Elsevier

Journal of Hydrology

Volume 553, October 2017, Pages 88-104
Journal of Hydrology

Research papers
Performance of SMAP, AMSR-E and LAI for weekly agricultural drought forecasting over continental United States

https://doi.org/10.1016/j.jhydrol.2017.07.049Get rights and content

Highlights

  • Remotely sensed products (i.e., LAI, AMSR-E, SMAP) can improve drought forecast.

  • Coupled SVM-DA model can improve drought forecast at longer lead time.

  • Performance of drought forecast varies at local scales.

Abstract

This study applied support vector machines (SVMs) and data assimilation (DA) methods to investigate the performance of in-situ and remotely sensed products (i.e., leaf area index (LAI), AMSR-E and SMAP soil moisture retrievals) for near-real time agricultural drought forecasting for in-situ stations located in continental United States (CONUS). The agricultural drought was quantified using soil water deficit index (SWDI) derived based on available soil moisture and basic soil water parameters. It is observed that SVMs or SVM-DA with limited meteorological variables as inputs able to forecast SWDI at most of the in-situ stations up to 1–2-week lead time. Addition of remotely sensed products (i.e., LAI, AMSR-E, SMAP) either individually or simultaneously as inputs to SVMs can able to improve SWDI forecast at most of the stations where the strong relationship exists between LAI (and/or AMSR-E, SMAP) with SWDI. Such improvement can persist up to 2–4-week lead time at some of the stations. But the efficiency tends to decrease with the increase in lead time. The addition of both LAI and SMAP (AMSR-E) performs better than the independent addition of LAI or SMAP (AMSR-E). Typically, the performance of drought prediction varies at local scales; therefore it is difficult to generalize our findings at regional scale.

Introduction

Soil moisture plays an important role in the global water and energy cycle. The anomalies of soil moisture may exert great impact on the subsequent climate variables, thus, leading to the evolution of climate extremes (e.g., flood, drought and heat wave) by linking hydroclimatic fluxes at different spatio-temporal scales (e.g., Jaeger and Seneviratne, 2011, Liu et al., 2014a, Liu et al., 2014b). Among the three commonly recognized categories of drought (i.e., meteorological, hydrological and agricultural) (Mishra and Singh, 2010), agricultural drought has a more direct and immediate impact on food security. Generally, agricultural drought is considered to begin when the soil moisture availability to plants drops to a level, which would adversely affects the crop yield and, hence, agricultural production (Panu and Sharma, 2002, Mishra et al., 2015). Therefore, it is reasonable to consider soil moisture databases as one potential resources for agricultural drought monitoring.

Currently, the soil moisture information at a coarser resolution is available for different sectorial applications with the development of remote sensing techniques, such as, the Advanced Microwave Scanning Radiometer-Earth Observing System sensor (AMSR-E) (Njoku et al., 2003), the Advanced Scatterometer (ASCAT) soil moisture retrievals (Wagner et al., 1999), the Soil Moisture and Ocean Salinity (SMOS) Soil Moisture Level 2 User Data Product (SMUDP2 file) (Kerr et al., 2010) and the newly launched Soil Moisture Active Passive (SMAP) mission (Entekhabi et al., 2010). These remotely sensed SM retrievals can provide an unprecedented spatial and temporal resolution of SM data across a range of scales, but is limited in terms of sensing at different depth. There is an ongoing effort to retrieve (develop) root zone SM profile using the surface SM values (Li et al., 2012, Tran et al., 2013, Ridler et al., 2014).

The growing of multiple SM databases activate the proposition of SM related drought indices in the agricultural drought monitoring (e.g., Hunt et al., 2009, Martínez-Fernández et al., 2015, Martínez-Fernández et al., 2016, Mishra et al., 2015, Keshavarz et al., 2014, Scaini et al., 2015, Sridhar et al., 2008). For example, Sridhar et al. (2008) and Hunt et al. (2009) proposed and revised the soil moisture index (SMI) to characterize the water deficit in the soil for drought. Mishra et al. (2015) applied the Standardized Soil Moisture Index (SSMI) at different soil layers and the Standardized Soil Water Availability Index (SSWI) to diagnose the agricultural drought severity based on crop modeling. Scaini et al. (2015) used SMOS-derived SM anomalies for drought monitoring assessment. Martínez-Fernández et al. (2015) defined the soil water deficit index (SWDI) using soil moisture series solely for agricultural drought monitoring based on the percentile method proposed by Hunt et al. (2009). However, unlike the SMI defined in (Hunt et al., 2009, Sridhar et al., 2008), the deficit threshold for SWDI is fixed by the field capacity instead of the wilting point. Martínez-Fernández et al. (2016) further assessed the SWDI for agricultural drought monitoring with SMOS soil moisture retrievals.

However, some previous literatures stated that the water supplied to crops depends on the available SM in the root zone. Therefore, agricultural drought should not be treated equal for all crops, instead, it should be calculated based on the root zone SM rather than at a fixed soil layer depth (Mishra et al., 2015). Among the proposed SM related drought index, the SWDI proposed by Martínez-Fernández et al. (2015) directly include the available SM at the root zone for the crop growth. SWDI considered to be an effective agricultural drought index as it is based on soil moisture and basic soil water parameters, including field capacity and wilting point (Martínez-Fernández et al., 2015, Martínez-Fernández et al., 2016). Further, the SWDI can provide accurate status of agricultural drought because it is related to the soil moisture suction capacity of different type of crops. SWDI is a simple and useful way for the drought monitoring over large area with the increasing availability of soil moisture databases worldwide. The root zone SM predictions are important for agricultural drought monitoring and forecasting and to improve water resources management. But it is still a challenge to accurately predict the SM that varies in depth, space and time.

According to previous studies, the data-driven techniques, such as support vector machines (SVMs), are able to estimate or predict SM at root zones with antecedent meteorological variables (i.e., surface air temperature, precipitation, solar radiation, relative humidity) (e.g., Gill and McKee, 2007, Kaheil et al., 2008, Liu et al., 2010, Yu et al., 2012, Liu et al., 2016). In addition to that, data assimilation (DA) is proved to be a promising technique to optimally estimate land surface’s variables (i.e., SM) by merging observed information into models (e.g., Kumar et al., 2008, Das et al., 2011, Li et al., 2012, Tran et al., 2013, Han et al., 2014, Kornelsen and Coulibaly, 2014, Yin et al., 2015, Montzka et al., 2011). Ensemble Kalman Filter (EnKF) is a kind of Monte Carlo approximation of sequential Bayesian filtering process, which alternates between an ensemble forecast step and a state variable update step (Reichle et al., 2002). EnKF is a popular DA technique used in hydrology (Gill and McKee, 2007, Liu et al., 2010, Yin et al., 2014, Mishra et al., 2015, Yin et al., 2015). The dual EnKF technique requires two separate state-space representation for the state and parameters through two interactive filters by updating model parameters and model states, which makes it more efficient in the improvement of estimations (i.e., Liu et al., 2016, Moradkhani et al., 2005).

This study investigates the potential of individual SVMs or coupled with dual EnKF DA technique (SVM-DA) to predict agricultural drought using in-situ meteorological variables and the remotely sensed products (i.e., soil moisture, leaf area index). Although the remote sensing products are applied for drought forecasting, however, the application of SMAP is limited. This study can provide insight to the potential improvements that can be achieved in agricultural drought forecasting by addition of a variety of remote sensing products including SMAP. The main objectives are: (1) To investigate the efficiency of meteorological variables as input of SVM for the agricultural drought forecast at subsequent weekly time scales; (2) To assess whether the addition of remotely sensed products as forcing variables can improve the performance of drought forecasting; and (3) To evaluate the performance of coupled SVM-DA technique to forecast agricultural drought in multiple in-situ stations located in CONUS.

The remaining structure of the paper is organized as follows: data description is given in Section 2; the model setup, experimental design and evaluation methodologies are introduced in Section 3; the results are analyzed in Section 4 and Section 5 concludes the main findings.

Section snippets

In-situ and remotely sensed data sets

The in-situ observed hydrometeorological variables, including air temperature (T2m), solar radiation (SR), precipitation (P), relative humidity (RH), and soil moisture (SM) at different soil depth from 5 cm to 100 cm, obtained from the United States Climate Reference Network (USCRN) stations with automated measured instrument are adopted in this study. The hourly, daily and monthly data are available at USCRN website (https://www.ncdc.noaa.gov/crn/qcdatasets.html). The daily time series during

Experimental design

The SVMs and dual EnKF technique used in (Liu et al., 2016) is adopted for the near real time SWDI forecast. Consistently, for SVMs, it has training and predicting processes, while the SVM-DA approach include one more step (i.e., updating process). For example, consider the SVM-DA framework (Fig. S2), which suggests if the observed data are available at the predicted time, then the initial predictions (Yp) and observed data are used to update the model state and provide forcing data with the

Results analysis

Due to the non-overlapping of AMSR-E and SMAP data, we selected two time periods for analysis: (a) the 1st period for AMSR-E (October 1st 2009 to September 30th 2011); and (b) the 2nd period for SMAP (April 1st 2015 to March 31st 2016). The results are discussed in two sections: Firstly, we evaluated the performance of AMSR-E, SMAP and LAI for the drought forecasting at two different time periods at the 14 selected stations; Secondly, generalize our findings at the selected 59 stations for

Conclusions

This paper investigates the performance of limited in-situ meteorological variables (i.e., T2m, P, SR and RH) and remotely sensed products (including LAI, AMSR-E and SMAP) for the near-real time agricultural drought prediction using data-driven technique SVMs, solely or coupled with DA technique (SVM-DA). Due to limited time length of meteorological data and remotely sensed products, two time periods are selected in this study: 1st study time period during October 1st 2009 to September 30th

Conflict of interest

Authors do not have COI for their manuscript.

Acknowledgement

This study was supported by the United States Department of Agriculture (USDA) award 2015-68007-23210, the Fundamental Research Funds for the Central Universities of China (Grant No. 2015B00214), the National Natural Science Foundation of China (Grant Nos. 41471016; 41323001; 51539003), the National Key Research and Development Program of China (Grant Nos. 2016YFC0402706; 2016YFC0402710), the Special Fund of State Key Laboratory of Hydrology-Water Resources (Grant No. 20145027312) and the open

References (57)

  • A.K. Mishra et al.

    Anatomy of a local-scale drought: application of assimilated remote sensing products, crop model, and statistical methods to an agricultural drought study

    J. Hydrol.

    (2015)
  • A.K. Mishra et al.

    A review of drought concepts

    J. Hydrol.

    (2010)
  • C. Montzka et al.

    Hydraulic parameter estimation by remotely-sensed top soil moisture observations with the particle filter

    J. Hydrol.

    (2011)
  • H. Moradkhani et al.

    Dual state–parameter estimation of hydrological models using ensemble Kalman filter

    Adv. Water Resour.

    (2005)
  • J. Samuel et al.

    Assessing model state and forecasts variation in hydrologic data assimilation

    J. Hydrol.

    (2014)
  • W. Wagner et al.

    A method for estimating soil moisture from ERS scatterometer and soil data

    Remote Sens. Environ.

    (1999)
  • H. Yan et al.

    A probabilistic drought forecasting framework: a combined dynamical and statistical approach

    J. Hydro.

    (2017)
  • Z. Yu et al.

    A multi-layer soil moisture data assimilation using support vector machines and ensemble particle filter

    J. Hydrol.

    (2012)
  • Allen, R.G., Pereira, L.S., Raes, D., et al., 1998. Crop Evapotranspiration: Guidelines for Computing Crop Water...
  • M.E. Brown et al.

    NASA's soil moisture active passive (SMAP) mission and opportunities for applications users

    Bull. Amer. Meteor. Soc.

    (2013)
  • K.S. Cheng et al.

    On the criteria of model performance evaluation for real-time flood forecasting

    Stoch. Environ. Res. Risk Assess.

    (2016)
  • N. Das et al.

    Algorithm for merging SMAP radiometer and radar data for high resolution soil moisture retrieval

    IEEE Trans. Geosci. Remote Sens.

    (2011)
  • P.A. Dirmeyer et al.

    Precipitation, recycling, and land memory: An integrated analysis

    J. Hydrometeorol.

    (2009)
  • M. Drusch et al.

    Observation operators for the direct assimilation of TRMM microwave imager retrieves soil moisture

    Geophys. Res. Lett.

    (2005)
  • Entekhabi, D., et al., 2010. The Soil Moisture Active Passive (SMAP) mission. Proc. IEEE 98, pp....
  • M.K. Gill et al.

    Soil moisture data assimilation using support vector machines and ensemble Kalman filter

    J. Am. Water Resour. Assoc.

    (2007)
  • X. Han et al.

    Soil moisture and soil properties estimation in the Community Land Model with synthetic brightness temperature observations

    Water Resour. Res.

    (2014)
  • E.D. Hunt et al.

    The development and evaluation of a soil moisture index

    Int. J. Climatol.

    (2009)
  • Cited by (0)

    View full text