DEVELOPMENT AND APPLICATION OF A REMOTE SENSING MODEL FOR ESTIMATING SOIL WATER STATUS AND EVAPOTRANSPIRATION IN SEMIARID REGION

Transcription

1 DEVELOPMENT AND APPLICATION OF A REMOTE SENSING MODEL FOR ESTIMATING SOIL WATER STATUS AND EVAPOTRANSPIRATION IN SEMIARID REGION YANJUN SHEN*, AKIHIKO KONDOH +, CHANGYUAN TANG #, JIEYING XIAO #, TAIKAN OKI*, SHINJIRO KANAE* *JST/CREST, Hydrol. and Water Res. Eng. Division, Institute of Industrial Science, the Univ. of Tokyo, Komaba 4-6-1, Meguro, Tokyo , Japan + Center for Environmental Remote Sensing, Chiba Univ.y, Yayoi 1-33, Inage, Chiba , Japan # Graduate School of Science and Technology, Chiba Univ., Yayoi 1-33, Inage, Chiba , Japan ABSTRACT: Accurate estimation of evapotranspiration in a regional scale is very important for agricultural water management, drought monitoring, and fire risk assessment. This is more important and urgent in the North China Plain, where water shortage is becoming the biggest barrier to sustainable development. So, right monitoring the water status of land surface is necessary for making water use plan and improvement of water management. This study attempts to develop a model for estimating soil moisture status and evapotranspiration at regional scale through assimilating satellite remote sensing data. Based on the concept of soil water deficit and a dual-source conceptual evapotranspiration model relation to soil moisture dynamics, this model is composed of two sub models of calculating soil water deficit inde and potential evapotranspiration. Application to North China Plain demonstrates the proposed model plays a good performance by using 13 sub-scenes of Landsat TM and ETM+ data and ground micrometeorological observation. INTRODUCTION Nowadays, we can fairly well understand the major hydrological processes at a plot scale, however, knowledge of these processes, e.g. evapotranspiration or infiltration, at a catchment or larger spatial scale is indispensable for prediction of floods or assessing the impact of droughts Choudhury [4]. In an agricultural region, it is more important for efficient water management to instantly and accurately obtain the spatial information, such as evapotranspiration and soil water status, as well. In the past decade, remote sensing has been widely used in estimating evapotranspiration and heat balance in a large scale. Generally, the evaporanspiration algorithms can be divided into 3 categories: empirical approach (e.g., Kondoh and Higuchi [8]), energy balance approach (e.g., Moran et al.[9]; Chouldhury [4]; Bastiaanssen et al. [2]), and sophisticated SVAT model with assimilating remote sensing data (e.g., Zhan and Kustas [16]). Due to its limitation at accuracy and applicability, the empirical method is not widely used. Also, the application of SVAT model in remote sensing is very difficult to implement in a large heterogeneous region without considerable amount of surface information for all the input parameters, even though it has high accuracy. In the literature of remotely estimating land surface 1

2 evapotranspiration, the VITT method proposed by Moran et al. [9] has attracted wide attentions, but due to the difficulties in calculating the theoretical four characteristic points, the accuracy of VITT method is limited. Here, we proposed an improved VITT method for determining the soil water status in irrigated agricultural region under semiarid climate and applied it to estimating the energy and water fluxes from land surface. At the same time, the dual-source conceptual evapotranspiration model developed in the former chapter will applied to estimate the potential and then actual evapotranspiration as a theoretical base. At last, the remotely estimated energy and water fluxes are compared with the ground surface measurements for validation. ALGORITHM DESCRIPTION In this research, we assume that no significant advective heat flow occurred near land surface and use a pixel-based energy balance algorithm for estimating the surface evapotranspiration in a regional scale. The schematic flow of the algorithm is illustrated in Figure 1. Besides remote sensing data, some ancillary data such as air temperature (Ta), Vapour Pressure (e a ), and wind speed (U) are used in the algorithm. 2 Figure 1 Schematic flowchart of the land surface evaporation and transpiration estimating algorithm First of all, the geographical and atmospheric rectified satellite image is processed with multi-spectral analysis to obtain the initial maps, e.g. NDVI, surface temperature (), albedo, and net radiation (Rn). Second, using the information for vegetation cover and soil thermal properties from NDVI,, and albedo calculate the ground heat flux (G). Then, assimilating some ground meteorologically ancillary data such as air temperature, vapour pressure, and wind speed, with the precedent maps of Rn, G, NDVI, and, using Penman-Monteith method to calculate potential latent heat (LEp). The soil water stress status map is obtained through analyzing VI- scattergram. Finally, to computing actual evapotranspiration using the pixel-based potential ET and soil water deficit index (WDI).

3 Determination of soil water deficit inde WDI The soil water deficit index (WDI) is determined by analyzing the spatial distribution relationship of surface temperature fractional vegetation cover (Shen [13]). The WDI is defined as WDI( min ( ( ( ( min max (1) where, min and max are the theoretical minimum and maximum surface temperature, at which the evapotranspiration of land surface are equal to potential ET and 0 respectively, under a given vegetation cover; The notation of ( means a specific parameter varies in the horizontal space domain with a resolution of one pixel. The parameter without ( notation is considered as constant spatially. And, min and max can be expressed as = a + b f ( (2) min ( 1 1 max ( 2 2 v = a + b f ( (3) v where, a 1, a 2, b 1, and b 2 are parameters; f v is the fractional vegetation cover of one pixel, and calculated from NDVI. Estimation of net radiation, Rn The net radiation, R n (, is the principal energy source that drives the land surface heat fluxes. It is the algebraic sum of all incoming and outgoing radiation integrated over all wavelengths, and can be calculated as: R ( = R (1 r ( ) + L L ( (4) n s o where, R s is the global radiation; r o is the surface albedo. L and L are the downwelling and upwelling long wave radiation, respectively. R s is from the surface observation. The surface albedo, r o, is estimated from multi-spectral information obtained by satellite.for Landsat remote sensing data, we use the following equation to calculate the albedo (Wang et al. [15]). r0 ( = B1( B2( B3( B4( B4( B5( B7( (5) where, B1 to B7 are the surface reflectance at nadir calculated from TM/ETM+ band 1 to band 7 data. In the present study, we employ the equation (6-5) to calculate land surface albedo. The surface efflux of longwave radiation, L, is a function of surface temperature, T s (, and the average surface emissivity, ε(. According to Stefan-Boltzmann law, the surface longwave eradiation can be calculated as 3

4 4 4 s L ( = ε ( σ T ( (6) where, σ is the Stefan-Boltzmann constant. The downward atmospheric longwave radiation, L, is determined predominantly by the humidity and temperature profiles through the atmosphere. Here, we use a simply empirical method proposed by Brutsaert [3] and modified by Culf and Gash [5] to determine the atmospheric downward radiation. e 4 1 / 7 = 1.31 σ a ( ) (7) T a L T where, e is the vapour pressure, and T a is the near surface air temperature. The surface emissivity of a partial vegetated pixel is described as a linear combination of surface emissivity of vegetation cover and bare soil (Valor and Caselles[14]). ε = f ε + (1 f ) ε + 4 dε f (1 f ) (8) ( v veg v soil v v where, ε veg and ε soil are the emissivities of full vegetation covered and bare soil surface; dε is vegetation structure parameter. The values of ε veg, ε soil, and dε are set to 0.99, 0.91 and 0.02, respectively, as suggested by Valor and Caselles [14]. Estimation of soil heat flu G The soil heat flux (G) is the energy used for warming or cooling the subsurface soil volume. The previous investigations have shown that mid-day G fraction is reasonably predictable from remote sensing determinants of vegetation characteristics, and can be expressed as (Daughtry et al. [6]): G = C (9) R n where C is the soil heat flux/net radiation fraction (G/R n ). However, the attenuation of radiative and conductive heat transfer in canopy and soil, respectively, changes significantly with soil cover. From the literature in study of soil heat flux/net radiation fraction, it is found that the fraction of G/Rn shows a non-linear relation with NDVI. In this study, we adopt the method suggested by Bastiaanssen et al. [2], which integrated the effects of surface temperature, albedo, and NDVI, and expressed as G( ( 2 4 = (0.0032r o( r o ( ) ( NDVI( ) (10) R ( r ( n o For bare soil, the Equation 12 is used for calculating soil heat flux. G=0.23*Rn (11)

5 5 Estimation of leaf area inde LAI Kanemasu et al. [7] established a simple relationship between the fractional vegetation cover and leaf area index for various agricultural plants: LAI( = a + b ln(1 f ( ) (12) v where, a and b are empirical coefficients. We analyzed the ground observation results of LAI and f v from space remote sensing and specified a and b as and for wheat, and for maize, respectively. Calculation of potential latent heat flu LEp It is of importance to calculate correctly potential evapotranspiration of the land surface for estimating the actual evapotranspiration and then therein energy fluxes accurately. Here, we use Penman-Monteith method to calculate potential evapotranspiration. So, the potential latent heat flux (LE p ) can be written as LE p ( Rn G) + ρ Cp D / ra = (13) + γ (1 + r / r ) cp a where, r cp is canopy resistance at the situation of potential evapotranspiration; r a is aerodynamic resistance; D is vapour pressure deficit; and γ are the slope of saturated vapour pressure-air temperature curve and the psychrometric constant. The resistance of plant canopy under potential evapotranspiration, r cp, is calculated from the following equation (Allen et al.[1]). r cp rsm = (14) a LAI r sm is minimum stomatal resistance. It is suggested that 0.5 is an appropriate value of a for grass (Allen et al., [1]). However, since the resistance of woody or herbaceous natural vegetation is approximately double that of agricultural crops (Radersma and Rider, [10]), we use the value of 0.25 after Rey [11]. The minimum stomatal resistance is estimated according our field measurements as described in Shen et al. [12]. The values of minimum stomatal resistance for different growing seasons are listed in Table 1. Table 1 The minimum stomatal resistance used in this study (Unit: s/m) Growth stages ~ Jointing Heading Blooming Milking Maturing Winter wheat Maize Calculation of aerodynamic resistance, r a An essential parameter in estimating surface heat or vapor flux is aerodynamic resistance (r a ), the resistance to vapor transferring from near surface air to boundary layer atmosphere.

6 6 In neural conditions, the aerodynamic resistance, r a, is usually estimated as z d 2 2 ra = (ln ) /( k uz ) (15) z 0m where, z 0m and d are roughness length for momentum transfer and zero displacement height, respectively; z is reference height; k is Von Karman s Constant; u z is wind speed at the reference height. APPLICATION TO NORTH CHINA PLAIN Data and processing Totally, 13 sub-scenes of Landsat TM and ETM+ images are used in this study. Each image covers around a region of 1000 km 2 near our ground experimental site. Basically, the images can well demonstrate the major growth stages of winter wheat and maize through a whole crop-year. Before calculate the energy and evapotranspiration fluxes, all images were preprocessed for geometric rectification (or geographical encoding) and atmospheric radiance rectification. Remotely estimatedevapotranspiration and energy balance Figure 2 shows the results of remotely estimated evapotranspiration and energy balance at our experimental site. The results of Apr. 5, 2000, Sep. 12, 2000, Apr. 11, 2002, and Apr 27, 2002 are also plotted in the figure. The results show very smooth seasonal changes. The change of LE shows 2 peaks at April and August respectively. Figure 2 The estimated results of land surface energy balance at LESA VALIDATION Accuracy of the proposed algorithm is strongly dependent on the accuracy in estimating soil water status, WDI, and the potential evapotranspiration, LEp. In order to assess the

7 accuracy of the remote sensing algorithm, we compared the remotely estimated value of evapotranspiration flux and soil water deficit index with ground measurements at LESA. Soil water deficit inde WDI The relationship between WDI and extractable soil water (ESW) can be expressed as (Shen [13]) θ θ θ θ w w ( 1 WDI) = > ESW = θ * θ w θ f θ w 7 (16) where, θ, θ f, and θ w are actual soil moisture, field capacity, and wilting point, respectively; θ* is called incipient stress point of soil moisture. Figure 3a compares the soil water deficit status with ground measurements of extractable soil water at 0-40 cm layer. It is shown that most of the data lies on the up side of the 1:1 line meaning the unequal relation in Equation 16. This fact implies that the simplified method for determine WDI plays a good performance. (a) Figure3 Comparison of remotely estimated WDI and measured ESW of 0-40cm layers(a); and the remotely estimated LE with ground measurement (b). Evapotranspiration flu ET The remotely estimated latent heat flux (ET_TM) at LESA is compared with the ground measurement by Bowen ratio system (ET_BW) (Figure 3b). Figure 3b illustrates the correlation between estimated and measured ET at LESA. The Pearson s coefficient is 0.99, and regression coefficient is around The facts imply that the proposed algorithm can estimate the evapotranspiration well. CONCLUSIONS In this research, we proposed an algorithm for estimating regional evaporation and transpiration of land surface using remote sensing data. The application in irrigated agricultural region, i.e. NCP, shows a good performance when the vegetation cover is low. However, during the high vegetation cover period, it seems that the algorithm need to be improved. (b)

8 8 REFERENCES [1] Allen RG, Jensen ME, Wright JL, Burman DR Operational estimates of reference evapotranspiration. Agon. J. 81: [2] Bastiaanssen WGM, Menenti M, Feddes RA, Holtslag AAM A remote sensing surface energy balance algorithm for land (SEBAL): 1. Formulation. J. Hydrol : [3] Brutsaert W On a derivable formula for long-wave radiation from clear skies. Water Resources Research, 11: [4] Choudhury BJ Synergism of multispectral satellite observations for estimating regional land surface evaporation. Remote Sens. Environ. 49: [5] Culf AD and Gash JH Longwave radiation from clear skies in Niger: A comparison of observations with simple formulas. Journal of Applied Meteorology, 32: [6] Daughtry CST, Kustas WP, Moran MS, Pinter PJ Jr., Jachson RD, Brown PW, Nichols WD, Gay LW Spectral estimates of net radiation and soil heat flux. Remote Sens. Environ. 32: [7] Kanemasu T, Rosenthal UD, Raney PJ, Stone ThP Evaluation of an evapotranspiration model for corn. Agron. J. 69: [8] Kondoh A. and Higuchi A Relationship between satellite-derived spectral brightness and evapotranspiration from a grassland. Hydrol. Process. 15: [9] Moran MS, Clarke TR, Inoue Y, Vidal A Estimating crop water deficit using the relation between surface-air temperature and spectral vegetation index. Remote Sens. Environ. 49: [10] Radersma S and Ridder N Computed evapotranspiration of annual and perennial crops at different temporal and spatial scales using published parameter values. Agric. Water Manag. 31: [11] Rey JM Modelling potential evapotranspiration of potential vegetation. Ecol. Modelling. 13: [12] Shen Y, Kondoh A, Tang C, Zhang Y, Chen J, Li W, Sakura Y, Liu C, Tanaka T, Shimada J Measurement and analysis of evapotranspiration and surface conductance of a wheat canopy, Hydrological Processes, 16: [13] Shen Y Study on Hydrological Processes of Land and Atmospheric System in Semiarid Agricultural Region. Doctoral thesis, Chiba University, Japan. pp [14] Valor E and Caselles V Mapping land surface emissivity from NDVI: application to European, African, and South American areas. Remote sensing of Environment, 57: [15] Wang J, White K, Robinson GJ Estimating surface net solar radiation by use of Landsat-5 TM and digital elevation models. Int. J. Remote Sens. 21: [16] Zhan X, Kustas WP, Humes KS An intercomparison study on models of sensible heat flux over partial canopy surfaces with remotely sensed surface temperature. Remote Sens. Environ. 58: