Simultaneous soil moisture and properties estimation for a drip irrigated field by assimilating cosmic-ray neutron intensity
Introduction
Globally, 70% of fresh water is used by agriculture (FAO – Food and Agriculture Organization of the United Nations). Therefore, it is necessary to increase the water use efficiency and reduce the water need for crop production, while maintaining crop yield. Enough water should be applied to meet the requirement of maximum crop evapotranspiration (ET). Farmers usually base irrigation scheduling on their own experience taking into account soil water status and crop growth. However, it is unlikely that the optimal scheduling of irrigation is acquired without the knowledge of crop water needs. Low cost sensors that measure soil moisture content can be of advantage. However, these sensors typically have a very small measurement volume which is much smaller than the scale of the fields where the crops are grown. Numerical models like crop growth models (Heng et al., 2009) and land surface models (Wood et al., 2011) can be used for the quantitative estimation of the irrigation requirement under specific soil water and crop growth conditions. The estimated irrigation amount can be applied accurately with new agricultural technology like drip irrigation (Sampathkumar et al., 2012). However, uncertain model input data and deficits in the model structure result in biased estimates of soil water status, crop transpiration and therefore irrigation requirement.
The optimal scheduling of irrigation is complicated given the high heterogeneity of soil moisture content in drip irrigated fields. An estimate of soil moisture content for the complete root zone is important in this context. It is difficult to achieve this with small-scale measurements (e.g., TDR – Time Domain Reflectometry, FDR – Frequency Domain Reflectometry or TDT – Time Domain Transmission) as a prohibitively large number of sensors is needed to cover large irrigated areas. Soil moisture information from remote sensing on the other hand is limited to the upper few soil centimeters, and often has a very coarse horizontal resolution (>10 km) (Entekhabi et al., 2010, Kerr et al., 2010, Montzka et al., 2013). A further limitation of satellite-derived soil moisture content is that it is not reliable for highly vegetated areas (Njoku and Chan, 2006) and high uncertainties (Merlin et al., 2009, Montzka et al., 2013). The spatial variability of soil moisture is controlled by soil hydraulic properties, meteorological forcing, land cover patterns and topographic features at different measurement scales. Small scale variability is more driven by soil hydraulic properties while large scale variability is also more driven by the other factors. Hence, strengths and weaknesses of each measurement method rely on the additional uncertainty given by these additional controlling factors (Crow et al., 2012).
A new promising method which can determine integral root zone soil moisture from the measured above ground fast neutron intensity (defined as the number of counted neutrons per unit of time – e.g., counts per hour) has been proposed (Zreda et al., 2012). This synthetic study focuses on the assimilation of cosmic-ray probe (CRP) neutron intensity (Bogena et al., 2013, Desilets et al., 2010, Rosolem et al., 2014, Shuttleworth et al., 2013, Zreda et al., 2008, Zreda et al., 2012). Soil moisture measurements at the intermediate scale of the cosmic ray probe have the advantage that they are less affected by small scale variability of soil hydraulic properties. A further advantage is that soil moisture can be determined for a deeper layer (10–70 cm) in higher temporal frequency than remote sensing (Rosolem et al., 2014).
Primary cosmic rays originate from our galaxy and eventually collide with atmospheric nuclei, generating secondary cosmic rays mainly consisting of neutrons (Lal and Peters, 1967). Primary cosmic rays create cascades of secondary high-energy neutrons through colliding with atmospheric nuclei and the high-energy neutrons can penetrate the atmosphere and collide with nuclei in soils. These collisions in the soil generate fast neutrons. Some of these fast neutrons are eventually scattered back to the atmosphere and the fast neutron intensity can be measured with the CRP. The measured intensity of fast neutrons above the ground depends strongly on soil moisture content (Hendrick and Edge, 1966, Zreda et al., 2012). CRPs make use of this principle to estimate soil moisture content for an area of about 600 m diameter and variable measurement depth (∼10–70 cm) depending on the soil moisture conditions (Zreda et al., 2012).
Measured neutron intensities above ground need to be corrected for variations in incoming high-energetic neutrons and atmospheric pressure (Zreda et al., 2012). Moreover, as the measured neutron intensity depends on additional sources of hydrogen (besides of soil moisture), these need to be taken into account in order to isolate the soil moisture signal. Corrections have been proposed for other hydrogen sources like atmospheric vapor (Rosolem et al., 2013), lattice water and organic carbon in the soil (Franz et al., 2013), hydrogen atoms stored in the litter layer (Bogena et al., 2013) and above-ground biomass (Baatz et al., 2015). Data assimilation studies have shown the advantage of using measured multi-source soil moisture observations for improving the soil moisture profile characterization of a land surface model (Crow et al., 2008, De Lannoy et al., 2007b, Han et al., 2012, Huang et al., 2008, Reichle et al., 2008, Walker et al., 2001). Measured neutron intensities have already been used for assimilation in a land surface model to improve estimates of soil moisture profiles, but the model parameters were calibrated a priori (Han et al., 2015a, Rosolem et al., 2014, Shuttleworth et al., 2013).
In this paper we will investigate the benefits of assimilating coarse scale (600 m) neutron intensity data into the Community Land Model (CLM) for the application of drip irrigation for citrus trees on a finer scale (100 m) than the CRP scale. The neutron intensity measured by a synthetic CRP affects a larger area than a typical irrigation management unit (1 ha in this work). In order to study the impact of soil moisture data assimilation on irrigation scheduling, the drip irrigation was therefore simulated at a finer spatial scale than the footprint of a CRP. The drip irrigation was applied at the vegetated area and resulted in a very heterogeneous soil moisture distribution with the alternation of patches of wet and dry soil. It is very CPU-intensive to explicitly model the irrigated patches and the non-irrigated parts, and a simplified implementation was adopted in this work, which will be further detailed in the methodology section. In the simulation experiments, CLM was driven by biased soil properties to mimic the intrinsic model uncertainties. The coarse scale CRP neutron intensity observations were used to update the field scale heterogeneous soil moisture field through data assimilation. The joint soil moisture and soil properties (or soil moisture bias) estimation scheme was evaluated. This is important because soil moisture content and crop transpiration are sensitive to model parameters (Hou et al., 2012, Rosolem et al., 2012, Schwinger et al., 2010). Typically, field measurements of parameter values are scarce and very uncertain, especially because of the scale mismatch between a local measurement and the model scale (Waller et al., 2014). Model parameter estimation in the context of a data assimilation framework was proven to be successful, using either an augmented state vector approach (Chen and Zhang, 2006), dual state parameter estimation (Moradkhani et al., 2005b) or parameter estimation in a loop external to the data assimilation filter (Vrugt et al., 2005). Successful applications are reported for such diverse areas as groundwater hydrology (Franssen and Kinzelbach, 2008, Kurtz et al., 2014, Schöniger et al., 2012), rainfall–runoff models (Moradkhani et al., 2005a, Vrugt et al., 2006), land surface models (Han et al., 2014a, Pauwels et al., 2009), vadose zone hydrology (Montzka et al., 2011, Wu and Margulis, 2013) and atmospheric models (Ruiz et al., 2013). A data assimilation framework can consider uncertain model forcing, model structure and initial conditions, as well as parameter uncertainties. Data assimilation has become a commonly used method for parameter estimation, especially for large scale applications (Wanders et al., 2014).
Joint soil moisture and soil moisture bias estimation has been proven to be helpful for improving data assimilation results (De Lannoy et al., 2007a, Kumar et al., 2012b) like soil temperature assimilation with bias correction (Bosilovich et al., 2007, Reichle et al., 2010). In this study, we also evaluated the impact of the soil moisture bias estimation method (Dee, 2005) on improving the soil moisture assimilation and irrigation scheduling and compared it with joint state-parameter estimation.
In Han et al. (2015a), we studied the joint updating of soil moisture, soil temperature and leaf area index by assimilating CRP neutron intensity and land surface temperature. In this study however, we considered in addition the joint updating of soil moisture and soil properties, or soil moisture and soil moisture bias, and the vertical and horizontal weighting for updating soil moisture in the footprint of a CRP. This implies that in this work states and parameters for many model grid cells in the CRP footprint are updated with a single CRP neutron intensity observation. This is therefore a small multiscale data assimilation experiment with the irrigation scheduling as one of the objectives.
It is expected that a more accurate characterization of the heterogeneous soil moisture distribution can be obtained if the coarse scale CRP neutron intensity data are assimilated using a combination of data assimilation and parameter estimation (or bias estimation). Based on such results, it is then assumed that the estimated irrigation requirement could be improved. The objective of this study is to evaluate with help of a synthetic study: (1) the potential of measured neutron intensity data by the CRP for improving the characterization of soil moisture content and soil properties (or soil moisture bias), and (2) the impact of the assimilation of neutron intensity on better irrigation scheduling and the potential for real-time irrigation optimization. In this study, the spatial variability of soil properties and crop status will be considered in the data assimilation.
Section snippets
Methodology
The main components of the methodology are: (i) measurement of above-ground neutron intensity, which is linked to field scale soil moisture content by a measurement operator (Section 2.1) and horizontal weights (Section 2.3); (ii) the land surface model CLM (version 4.5) which simulates the transport of water and energy in the soil–plant–atmosphere continuum (Section 2.2); (iii) data assimilation according to the Local Ensemble Transform Kalman Filter (LETKF) methodology (Hunt et al., 2007)
Synthetic experiment
A synthetic study was conducted to evaluate the methodology outlined in the previous sections. The synthetic study mimicked the Picassent site (close to Valencia, Spain) with citrus trees, which receives drip irrigation. The site is situated in a semi-arid region (39.38°N, 0.47°E) with yearly average precipitation of 454 mm (44 precipitation days), average daily maximum temperature of 22.3 °C and average daily minimum temperature of 13.4 °C, and with a yearly irrigation period from April to
Results
In this section we evaluate time series for the different simulation scenarios at the CRP location. Spatial patterns of (estimated) soil properties and optimized irrigation amounts for different simulation scenarios are also compared. This comparison is made at the scale of the complete CRP footprint. The temporal evolution of soil moisture content at 30 cm and 50 cm depth for different simulation scenarios is shown in Fig. 2. The scenario No_DA underestimated soil moisture content even although
Discussion
The proposed data assimilation and parameter estimation (or bias estimation) can improve the soil moisture and irrigation estimation. The joint state-parameter estimation is the best scenario, and reduced the RMSE values of soil moisture content more than 50%, the spatial similarity of irrigation amount was increased and the HD values were decreased by 86% on average. The novelty of this work was the assimilation of the new CRP data in combination with irrigation scheduling. In general,
Conclusions
This study investigated the assimilation of synthetic measurements of coarse scale CRP neutron intensity in CLM for updating field scale root zone soil moisture content. The synthetic study mimicked a drip irrigated citrus farmland near Valencia, Spain. CLM was driven by biased soil properties and the joint estimation of soil moisture and soil properties (or soil moisture bias) was evaluated in a data assimilation framework using the state augmentation method. The non-linear measurement
Acknowledgments
This work was supported by AGADAPT (adapting water use by the agriculture sector) financed by Climate Knowledge and Innovation Community (Climate-KIC) of the European Union. AGADAPT focuses on the development and deployment of novel methods to reduce and optimize the water usage of rain-fed and irrigated agriculture by combining knowledge-based innovative technologies, modelling and transfer of technologies and innovative practices. The work was also supported by Transregional Collaborative
References (80)
- et al.
Response of Clementina de Nules citrus trees to summer deficit irrigation. Yield components and fruit composition
Agric. Water Manage.
(2011) A comparison study of two different control criteria for the real-time management of urban groundwater works
J. Environ. Manage.
(2012)- et al.
Data assimilation for transient flow in geologic formations via ensemble Kalman filter
Adv. Water Resour.
(2006) - et al.
Monitoring root-zone soil moisture through the assimilation of a thermal remote sensing-based soil moisture proxy into a water balance model
Remote Sens. Environ.
(2008) - et al.
Experiments of one-dimensional soil moisture assimilation system based on ensemble Kalman filter
Remote Sens. Environ.
(2008) - et al.
Efficient data assimilation for spatiotemporal chaos: a local ensemble transform Kalman filter
Physica D
(2007) - et al.
Estimation of unsaturated soil hydraulic parameters using the ensemble Kalman filter
Vadose Zone J.
(2011) Hydraulic parameter estimation by remotely-sensed top soil moisture observations with the particle filter
J. Hydrol.
(2011)- et al.
Dual state–parameter estimation of hydrological models using ensemble Kalman filter
Adv. Water Resour.
(2005) - et al.
Vegetation and surface roughness effects on AMSR-E land observations
Remote Sens. Environ.
(2006)