Potential of water balance and remote sensing-based evapotranspiration models to predict yields of spring barley and winter wheat in the Czech Republic
Introduction
Agricultural drought can significantly impact crop yields, as demonstrated both recently (e.g. Blauhut et al., 2016; Jakubínský et al., 2019) and in the more distant past, e.g. in the pre-instrumental period (e.g. Brázdil et al., 2019), and can have global implications for food production and security (Trnka et al., 2019). The impacts of agricultural drought on crops can be most directly identified by the reduction of yield (Fiala et al., 2014); however, not only the quantity but also the quality of production can be affected by drought occurrence (Jensen et al., 1996). Understanding of the relationship between water status indicators and crop yields is crucial for assessing the drought impacts on agricultural crops (Anderson et al., 2016a, Yang et al., 2018). While the final impacts of drought on agricultural crop yields can be fully assessed only at harvest (yield observation), early estimates of yield (yield forecasting) are very useful for farmers, traders and policy makers (Zhang and Kovacs, 2012). Yield forecasting is typically realized using a set of real-time predictor time series, with forecast accuracy gradually increasing as the growing season progresses toward the peak performance in the interval prior to harvest (Bognár et al., 2017). In this way, the final crop yield can be to some degree predicted several weeks up to months before harvest, providing useful support for planning of agricultural operations (Hlavinka et al., 2019, Nagy et al., 2018).
One of the most widely used approaches for monitoring crop conditions is based on remote sensing (Tucker et al., 1980, Johnson, 2016; Wardlow et al., 2012), which utilizes the connection between the spectral reflectances of certain plant species and their yield (e.g. Tucker et al., 1980). This approach has been used since 1972 when Landsat 1 was launched (Rembold et al., 2013), representing the first civil satellite orbiting the Earth. In addition to satellites, data from airborne imaging and unmanned aerial vehicle surveys can be used for monitoring of vegetation (Zhang and Kovacs, 2012). Commonly used remote sensing indicators embody vegetation indices that have the capability to track crop progress and evolution in green biomass (Becker-Reshef et al., 2010), including the normalized difference vegetation index (NDVI), the enhanced vegetation index (EVI), or its two-band version (EVI2).
Another effective indicator of crop health is evapotranspiration (ET), which over agricultural lands represents water consumed by crops in the process of transpiration and water evaporated from the soil, leaves and other surfaces. During the growing cycle, anomalously low rates of actual ET relative to maximum expected ET may indicate depleted soil moisture and stress-related reductions in transpiration fluxes (Anderson et al., 2007a). Therefore, ET-based indicators have been used to monitor agricultural drought (Anderson et al., 2011, Anderson et al., 2016b, Potopová et al., 2015) and impacts on crop yields (Anderson et al., 2016a, Crocetti et al., 2020, Yang et al., 2018) including indices based on actual ET (ETa) and ETa normalized by reference ET (ETo). The relationship between ET and crop health and productivity has been studied for a long time and provides basis for various approaches to monitor crop water stress. These included e.g. the crop coefficient model (Allen et al., 1998) combined with the approach based on radiation interception and light use efficiency (Monteith, 1977) or a crop simulation model based on the concept of water use efficiency (Raes et al., 2009, Steduto et al., 2012).
For regional monitoring of ET, two modeling approaches are generally used (e.g. Yilmaz et al., 2014; Anderson et al., 2007a): prognostic and diagnostic models, each with its advantages and disadvantages. ET as a water balance component can be estimated by the use of prognostic land surface models (Yilmaz et al., 2014, Mendiguren et al., 2017, Arnold et al., 1998, Abbott et al., 1986). In this study, we considered the case of the SoilClim water balance model widely used in the Czech Republic (Hlavinka et al., 2011, Trnka et al., 2020). Other water balance models include the Variable Infiltration Capacity (VIC) model (Wood et al., 1992), the Soil and Water Assessment Tool (SWAT) (Arnold et al., 1998) and the FAO model AquaCrop (Raes et al., 2009, Steduto et al., 2012). Prognostic assessment of ET can be also performed using land surface models (LSMs) such as Noah LSMs (e.g. Ek et al., 2003). Prognostic models use basic weather variables (e.g. precipitation, wind, radiation, air temperature) together with information about soil (e.g. field capacity and wilting point) and vegetation (e.g. leaf area index – LAI, rooting depth, crop coefficient) to model the complete water and energy balance in the soil, plant and atmosphere interface. Predicted ET rates depend strongly on the accuracy and availability of input weather variables and parameters related to soil and vegetation, as well as on the accuracy of the model itself (Beljaars et al., 1996). Many of these inputs can be difficult to specify at regional or continental level, and biases may result in cumulative errors in soil water content and associated fluxes. These biases can even become larger in the long-term perspective if there are no mechanisms for periodic correction (Schaake et al., 2004).
In contrast, diagnostic models of ET require less information about soil and vegetation properties and generally do not require precipitation as input. To estimate ET, these models rely on remote sensing retrievals of land surface temperature (LST), albedo, vegetation cover fraction or LAI. Therefore, diagnostic models can provide the information about the plant water status whenever remote sensing data are available independently on the previous imaging. As estimation of ET is based on remote sensing data, these models are largely independent from ground measurements and ET-based indicators can be calculated over the large areas, from country to continent (e.g. Europe) to global scales. On the other hand, diagnostic models based on thermal and visible/near-infrared remote sensing data inherently have one significant limitation compared to prognostic ground-based models: input data can be obtained only during clear-sky conditions. Examples of diagnostic ET models include the Surface Energy Balance System (SEBS) (Su, 2002), the Surface Energy Balance Algorithm for Land (SEBAL) (Bastiaanssen et al., 1998) and the Two-Source Energy Balance model (TSEB) (Norman et al., 1995). The diagnostic model considered here is the Atmosphere–Land Exchange Inverse (ALEXI) model (Anderson et al., 1997, Anderson et al., 2007a), which is based on TSEB and uses time changes in LST derived from remote sensing to partition the surface energy budget. The evaporative stress index (ESI) is an ALEXI based product representing standardized anomalies in the ratio of actual to reference ET (ETa/ETo) (Anderson et al., 2007b). ESI has been demonstrated to be consistent with standard precipitation-based drought indices while additionaly capable to provide quick response to changes in vegetation health during rapid drought onset events (Otkin et al., 2016).
Recent efforts toward developing operational yield forecasting systems based on different predictors have been conducted within several regions in Central Europe. In Hungary, for example, several studies have investigated the relationship between remote sensing data and crop yields (Bognár et al., 2017, Kern et al., 2018, Nagy et al., 2018). In the Czech Republic, ESI was tested for spring barley and winter wheat yields (Anderson et al., 2016a), the two most widely cultivated cereal crops in the region. Based on the study of Anderson et al. (2016a) and several analyses focusing on vegetation indices (e.g. Johnson, 2016), a prototype remotely sensed forecasting method for key crops for the whole country was put into use in 2018. This system was initially based on vegetation indices (NDVI, EVI, EVI2) from the Moderate Resolution Imaging Spectroradiometer (MODIS) with 250 m spatial resolution. Calibration of yield forecasts for 2018 were based on yield data and vegetation indices from 2000 to 2017. This operational forecasting method has capability to predict the final yields more than two months prior to harvest.
In this study, we tested the predictive capability of ET-based indicators (ETa and ESI) toward incorporating them into the operational yield forecasting system. The ET-based indicators were tested for the period 2001–2018 over 33 districts in the Czech Republic and were generated with models representing two different approaches: prognostic ground-based (SoilClim) versus diagnostic remote sensing-based model (ALEXI). In the previous study of Anderson et al. (2016a), linear regressions between yields and ESI from ALEXI were tested at regional and district levels. In the current study, the method of the artificial neural networks (ANNs) at the district level was investigated. The main objective was to (i) to determine whether an indicator from a prognostic or diagnostic model is preferable to be included into the operational yield forecasting system. The other objective was (ii) to identify the optimal period during the growing season when the capability to predict yield is the highest. This was tested by the adjusting number and timeframe of ANN inputs during the growing season.
Section snippets
Study area
Only districts where yield data were available for all years of the period 2001–2018 were analyzed: yield data were available for 33 of total 76 districts of the Czech Republic (Table 1 and Fig. 1). Districts of the Czech Republic are identical with the local administrative units of level 1 used in states of the European Union (LAU 1; https://www.czso.cz/csu/czso/kraje-nuts-3-a-okresy-lau-1-ceske-republiky). The district area differs within the country with the lowest area around 200 km2 and
Identification of the optimal period for yield prediction
For all ET-based indicators, the RMSE values decreased and R2 values increased with the higher number of ANN inputs and with DOY within the growing season as more data were included in the yield prediction. For each indicator, there could be identified the optimal period for yield prediction. This period started when the lowest RMSE and highest R2 values were approached and these values did not significantly improve as the growing period proceeded. This could be identifiable already at the
Comparison with previous studies
The relationship between ESI from ALEXI (in particular ESI-12WKALEXI) and winter wheat and spring barley yields in the Czech Republic was previously examined by Anderson et al. (2016a). These two crops were analyzed at the district level (LAU 1), as well as at the regional level (NUTS 3). Anderson et al. (2016a) calculated yield anomalies at the district level from a linear regression over the period 2002–2014, then computed regression statistics between ESI-12WKALEXI and observed yields lumped
Summary and conclusions
This study focused on the capability of ET-based indicators to predict yield of spring barley and winter wheat prior to harvest and identification of the optimal period for the yield prediction. The optimal period was defined as the date when the lowest RMSE and highest R2 values were reached and these did not significantly improve as the growing season proceeded. The optimal period for both key crops of the Czech Republic started between DOYs 155 and 169 (roughly early-June to mid-June). The
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
The study was conducted with support of the SustES – Adaptation Strategies for Sustainable Ecosystem Services and Food Security Under Adverse Environmental Conditions (CZ.02.1.01/0.0/0.0/16_019/0000797). F.J. acknowledges support from the project IGA AF MENDELU no. IP 9/2017 with the support of the Specific University Research Grant, provided by the Ministry of Education, Youth and Sports of the Czech Republic in 2017.
References (71)
- et al.
An introduction to the European Hydrological System — Systeme Hydrologique Europeen, “SHE”, 1: History and philosophy of a physically-based, distributed modelling system
J. Hydrol.
(1986) - et al.
New globally convergent training scheme based on the resilient propagation algorithm
Neurocomputing
(2005) - et al.
A two-source time-integrated model for estimating surface fluxes using thermal infrared remote sensing
Remote Sens. Environ.
(1997) - et al.
The Evaporative Stress Index as an indicator of agricultural drought in Brazil: an assessment based on crop yield impacts
Remote Sens. Environ.
(2016) - et al.
A remote sensing surface energy balance algorithm for land (SEBAL). 1. Formulation
J. Hydrol.
(1998) - et al.
A generalized regression-based model for forecasting winter wheat yields in Kansas and Ukraine using MODIS data
Remote Sens. Environ.
(2010) - et al.
Mapping daily evapotranspiration at field scales over rainfed and irrigated agricultural areas using remote sensing data fusion
Agric. Meteorol.
(2014) - et al.
A within-season approach for detecting early growth stages in corn and soybean using high temporal and spatial resolution imagery
Remote Sens. Environ.
(2020) - et al.
Effect of drought on yield variability of key crops in Czech Republic
Agric. Meteorol.
(2009) - et al.
Development and evaluation of the SoilClim model for water balance and soil climate estimates
Agric. Water Manag.
(2011)
Seed glucosinolate, oil and protein contents of field-grown rape (Brassica napus L.) affected by soil drying and evaporative demand
F. Crop. Res.
A comprehensive assessment of the correlations between field crop yields and commonly used MODIS products
Int. J. Appl. Earth Obs. Geoinf.
Statistical modelling of crop yield in Central Europe using climate data and remote sensing vegetation indices
Agric. Meteorol.
Global products of vegetation leaf area and fraction absorbed PAR from year one of MODIS data
Remote Sens. Environ.
Wheat and maize yield forecasting for the Tisza river catchment using MODIS NDVI time series and reported crop statistics
Comput. Electron. Agric.
Source approach for estimating soil and vegetation energy fluxes in observations of directional radiometric surface temperature
Agric. Meteorol.
Impacts and adaptation of European crop production systems to climate change
Eur. J. Agron.
Assessing the evolution of soil moisture and vegetation conditions during the 2012 United States flash drought
Agric. Meteorol.
Performance of the standardised precipitation evapotranspiration index at various lags for agricultural drought risk assessment in the Czech Republic
Agric. Meteorol.
New refinements and validation of the Collection-6 MODIS land-surface temperature/emissivity products
Remote Sens. Environ.
Remote Sensing of Environment Field-scale mapping of evaporative stress indicators of crop yield: an application over Mead, NE, USA
Remote Sens. Environ.
Impacts of water availability and drought on maize yield – a comparison of 16 indicators
Agric. Water Manag.
Relationships between the evaporative stress index and winter wheat and spring barley yield anomalies in the Czech Republic
Clim. Res.
Evaluation of drought indices based on Thermal remote sensing of evapotranspiration over the continental United States
J. Clim.
A climatological study of evapotranspiration and moisture stress across the continental United States based on thermal remote sensing: 1. Model formulation
J. Geophys. Res.
A climatological study of evapotranspiration and moisture stress across the continental United States based on thermal remote sensing: 2. Surface moisture climatology
J. Geophys. Res.
Large area hydrologic modeling and assessment. Part I: Model development
J. Am. Water Resour. Assoc.
Situační a výhledová zpráva: Půda (in Czech)
The anomalous rainfall over the United States during July 1993: sensitivity to land surface parameterization and soil moisture anomalies
Mon. Weather Rev.
Estimating drought risk across Europe from reported drought impacts, drought indices, and vulnerability factors
Hydrol. Earth Syst. Sci.
Yield estimation and forecasting for winter wheat in Hungary using time series of MODIS data
Int. J. Remote Sens.
The extreme drought of 1842 in Europe as described by both documentary data and instrumental measurements
Clim
Earth Observation for agricultural drought monitoring in the Pannonian Basin (southeastern Europe): current state and future directions
Reg. Environ. Change
Implementation of Noah land surface model advances in the National Centers for Environmental Prediction operational mesoscale ETa model
J. Geophys. Res.
Cited by (10)
A brief history of the thermal IR-based Two-Source Energy Balance (TSEB) model – diagnosing evapotranspiration from plant to global scales
2024, Agricultural and Forest MeteorologyCrop yield anomaly forecasting in the Pannonian basin using gradient boosting and its performance in years of severe drought
2023, Agricultural and Forest MeteorologyAttributing the drivers of runoff decline in the Thaya river basin
2023, Journal of Hydrology: Regional StudiesApplicability of machine learning techniques in predicting wheat yield based on remote sensing and climate data in Pakistan, South Asia
2023, European Journal of AgronomyCopulas modelling of maize yield losses – drought compound events using the multiple remote sensing indices over the Danube River Basin
2023, Agricultural Water Management