire after channelization, owing mainly to the lowering of the groundwater level and the increased nutrient input, resulting in the gradual shrinking of the mire. These figures show that human activities have changed the water cycle in the Kushiro Mire and thus vegetation succession.
 The water dynamics in the Kushiro River catchment, including the Kushiro Mire, are changing spatially. Deep groundwater levels in forested areas and shallow groundwater levels in the mire, for example, change on account of the sudden topography fluctuations over short distances and seasonal variations in vegetation. When a hydrological model is applied to this area, it is recommended to take into account surface runoff and unsaturated-saturated water processes and land-surface processes assimilated with satellite data to describe temporal variation in vegetation growth and phenology.
 Only SHE, SWATMOD, and WEC-C combine surface runoff with groundwater flow. SHE utilizes the major hydrological processes of water movement to describe overland and channel flow, unsaturated and saturated subsurface flow, canopy interception, evapotranspiration, and snowmelt. However, in this model the modeled land surface process of vegetation phenology is treated as constant, and the unsaturated-saturated water processes are quasi-lumped as a constant. SWATMOD, consisting of SWAT [Arnold et al., 1993] and MODFLOW [Sophocleous et al., 1999], is a semiconceptual model, which includes unsaturated-saturated water processes. WEC-C is a distributed, deterministic, catchment-scale model of water flow and solute transport with a rectangular grid of uniform cell size in the lateral plane combined with a system of soil layers in the vertical direction. However, the land-surface process is treated as constant and does not reflect seasonal changes. Therefore it is necessary to develop process-based distributed models including these two critical requirements, which are necessary to clarify the relationship between the water cycle and various indices, such as soil moisture, groundwater level, surface temperature, and evapotranspiration, caused by the spatial variability in vegetation type, soil texture, and topography.
 Furthermore, vegetation phenology and water cycle are closely related. There are only a few studies which have simulated both soil moisture and groundwater levels in a larger catchment and assimilated them with, for example, Moderate Resolution Imaging Spectroradiometer (MODIS) satellite data from storm events and over an annual timescale. A very powerful tool for simulating precise land surface processes and for predicting a time series of root zone soil moisture content is the assimilation of remote-sensing measurements of surface soil moisture into land surface models, for example, soil vegetation atmosphere transfer (SVAT) or a simple biosphere model [Ragab, 1995; Li and Islam, 1999; Wigneron et al., 1999; Montaldo et al., 2001]. However, the studies do not consistently use higher-order products such as fraction of photosynthetically active radiation (FPAR) or leaf area index (LAI), important parameters for evaluating vegetation growth and phenology [Christopher et al., 1998].
 The objective of the current research is to simulate the water cycle change and drying phenomena in the Kushiro Mire due to the effects of vegetation changes (Figure 2). Thus a new integrated catchment model which simulates land surface processes, including the effect of temporal and spatial vegetation changes and surface unsaturated-saturated water processes, is necessary in order to evaluate the water cycle changes in the whole catchment. We developed the National Institute for Environmental Studies (NIES) integrated catchment-based ecohydrology (NICE) model, which reproduces water cycle changes and drying phenomena in the Kushiro Mire. The details are presented in this paper.
2. Model Description
2.1. General Model Framework
 The NICE model consists of SiB2 [Sellers et al., 1996] for soil moisture and heat flux, the U.S. Geological Survey MODFLOW model of three-dimensional groundwater flow [McDonald and Harbaugh, 1988], and a grid-based hydrology model (Figure 3). MODIS satellite data with a 1 km mesh were input into the model to describe the spatial and temporal changes of vegetation phenology. The water flux between recharge layer and groundwater layer was calculated in order to combine the soil moisture model and the groundwater flow model in each time step. The effective precipitation and the seepage between river and groundwater are included in the model. Therefore the model can reproduce long-term components of river flow discharge due to recharge rates in addition to short-term components.
2.2. Biophysical and Soil Moisture Models
 Since a detailed description of SiB2 can be found in a previous study [Sellers et al., 1996], only a brief description of heat and water transfer is given here. SiB2 divides canopy into two layers (canopy layer and ground surface) and soil into three layers (upper layer, intermediate layer, and lower layer) in the vertical dimension. The governing equations for SiB2 prognostic variables consists of temperatures, interception stores, soil moisture stores, and canopy conductance to water vapor.
2.2.1. Canopy, Ground Surface, and Deep Soil Temperatures
The subscript c refers to the canopy, g refers to the soil surface, and d refers to the deep soil. Tc, Tg, and Td (K) are canopy, ground surface, and deep soil temperatures; Rnc and Rng (W m−2) are absorbed net radiation of canopy and ground; Hc and Hg (W m−2) are sensible heat flux; Ec and Eg (kg m−2 s−1) are evapotranspiration rates; Cc, Cg, and Cd (J m−2 K−1) are effective heat capacities; λ (J kg−1) is latent heat of vaporization; τd (s) is day length; and ξcs and ξgs (W m−2) are energy transfer due to phase changes in Mc and Mg (described in section 2.2.2), respectively.
2.2.2. Interception Stores
Mc and Mg (m) are water or snow/ice stored on the canopy and on the ground, P (m s−1) is precipitation rate, Dd (m s−1) is canopy throughfall rate, Dc (m s−1) is canopy drainage rate, Eci and Egi (kg m−2 s−1) are interception loss of canopy and ground, and ρw (kg m−3) is density of water.
2.2.3. Soil Moisture Stores
Wi is the soil moisture fraction of the ith layer (=θi/θs), θi (m3 m−3) is volumetric soil moisture in the ith layer, θs (m3 m−3) is the value of θ at saturation, Di (m) is the thickness of the soil layer, Qi,j (m s−1) is the flow between layers i and j, Q3 (m s−1) is gravitational drainage from recharge soil moisture store, Ect (kg m−2 s−1) is canopy transpiration, Egs (kg m−2 s−1) is ground evaporation, and Pw1 (m s−1) is infiltration of precipitation into the upper soil moisture store.
2.2.4. Canopy Conductance to Water Vapor
where gc (m s−1) is canopy conductance, kg (s−1) is a time constant, and gc,inf (m s−1) is the estimated value of gc at t → 8. In addition to equations (1)–(9), SiB2 models the radiative transfer, aerodynamic resistance, turbulent transfer, and canopy photosynthesis.
2.3. Groundwater Model
 A partial differential equation of three-dimensional groundwater flow is expressed in the following equation [McDonald and Harbaugh, 1988]:
Kxx, Kyy, and Kzz (m s−1) are values of hydraulic conductivity along the x, y, and z coordinate axes, respectively; hg (m) is the potentiometric head; W (1 s−1) is the volumetric flux per unit volume representing sources and/or sinks of water; and Ss (m−1) is the specific storage. Equation (10) is discretized by using the finite difference method on the assumption that the variables between two cells change linearly:
where hi,j,km is head at cell i, j, k at time step m; CV, CR, and CC are hydraulic conductances or branch conductances between node i, j, k and a neighboring node in the horizontal, lateral, and vertical directions; Pi,j,k is the sum of coefficients of head from source and sink terms; Qi,j,k is the sum of constants from source and sink terms, where Qi,j,k < 0.0 for flow out of the groundwater system and Qi,j,k > 0.0 for flow in; SSi,j,k is specific storage; DELRj is the cell width of column j in all rows; DELCi is the cell width of row i in all columns; THICKi,j,k is the vertical thickness of cell i, j, k; and tm is the time at time step m. For steady state stress periods the right-hand side of equation (11), the storage term, is set to zero. For solution by computer, equation (11) is modified into the following form at time step m:
In equation (12) the time superscript is removed for simplicity. HCOFi,j,k contains Pi,j,k and the negative of the part of the storage term, which includes the head in the current time step m (the negative sign comes from moving the term to the left-hand side). RHS includes −Q (the negative sign comes from moving Q to the right-hand side) and the part of the storage term that is multiplied by the head at time step (m − 1).
 The horizontal conductance between cells i, j, k and i, j + 1, k is given by using the equivalent conductance in a set of conductances arranged in series as follows:
where TRi,j,k is transmissivity in the row direction at cell i, j, k, DELRj is the grid width of column j, and DELCi is the grid width of row i.
 In the quasi-three-dimensional approach [McDonald and Harbaugh, 1988] the semiconfining unit makes no measurable contribution to the horizontal conductance or the storage capacity of either model layer; the only effect of the confining bed is to restrict vertical flow between the model cells. Under these assumptions the impact of the semiconfining unit can be simulated without using a separate layer in the finite difference grid. Because three intervals (the lower half of the upper aquifer, the semiconfining unit, and the upper half of the lower aquifer) must be represented in the summation of conductance between the nodes, the vertical conductance can be expressed as follows:
where VKi,j,k is vertical hydraulic conductivity of cell i, j, k, VKCBi,j,k is the hydraulic conductivity of the semiconfining unit between cells i,j,k and i,j,k + 1, and THICKCB is the thickness of the semiconfining unit.
2.4. Surface Hydrology Model
 The surface hydrology model consists of a hillslope hydrology model based on a kinematic wave theory and a distributed stream network model based on both kinematic and dynamic wave theories. The hillslope hydrology model consists of a water and thermal energy budget model and a surface runoff model. The kinematic wave model for distributed surface runoff can be expressed by the following equations [Takasao and Shiiba, 1988]:
where q (m2 s−1) is the discharge of unit width, r(x, t) (m s−1) is the effective rainfall intensity at position x and time t, b(x) (m) is the width of the flow, θ(x) is the riverbed gradient, k (m s−1) is the hydraulic conductivity in the “A layer” with a depth of D (m) near the ground surface, n (m s−1) is the Manning coefficient, and m = 5/3. When H (m) is defined as the depth of the rainwater flow in the A layer, ha (m) is given as the apparent flow depth (=γH), γ is defined as the porosity of the A layer, and d = γD, then the true flow depth is given by ha/γ for ha < d, and by d/γ + ha − d for ha > d.
 The basic equation of one-dimensional unsteady flow is expressed by the stream network model in the following continuity equation (equation (17)) and momentum equation (equation (18)):
where A (m2) is the cross-sectional area, Q (m3 s−1) is the discharge, ql (m2 s−1) is the lateral inflow q entering along the side of the river channel simulated by a hillslope model, hr (m) is the flow depth in the river, g (m s−2) is the gravitational acceleration, If is the friction slope, and i is the bed slope. The first term on the left-hand side of equation (18) is the local acceleration term, the second is the convective acceleration term, the third is the pressure force term, the fourth is the friction force term, and the fifth is the gravity force term. The local and convective acceleration terms represent the effect of inertial forces on the flow. The alternative distributed flow routing models are produced by using the full continuity equation while eliminating some terms of the equation (18). The simplest distributed model is the kinematic wave model, which neglects the local acceleration, convective acceleration, and pressure terms in equation (18); that is, it assumes If = i and the friction and gravity forces balance each other. The diffusion wave model neglects the local and convective acceleration terms but incorporates the pressure term. The dynamic wave model takes in all the acceleration and pressure terms in equation (18).
Equations (17) and (18) are discretized by using Preissmann's implicit scheme in the following nonlinear simultaneous equations:
where the superscript n refers to the time step, the subscript i to the number of simulation section, and θ refers to the weighting parameter (0.5 < θ < 1.0). When equations (19) and (20) are applied to a single channel component with N simulation section, the equations consist of 2N − 2 nonlinear simultaneous equations, and the unknown variables are the discharge and flow depth (i = 0, 1, …, N − 1) at time t = (n + 1)Δt, which account for 2N. Then two equations for the unknown variables at both ends of a single channel (ΔQ0, Δh0, ΔQN−1, ΔhN−1) are produced by applying the Taylor expansion to the first-order term around the unknown approximations and eliminating the unknown variables except the ends. Because there are the continuity conditions on discharge and flow depth at the confluent/effluent points of several channels in a grid box, the simultaneous equations consist of only the unknown variables at the intersections between the river channel and the grid boundary in the grid box. The simultaneous equations for the unknown variables for discharge and flow depth in the whole catchment are solved for the continuity conditions at each grid boundary and for the boundary conditions at the farthest upstream and downstream ends by using the Gauss method. This process was repeated by using the Newton-Raphson iteration method for convergence and correcting the approximate solution until the variables (ΔQi, Δhi) become small enough. Finally, the discharge and flow depth at each point are given.
2.5. Integration of Models
 In a region where the gradient of an elevation or a wetting front is much smaller than 1 the vector of hydraulic gradient is approximately downward. So the flow in an unsaturated layer can be estimated as vertically one-dimensional. The transfer of water between adjacent soil layers in equations (6), (7), and (8) is given by
where Qi,i+1 (m s−1) is downward flow from soil layer i to soil layer i + 1, Di (m) is thickness of the soil layer, Ki (m s−1) is hydraulic conductivity for ith layer (=KsWi2B+3), (m s−1) is the estimated effective hydraulic conductivity between layers, ψi is matric potential for the ith layer (= ψsWi−B), Ks (m s−1) is hydraulic conductivity at saturation, ψs (m) is soil moisture potential at saturation, and B is an empirical constant.
 To combine unsaturated flow and saturated flow, the water flux qf is expressed by using the gradient of hydraulic potentials between the deepest layer of unsaturated flow and the groundwater level in the expansion of the drainage in the original SiB2 model [Sellers et al., 1996] in the following equation:
where (m s−1) is the estimated effective hydraulic conductivity between unsaturated and saturated layers, Ψg(= hg) and Ψ3(= ψ3 + Dg + D3/2) (m) are hydraulic potentials at the groundwater surface and the lower layer of unsaturated flow, Dg (m) is the distance between the top of the second layer and the bottom of the 20th layer in the groundwater model, and hg (m) is the hydraulic head simulated by the groundwater model. When the groundwater level rises and enters into the soil moisture layer, the partial pressure is set at the bottom of unsaturated layer (Ψ3 = ψp) to simulate soil moisture. After the water flux qf is calculated in each time step, the flows between each unsaturated soil layer Qi,j are simulated in equations (21) and (22) by using an improved backward implicit scheme in order to simulate soil moisture θi in the ith layer in equations (6) to (8). Furthermore, this flux is input into the groundwater flow model as recharge rate at the highest active cell as the upper boundary condition, and the groundwater flow model is simulated.
 For the treatment of effective rainfall intensity r in equation (15) of the surface hydrology model the effective precipitation can be calculated from the precipitation rate P (m s−1), the infiltration of precipitation into the upper soil moisture store Pw1 (m s−1), and the evapotranspiration rates (Ec + Eg) (kg m−2 s−1) by the following equation:
When the volumetric soil moisture θi takes greater a value than that of saturation θs, the surplus of each value is added to the right-hand side of equation (24) as return flow to the surface. The seepage between river and groundwater depends on the interaction between flow depth at river and groundwater level of each cell. A volumetric flux between them is calculated by means of Darcy's law [McDonald and Harbaugh, 1988]:
where Qs (m3 s−1) is the volumetric flux of seepage between river and groundwater (=qll, where l is the channel length of each component), kb (m s−1) is the hydraulic conductivity of the riverbed, Ab (m2) is the cross-sectional area of the groundwater section, bb (m) is riverbed thickness, hg (m) is groundwater head, and Hb (m) is the hydraulic potential of the river. In a recharge situation (hg < Hb) the volumetric flux does not exceed the total volume of river flow. When the convergence procedure is conducted in a groundwater flow simulation, the flow depth of the river in the grid-based distributed runoff model is fixed.
 In this way, various models from the ground to the surface were connected by considering the flux, allowing groundwater level and soil moisture to be calculated only from the meteorological data, vegetation class, and soil texture, as shown in Figure 3. The ambiguity of effective precipitation seen in most of the previous catchment models is avoided because the model can simulate the change of infiltration flux in equation (24) at every time step. Furthermore, both short-term and long-term (base flow) components of river runoff can be simulated correctly owing to the effects of return flow and seepage in equations (24) and (25) by combining the land surface process model, grid-based surface runoff model, and groundwater flow model.
3. Data and Boundary Conditions for Simulation
3.1. Input Data
 The hourly observation data of downward short- and long-wave radiation, precipitation, atmospheric pressure, air temperature, air humidity, and wind speed at a reference level were calculated from 11 points of Automated Meteorological Data Acquisition System (AMeDAS) data in the catchment collected by the Japan Meteorological Business Support Center and from three meteorological stations data of NIES. The friction velocity is estimated by the bulk transfer formula. When the vapor pressure is unknown, this value was estimated from the Tetens formula by using the dew point temperature.
 The mean elevation of each 500 m grid cell was calculated by using the spatial average of a digital elevation model (DEM) of 50 m mesh (from Geographical Survey Institute of Japan) throughout the Kushiro River catchment (Figure 1). The vertical dimension was divided into 20 layers with a weighting factor of 1.1 (finer at the upper layers). The upper layer was set at 2 m depth, and the twentieth layer was defined as an elevation of −250 m from the sea surface. Furthermore, two vegetation characteristics, FPAR and LAI, obtained from MODIS satellite data (1 km mesh), were input every 8 days after validation and verification of the MODIS data by ground truth data collected at the Tomakomai Flux Tower (42°44′13.1″N, 141°31′7.1″E, mean elevation of 115–140 m), Hokkaido. For the hillslope runoff the kinematic wave theory was applied to each cell of the 50 m mesh. Then both kinematic and dynamic wave theories were applied to a stream network of 317 rivers by inputting the discharges at the upper river channels and the lateral flows into the hillslope model. The dynamic wave theory was applied to a large flat area around Kushiro Mire, where stream slopes generally do not exceed 3/1000 [Samuels and Skeels, 1990; Meselhe and Holly, 1997].
 Constant head values were used at the upstream boundaries (Lake Kussharo, annual mean elevation of 121 m; Lake Mashu, annual mean elevation of 351 m; northern edge of simulation area) for the groundwater flow model. About the other upstream boundaries where there are no lakes or observation points, reflecting condition on hydraulic head was used supposing that there is no inflow from the mountains in the opposite direction. At the sea boundary, constant head was set at 0 m (southern part of simulation area). Groundwater levels at about 150 sampling points obtained by Ohata et al.  were used in well cells. The hydraulic head values parallel to the ground level were inputted as the initial condition for the groundwater flow model. For the model simulating hillslope hydrology the flow depth and the discharge at the uppermost ridges of the mountains were set as zero throughout the simulation. In river cells, outflows from the riverbeds of −1 m mean elevation from the ground level were considered.
3.2. Observed Data
 Three meteorological stations, 30 groundwater level meters, and 13 flow depth meters were set throughout the overall catchment for the calibration and validation of numerical results. Three meteorological stations were established in vegetation typical of the Kushiro River catchment (mire, mean elevation of 8 m; grassland, mean elevation of 187 m; forest, mean elevation of 127 m) (Figure 1). Meteorological variables were automatically recorded hourly at each station. The data were collected from 1 April to 31 December 2001. The variables measured were air temperature (Kona-System, KDC-S2), humidity (Kona-System, KDC-S2), wind speed (Makino-Keiki, AC750), net radiation (Eikou-Seiki, CN-11), albedo (Eikou-Seiki, MR-22), precipitation (Ikeda-Keiki, RH-5), soil temperature (Chino, platinum), soil moisture (Delta-T, ML2x, and PR1/6), and groundwater level (Kona-System, Kadec-Mizu-II). To measure soil water content at each station, soil samples were taken at three depths (0.1, 0.2, and 0.3 m) to calibrate the values measured by water content reflectometers (ML2x) and profile probes (PR1/6) on the basis of time domain reflectometry. Because the correlation coefficient between soil water content measured by reflectometer and profile probe was better in grassland (Rr = 0.956), the measured data were calibrated against sampled data. In contrast, the correlation was not so good in forest (Rr = 0.600) because of the large amount of bamboo and macropores, which created difficulties in taking measurements.
 Changes in groundwater levels at 30 sites and river flow-depths at 13 sites were measured every hour during the same period. The water level (Kadec-Mizu-II) was automatically recorded in data loggers (LS-3000PtV) every hour. H – Q curves (usually with almost a 1:1 relationship between water level and stream discharge) were used to convert water level into river discharge for flow depth measurements in rivers. The relationship was poor in the mire because beds and cross sections were frequently variable in rivers there, each river flowing into the mire has its own of runoff hydrography, and bed slopes are very small and are thus affected by downstream flow depths and floods. To validate MODIS data against ground truth data, measurements were provided by the Center for Global Environment Research (CGER), NIES, for surface temperature (Minolta, R505) at 15 m above the ground surface, FPAR (Li-Cor, LI-190s) at 25 and 40 m, and net radiation (Eikou-Seiki, MR-40) at 25 m, recorded at the Tomakomai Flux Tower, which stands in a coniferous forest with a height of ∼15–20 m, mainly larch (Larix).
3.3. Estimation of Heat Flux Budgets from Meteorological Data
 The heat flux budgets were evaluated from the measured meteorological data at the grassland station. Equation (26) describes the relationship between the heat flux at the ground surface (with vegetation) based on net radiation (NR) (W m−2), latent heat flux (LH) (W m−2), sensible heat flux (SH) (W m−2), and ground transfer heat flux (GH) (W m−2):
In equation (26) the heat flux transferred to the ground surface is set at a positive value. NR (short-wave radiation plus long-wave radiation) is defined as follows:
where α is surface albedo, SR (W m−2) is solar radiation, R (W m−2) is atmospheric radiation, σ (= 5.56 × 10−8 W m−2 K−4) is the Stefan-Boltzmann constant, and ɛ (≈1) is emissivity. NR was directly measured by net radiometer (Eikou-Seiki, CN-11).
 GH is estimated as follows:
where ρs (kg m−3) is the density of soil, Cs (J kg−1 K−1) is the specific heat of soil, Tm is the mean soil temperature, zs (m) is the depth of the heat flow plate, and Qg is the heat flow on the soil measured by heat flow plate. The second term on the right-hand side of equation (28) is the temporal variation of heat storage between the ground surface and depth zs, estimated from the temporal gradient of soil temperature.
 Both LH and SH are estimated from the vertical gradients of measured values at two points by using a gradient method [Oke, 1990]:
where is Karman constant (≈0.41), ρa (kg m−3) is the density of air, Ca (J kg−1 K−1) is the specific heat of air, Ta (K) is air temperature, Wa (m s−1) is wind speed, qa is specific humidity of air, and za (m) is the measurement height of Ta, Wa, and qa from the ground. Φ is a nondimensional function displaying the degree of aerodynamic stability (stable, neutral, or unstable), and the subscripts M, H, and E refer to momentum, sensible heat, and latent heat, respectively. There is an empirical correlation between (ΦMΦX)−1 in equations (29) and (30) when the Richardson number Ri is used as follows (X = H or E):
where (K) is the mean temperature of two points. If Ri = 0 (neutral condition), (ΦMΦX)−1 = 1.
3.4. Vegetation and Soil Properties
 About 50 vegetation and soil parameters were calculated in each cell on the basis of vegetation class and soil texture obtained from the Digital National Land Information GIS data of Japan for 1993 [Clapp and Hornberger, 1978; Rawls et al., 1982]. The major parameters include vegetation cover, green fraction, albedo, surface roughness length and zero displacement height, soil conductivity and soil water potential at saturation, and some parameters of stomatal resistance that relate to environmental factors. Soil texture data were digitized and categorized from a soil map of arable land in Hokkaido [Hokkaido National Agricultural Experiment Station, 1985] into seven types for SiB2 and were converted to a 1 km mesh (Figure 4). The red square shows the study area for drying phenomena, identical to Figure 1. The soil of Kushiro Mire is mainly peat. Vegetation class data categorized into 11 types, identified by the Environment Agency of Japan , were also converted to a 1 km mesh (data not shown). There is natural vegetation where vegetation class takes higher value.
 Soil samples were taken at two depths (0.1 and 1.0 m) to identify soil hydraulic permeability, geological structure, and water content at eight sites (Figure 1 and Table 1). By using these soil samples and about 150 sample data points [Ohata et al., 1975], geological structure was divided into four types on the basis of hydraulic conductivity (Kh and Kv), the specific storage of porous material (Ss), and specific yield (Sy) after the calibration for fitting simulated hydraulic heads to the observed heads, while keeping these values in the initial estimated known range (Table 2). Kushiro Mire consists largely of soils finer than silt, mainly peat.
Table 1. Results of Soil Property and Field Permeability Testa
Hydraulic Conductivity, cm s−1
Read 4.43E-05 as 4.43 × 10−5.
organic soil and volcanic ash sand
organic soil and volcanic ash sand
volcanic ash sand
volcanic ash sand
volcanic ash sand
gravel & sand
volcanic ash sand
volcanic ash sand
Table 2. Geological Parameters Used in Numerical Simulationa
Horizontal Hydraulic Conductivity Kh, m h−1
Vertical Hydraulic Conductivity Kv, m h−1
Specific Storage Ss, m−1
Specific Yield Sy
Read 5.0E+01 as 5.0 × 101.
coarser than coarse sand
fine to medium sand
finer than silt
3.5. MODIS Data
 The MODIS land surface temperature (LST) data were converted to surface temperatures T0 by using the NASA MOD11-ATBD (algorithm theoretical basis documents) equation [Wan, 1999]. FPAR and LAI were calculated by using MOD15-ATBD from the surface reflectance product (MOD09) and the land cover type product (MOD12): T0 = 0.02DN1, FPAR = 0.01DN2, LAI = 0.1DN3, where T0 is surface temperature (°C), FPAR (μmol m−2 s−1) is fraction of photosynthetically active radiation, LAI is leaf area index, and DNi (i = 1, 2, 3) is the digital number of MODIS data for T0, FPAR, and LAI, respectively.
 Because MODIS has more data channels than previous satellites, such as Landsat, it is possible to analyze higher-order products, such as LAI and FPAR, which are important parameters for evaluating vegetation growth [Christopher et al., 1998]. Because the MODIS data are recorded at about 1030 LT every 8 days, we averaged the Tomakomai Flux Tower data at 1000, 1030, and 1100 LT for comparison with MODIS data. The MODIS data were averaged over each 3 × 3 squares of pixels because there is transformation error from sinusoidal to universal transverse Mercator coordinate system after noise and null values due to bad weather were eliminated. MODIS data were compared with the synchronized ground truth data at the Tomakomai Flux Tower for surface temperature (°C) and FPAR (μmol m−2 s−1). The estimate of surface temperature is highly accurate (Rr = 0.992). Although the MODIS FPAR value underestimates the observed value in winter (mainly because of higher cloud cover and falling snow), the correlation is still good (Rr = 0.788) (data not shown).
 MODIS LAI and FPAR (μmol m−2 s−1) images of the Kushiro River catchment were obtained every 8 days in the snow-free period (1 May to 31 October) of 2001 (1 km mesh; image areas) for input into the model after noise was removed and moving average was applied (Figures 5a and 5b). Seasonal changes in LAI and FPAR, particularly in the mire, imply that the vegetation phenology and the water cycle are closely related in the mire. Both parameters take their maximum values (green) from early summer to fall while vegetation is growing and decrease toward winter, and both variables are highly correlated (Rr = 0.891). LAI and FPAR values can integrate the net assimilation rate (subtracting leaf respiration rate from leaf photosynthetic rate) for leaves at the top of the canopy into the total assimilation rate over the depth profile of leaf multiplied by Π [Sellers et al., 1996]. Π is calculates as follows:
where is temporal mean (radiation weighted) extinction coefficient for PAR and gr is the greenness of canopy. Both LAI and FPAR values were input into the model by improving the original SiB2 model developed by Sellers et al. . The greenness of canopy was calculated by using equations (33) and (34) and input into equation (32), whereas previous research used a greenness of a constant value.
where ωv is leaf scattering coefficient in the visible wavelength interval, Kd is optical depth of direct beam per unit leaf area, ωi,j is the leaf reflectance, and δi,j is the leaf transmittance. In αi,j and δi,j, i = 1 and 2 denote visible and near infrared, and j = 1 and 2 denote live and dead, respectively. Consistent canopy conductance can be calculated because SiB2 combines a photosynthetic model with a leaf stomatal conductance model, and calculates the canopy transpiration rate and surface energy balance [Sellers et al., 1996].
3.6. Running the Simulation
 The simulation area is 50 km wide by 80 km long, covering the whole Kushiro River catchment (Figure 1). This area is discretized into a grid of 100 × 160 blocks, with a grid spacing of 500 m. The simulation was conducted on an NEC SX-6 supercomputer. Since the effects of a snow layer and a freezing/thawing soil layer on winter runoff and spring snowmelt runoff were not included, simulations were performed for the snow-free period of 6 months from 1 May to 31 October 2001. The first 6 months were used as a warm-up period until equilibrium conditions were reached, and parameters were estimated by comparison of simulated steady state values in steady state condition with the observed values published in the literature. A time step of Δt = 1 hour was used.
4.1. Soil Moisture in Various Land Covers
 Simulated results of soil moisture for different land covers (mire, grassland, and forest) were compared with observed values and the precipitation distribution from 1 August to 31 October 2001 (Figures 6a–6c). The higher precipitation around 10 September was due to typhoons. Soil moisture fluctuation in response to changes in precipitation, evaporation, and redistribution decreases in deeper layers, indicating that the influence of meteorological forcing is smoothed and the response times become longer with depth [Li and Islam, 1999; Yu et al., 2001]. In grassland (Figure 6b), where vegetation and soil structures are simpler, the simulation accurately reproduces the observed data at soil depths of 10 cm and 1.0 m. In forested areas (Figure 6c) the porosity changes greatly in the vertical direction owing to a more complex root distribution, which suppresses the response of the observed soil moisture at 1.0 m. The simulated value excellently reproduces the observed data at 10 cm. However, it does not reproduce it at 1.0 m with higher precipitation because porosity and hydraulic conductivity are constant in the vertical direction in this model despite the change in soil texture with depth [Yu et al., 2001].
 The surface soil moisture content in forest is higher than in grassland, indicating that the forest land cover retains more water in soil and vegetation than grassland does, and that grassland tends to quickly lose infiltrated water to the groundwater system and evapotranspiration. Furthermore, the differences in land cover become less marked with depth. In the mire (Figure 6a) the simulation takes on an almost constant saturated value (≈1.0) owing to the shallow groundwater level calculated by MODFLOW. This simulated value agrees well with the observed value of ∼0.99. This high value is characteristic of a mire because of the poor drainage at lower elevations, the almost flat surface, the peaty soil texture, and the soil's elasticity [Kellner and Halldin, 2002]. The simulations of soil moisture reproduce very well the measured values in the mire, grassland, and forest.
4.2. Soil Temperature and Heat Flux Budget
 The simulated values of soil temperature in mire, grassland, and forest from 1 September to 30 September 2001 at a depth of 30 cm agree strongly with the measured values (Figures 7a–7c). The simulated value at a depth of 10 cm in grassland does not reproduce the observed value around 15 September because of the occurrence of a typhoon, because rapid infiltration flow increases the effective heat capacity of soils despite the inclusion of soil moisture and soil porosity effects in SiB2 [Sellers et al., 1996]. To strongly reproduce the soil temperature, it is necessary to solve the simultaneous heat transfer equations for both the water and soil layers with a finer mesh. In September the temperature in the grassland is highest because the forest trees prevent the sunshine from warming the ground and the mire is almost saturated with a higher effective heat capacity. Globally, the simulated values of the heat flux budgets (W m−2) in grassland excellently reproduce the observed values from 1 to 11 August 2001, despite small discrepancies in sensible heat flux (SH) and latent heat flux (LH) (Figures 8a and 8b). Therefore the soil and vegetation parameters were correctly selected in the simulation, and the simulation model accurately described the vegetation and soil structures (Figures 6–8).
4.3. Groundwater Levels
 Simulated hydraulic heads in steady state average for the overall Kushiro River catchment were plotted against the measured values [Ohata et al., 1975]. Hydraulic head becomes smaller as the riverbed hydraulic conductance kr (m2 h−1) becomes larger (Figure 9). The simulated groundwater level for kr = 300 m2 h−1 agrees closely with the measured value, and therefore this kr value was used in the following simulations. In the vicinity of the Kushiro Mire (wells h-60 to 120), groundwater almost saturates the surrounding soil, which can be reproduced very well by the numerical simulation. Generally speaking, because a three-dimensional groundwater model is used in the current study, the simulation values of groundwater fluctuations agree strongly with the measured values from 1 May to 31 October 2001, which depend not only on precipitation but also on local topography (Figure 10). The large amount of precipitation due to typhoons in September 2001 greatly affected the groundwater fluctuations in all areas of the catchment. In mountainous areas away from Kushiro Mire (O-1 and I-1) the groundwater level decreases gradually from spring to summer after snowmelt. However, around the mire (Ku-2 and W-5) the groundwater level is almost constant from spring through the typhoon season, as can also be seen with soil moisture in Figure 6a, which shows the high soil water capacity in the mire [Winter, 1988; Kellner and Halldin, 2002]. The simulated groundwater levels reproduce actual levels very accurately, both in the mountainous areas and near the mire, owing to the contribution of recharge rates in the soil moisture model in the upper layer.
 The overall variations in both soil water content and groundwater levels are very similar, namely, a rapid increase before precipitation and a gradual decrease after precipitation (Figures 6a–6c and 10); similar fluctuation patterns were observed by Chen et al.  and Eltahir and Yeh . However, the groundwater level shows its peak only on 10 September with small fluctuations (Figure 10), whereas soil moisture fluctuates sharply in response to precipitation (Figures 6a–6c). The groundwater acts as a long-term reservoir storing excessive soil water. Furthermore, the simulations do not consider short-term outflow rigidly, causing a poor response to short-term groundwater fluctuations (W-5), which implies that it is necessary to improve the numerical models to include the effect of short-term outflow. The short-term groundwater fluctuations are due to the locations of groundwater gauges near river flows.
4.4. River Discharge
 The simulated discharges of both the kinematic wave (dashed line) and dynamic wave (solid line) methods from stream network modeling are compared with observed values (open circles) at several points in the Kushiro River catchment from May 1 to October 31 2001 (Figures 11a and 11b). In Figures 11a and 11b the observed discharge decreases gradually in the same way as the groundwater levels in the mountainous areas from the spring to the early summer [Ohata et al., 1975]. In the late summer and fall the discharge fluctuates greatly, depending on volume of precipitation from typhoons.
 The simulated river discharge at the Kucyoro water flow survey station in the Kucyoro River (bed gradient here is about 1/2000), a tributary of the Kushiro River flowing into the Kushiro Mire, reproduces very well the observed value in the typhoon period, and the differences between kinematic wave and dynamic wave models are smaller, showing that the dynamic wave effect is small at this point despite the lower bed gradient (Figure 11a). Furthermore, the base flow can be calculated correctly because the grid-based distributed model used includes recharge rates, seepage, and the return flow from the ground. However, the simulation cannot reproduce the observed data during the melting period, in particular, in early May, and the simulated discharge is smaller than the observed value. The integrated model in this study does not completely include the melting process of snow volume to surface flow, a necessary consideration in the future.
 At the Gojikkoku water flow survey station downstream of the main Kushiro River region (bed gradient is about 1/1700) the dynamic effect is important because of the influence from the Kushiro Mire (Figure 11b). The peak discharge comes later and milder (but with a larger volume) in the downstream regions (maximum delay is ∼5 days) owing to the lower discharge rate from the mire [Winter, 1988; Kellner and Halldin, 2002], which is 50–60% of the flow rate in the mire [Ohata et al., 1975]. The simulated results of the kinematic wave model overestimate the observed values, especially during the rainy periods. The dynamic wave model reproduces excellently the peak value of discharge in precipitation and the backwater effect after a flood. Thus the grid-based distributed model used is highly accurate for different topographies because it includes surface unsaturated-saturated water processes displaying both short- and long-term components of river discharge, effective precipitation, and the kinematic and dynamic wave effects at various river slopes.
4.5. Soil Moisture Changes (Drying) From 1977 to 2001
 The size of the simulated area for monthly averaged groundwater and soil moisture around the Kushiro Mire in 2001 (the same study area for drying phenomena shown in Figures 2a and 2b) is 14.0 km long by 10.0 km wide (Figures 12a and 12b). This region is a low-elevation mire surrounded by capes and low mountains, of which ∼10–20 km2 floods in very wet weather (Figure 2a). In those times, large amounts of sediment and gravel flow into the mire and piles are formed, as evaluated by the WTI (water turbidity index) from a Landsat TM image [Kameyama et al., 2001], showing the higher turbidity area during flood periods.
 The groundwater levels fluctuate seasonally and spatially (Figure 12a), and the monthly averaged levels were greatly affected by the typhoons in September 2001 (Figure 10). In particular, the regions of low-level groundwater closely correspond to the regions of higher turbidity at the southern and eastern sides of the Kirakotan Cape [Kameyama et al., 2001]. The eastern side of the cape is located downstream of the channelized river (Figure 2a); therefore the coarser sediments are deposited there as the flow velocity decreases. In contrast, on the southern side of the cape, floodwater spreads out to the lowest areas, depositing the finer sediments widely. In these regions the simulation showed that the outflow of groundwater to the river and the inflow of recharge became smaller because of the lowered groundwater levels. The simulated soil moisture for 2001 (Figure 12b) takes a greater value near the Kushiro Mire, especially from summer to fall, than in the surrounded areas, with a value >0.80. In the simulation the regions of higher soil moisture in the northeast and southwest correspond closely to those with higher groundwater levels (Figure 12a). Furthermore, the region of lower soil moisture near the mire closely depends on the region of higher turbidity and the invasion of alder (Figure 2b).
 In the simulation of groundwater levels and soil moisture for 1977 (Figures 13a and 13b) the same data of soil texture and geological structure were used as those for 2001. Only meteorological forcing data and the vegetation classes were changed, on the assumption that predominant changes in the underground structure had not occurred as had changes in vegetation phenology from 1977 to 2001. The simulated 6 month (1 May to 31 October) averaged groundwater levels (Figure 13a) are greatly affected by the local topography and fluctuate more as the slope becomes less steep, in particular near the Kushiro Mire. In some areas near the Kushiro Mire the low-lying region is inundated and has higher soil moisture throughout the year, which is characteristic of mire that is covered mainly by reeds. The groundwater levels in 1977 and 2001 are almost similar, but a modest decrease in groundwater levels can be observed near the alder invasion area (circled area) owing to the deposit of sediment load and the decrease of recharge rate to the groundwater. The simulated annual averaged soil moisture (Figure 13b) clearly indicates the drying phenomena near the circled area due to the invasion of alder (Figure 2b).
5. Discussion and Conclusion
 The NICE model shows extremely high accuracy in simulating river discharge, soil moisture, and groundwater flow over the Kushiro River catchment during the snow-free period of 6 months. The use of very accurate field measurement data, MODIS data, flux tower data, and various parameters categorized by GIS and geological structure support this precision. Seasonal vegetation change, mechanisms of vegetation-water relations, and surface unsaturated-saturated water processes were included in the model. This model explains water cycle change and drying phenomena in the Kushiro Mire associated with vegetation change.
 Although the overall simulated values excellently reproduced the observed values, some discrepancies existed. The land surface model simulation could not reproduce the rapid change in soil moisture in the vertical direction at higher precipitation (Figures 6a–6c). This occurred because the vertical mesh size was rough and porosity/hydraulic conductivity were treated as constants in the vertical direction in the unsaturated layer in this model, even though the soil texture changes with depth [Yu et al., 2001]. The rapid change in soil moisture can be reproduced more excellently when Richards' equation is solved with a finer mesh, the soil structure is evaluated by measuring pF moisture characteristics, and the vegetation structure including the root depth and density is better parameterized. In the groundwater flow model the rapid change of groundwater level is difficult to reproduce in the same way as soil moisture because correct recharge rates from the unsaturated layer are not attainable by observation (Figure 10). For better recharge rate simulation the unsaturated layers must be more correctly simulated. More groundwater level data are necessary near mountain areas because boundary conditions for groundwater levels are difficult to input correctly except with constant head values at lake or sea level.
 The effect of the snow layer and the freezing/thawing soil layer during spring snowmelt runoff must be included in the surface hydrology model to simulate the entire year because the spring surface runoff, which occurs when the ground is still partly frozen during the snowmelt period, can be a significant component of the water balance in the Kushiro River catchment. Simulating the stream network by using the dynamic wave theory requires a lot more computation time than by using kinematic wave theory (Figures 11a and 11b). In order to stop the sediment load influx and to recover the Kushiro Mire a new project started to remeander the channelized rivers in 2002. In most cases, kinematic wave theory reproduces primarily river discharge, except after a flood when the backwater effect is observed (Figure 11b). However, the dynamic wave theory is very important for the simulation of meandering rivers because the backwater greatly affects sedimentation near the meandering rivers.
 The water cycle and heat flux processes relate implicitly to the change in vegetation phenology, which is excellently modeled by the model. MODIS data, which have multiple channels and broader areas, are spatially moderate in resolution (1 km mesh) (Figures 5a and 5b), and it is required to composite 8 days of MODIS data of biological products such as LAI and net primary production (NPP) because there is noise due to cloud, rain, and snow. Despite these difficulties in MODIS data the vegetation changes are successfully observed spatially and seasonally around the Kushiro Mire. However, some aspects of the MODIS data need improvement. FPAR and LAI are calculated from MOD09 and MOD12, which are based on only six types of land cover (grasses/cereal crops, shrubs, broadleaf crops, savanna, broadleaf forests, and needle forests), as in MOD12-ATBD [Strahler et al., 1999]. Mires and paddy fields were forced to fit into these six categories. In future work, FPAR and LAI values for mires and paddy fields will require the new classification of land cover, and it is necessary to add the growth process of vegetation to the model in order to refine the water cycle mechanism determination for the area of the Kushiro Mire.
 From the comparison between simulated and observed results in snow-free periods between 1977 and 2001 (Figures 13a and 13b) the drying phenomena clearly occurred in the area where alder dominated (Figure 2b). This shows that alder absorbed more water from the roots and transpired more to the atmosphere than original mire vegetation of reeds and that the soil moisture decreases dramatically in the alder because the recharge rate to the groundwater decreases. Furthermore, the area of lower soil moisture and lower surface temperature closely corresponds to the area covered by alder (Figure 2b) (data not shown). The simulation clearly demonstrates that this drying phenomenon is closely related with the increased influx of sediments from the surrounding area, where agricultural development, reclamation, and channelization of the river occurred [Environment Agency of Japan, 1984, 1993]. Thus the positive correlations between soil moisture and groundwater levels indicate that the inflowing sediments during flood periods formed spatially distributed piles, and resistance to water supply from the surroundings was increased. Consequently, the soil was drying and the local groundwater levels were lowering, resulting in a further invasion of alder in this area. The areas of alder promote the deposit of sediments by their overland roots and shed leaves. The topography changes owing to sediment deposits, and the nutrient infiltration processes are important for the long-term simulation in order to clarify the relationships between water, heat, vegetation, sediment, and nutrients, and ultimately, to reproduce the invasion of alder in the Kushiro Mire (Figure 2) due to the channelization of rivers. The succession from mire vegetation (for example, reeds) to alder and finally to willow trees occurs during the mire's drying process. The extension of the model will be very powerful in estimating the effect of river channelization on the Kushiro Mire, in predicting the recovery of groundwater recharge by remeandering the channelized rivers, and in protecting the sediment loads from riparian forests, which are most sensitive for soil moisture and the succession in mire vegetation among alder, reeds and willows.
 We thank S. Murakami, Q. Wang, S. Hayashi, and S. Kameyama, National Institute for Environmental Studies (NIES), Japan, for valuable comments for setting observation systems and numerical simulations. Y. Ichikawa, Kyoto University, was helpful about GeoHyMoS (Geomorphologically-based Hydrological Modeling System), which is the base on the surface hydrology submodel of the integrated model. Some of the simulations were run on an NEC SX-6 supercomputer at the Center for Global Environment Research (CGER), NIES. We also thank CGER for providing Tomakomai Flux-Tower data.