Potential of water balance and remote sensing-based evapotranspiration models to predict yields of spring barley and winter wheat in the Czech Republic

Authors: JURECKA, FRANTISEK - Mendel University; FISCHER, M. - Czech Globe - Global Change Research Institute; HAVINKA, P. - Mendel University; BALEK, J. - Czech Globe - Global Change Research Institute; SAMERADOVA, D. - Czech Globe - Global Change Research Institute; BLAHOVA, M. - Czech Globe - Global Change Research Institute; Anderson, Martha; HAIN, C. - Nasa Marshall Space Flight Center; ZALUD, Z. - Czech Globe - Global Change Research Institute; TRNKA, M. - Mendel University

Submitted to: Agricultural Water Management
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 7/2/2021
Publication Date: 7/15/2021
Citation: Jurecka, F., Fischer, M., Havinka, P., Balek, J., Sameradova, D., Blahova, M., Anderson, M.C., Hain, C., Zalud, Z., Trnka, M. 2021. Potential of water balance and remote sensing-based evapotranspiration models to predict yields of spring barley and winter wheat in the Czech Republic. Agricultural Water Management. 256:107064. https://doi.org/10.1016/j.agwat.2021.107064.
DOI: https://doi.org/10.1016/j.agwat.2021.107064

Interpretive Summary: Since crop water use (evapotranspiration or ET) is a good indicator of crop health and water availability, spatial maps of ET through the growing season can provide valuable information for predicting yields over large areas. Maps of ET can be developed using satellite imagery, or they can be computed using a gridded soil water balance model. Each approach has strengths and weaknesses, and it is not known which approach is more reliable for operational yield mapping activities. This study compares the relative utility of satellite-based and water-balanced ET maps for estimating yields of spring barley and winter wheat over the Czech Republic using an artificial neural network (ANN) approach. In comparison with observed yields collected from 2001-2018 over 33 districts, both datasets performed well, providing R2 values of up to 0.8 and errors between 0.4 and 0.7 t ha-1, with reliable estimates available several weeks prior to harvest. These results will be used to inform and improve operational yield forecasting systems for Central Europe.

Technical Abstract: Yield indicators based on evapotranspiration (ET) provide useful information about surface water status, response of vegetation to drought stress, and potential growth limitations. The capability of ET-based indicators including actual ET and the evaporative stress index (ESI) to predict crop yields of spring barley and winter wheat was analyzed for 33 districts of the Czech Republic. In this study, the ET-based indicators were computed using two different approaches: (i) using a prognostic model, SoilClim, which computes the water balance based on ground observations and information about soil and land cover; (ii) using the diagnostic Atmosphere–Land Exchange Inverse (ALEXI) model based primarily on remotely sensed land surface temperature data. The capability of both sets of indicators to predict yields of winter wheat and spring barley was tested using artificial neural networks (ANNs) applied to the datasets using time composite windows moving over the growing season. Yield predictions based on ANNs were computed for both crops for all districts together, as well as for individual districts. The root-mean-square error (RMSE) and coefficient of determination (R2) between observed and predicted yields varied with date within the growing season and with the number of composites used for yield prediction. The period with the highest predictive capability started from early-June to mid-June. The RMSE values varied between 0.4 and 0.7 t ha-1 while R2 reached values of 0.5–0.8 during this period. For all indicators, the timing of RMSE values between observed and predicted yield was similar for both spring barley and winter wheat, starting between two and four composites. Results of the study demonstrated that ET-based indicators can be used for yield prediction prior to harvest and be a useful tool for assessment of drought and its impact on agricultural crops. At the same time, these predictors can be used in real time during the growing season and therefore have great potential for decision making at regional and district levels.