Data assimilation of LAI improved crop growth modeling: Comparison between in-situ measurements and satellite estimations - Abstract

Data assimilation of LAI improved crop growth modeling:Comparison between in-situ measurements and satellite estimations - AbstractMarija Kopanja 1, Gordan Mimić 1, Hilde Vaessen 2,, Nevena Stevanović 1, Nikola Stanković 1, Maša Buđen 1, Dragana Blagojević 1, Bernardo Maestrini 2, Nataša Ljubičić 1, Sanja Brdar 1, Frits van Evert 2 1 BioSense Institute, Zorana Đinđića 1, Novi Sad, 21000, Serbia 2 Wageningen University & Research, Droevendaalsesteeg 4, Wageningen, 6708 PB, Netherlands Summary A digital twin of a crop can be created by the integration of real-time measurements into existing crop growth models. In this study, we used data collected at field level during the 2023 growing season and the WOFOST crop simulation model. The field experiment was conducted in Čenej (Serbia), where maize and soybean were cultivated. The field was subdivided into 20 management zones, but the study focused on 12 zones characterized by the highest variability. Six zones were allocated to each crop based on multi-year average NDVI values. Daily weather data on maximum and minimum air temperature, precipitation amount, wind speed and relative humidity were collected from a meteorological station placed next to the field, and additionally surface solar radiation was downloaded from ERA-5 reanalysis available at Climate DataStore. When the intensive crop growth started, data on the in-situ measured leaf area index (LAI) were collected biweekly, while the satellite LAI estimations were calculated from the cloud-free Sentinel-2 images by using Sentinel Application Platform (SNAP). The yield data were obtained at the end of the growing season from a GPS-equipped yield monitor mounted on the combine harvester and the values were aggregated at the zone level. The study focused on the LAI and investigated the performance of the crop growth model after data assimilation of LAI, obtained in two different ways. Data assimilation was performed using the ensemble Kalman filter, with 50 ensemble members, varying the total initial dry weight (TDWI) as initial condition, and life span of leaves (SPAN) as a parameter, both related to the leaves. Adjustment of each model state was based on standard deviation of the observed LAI and simulated LAI, as a measure of the uncertainty of the observation and model state. The WOFOST model wasrun for each zone and the results for yield prediction were compared in three different cases: simulation without data assimilation, data assimilation of in-situ measured LAI (M), data assimilation of LAI estimated from satellite images (E). Performance of the WOFOST model was evaluated for the final yield, comparing it to the observed yield, by using standard error metrics such as mean absolute error (MAE) and root mean squared error (RMSE). For maize and soybean, the WOFOST model without data assimilation has an average MAE across all six zones 1198,50 kg/ha and 240,71 kg/ha, and an average RMSE 1492,57 kg/ha and 275,76 kg/ha, respectively. In the second case, the performance of WOFOST simulation with data assimilation using in-situ measured LAI was improved for both crops. For maize, a decrease is observed in both average MAE and RMSE across all six zones (MAEM=947,27 kg/ha and RMSEM=1011,03 kg/ha). Similarly for soybean, MAEM=217,43 kg/ha and RMSEM=231,92kg/ha. In the third scenario where the data assimilation is performed using LAI estimated from Book of Abstracts of HAICTA 2024, October 17–20, 2024, Karlovasi, Samos, Greece EMAIL: [email protected] (A. 1); [email protected] (A. 2); [email protected] (A. 3); [email protected] (A. 4); [email protected] (A. 5); [email protected] (A. 6); [email protected] (A. 7); [email protected] (A. 8); [email protected] (A. 9); [email protected] (A. 10); [email protected] (A. 11) ORCID: 0000-0003-1176-5677 (A. 1); 0000-0001-6879-8969 (A. 2); 0009-0009-7253-1322 (A. 3); 0000-0002-1613-7509 (A. 4); 0000-0002-8601-8070 (A. 5); 0000-0003-3825-5888 (A. 6); 0000-0002-6812-9113 (A. 7); 0000-0002-9438-0678 (A. 8); 0000-0001-5982-9401 (A. 9);0000-0002-2259-4693 (A. 10); 0000-0003-0302-668X (A. 11) © 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).60 satellite images, for maize the model overperformed the two other cases, where MAEE=543,69 kg/ha and RMSEE=632,59 kg/ha, while for soybean the model performed better compared to the model without data assimilation and worse compared to data assimilation with in-situ measured LAI, where MAEE=227,61 kg/ha and RMSEE=256,27 kg/ha. The results show that data assimilation can lead to an improvement in crop growth model performance in terms ofpredicted yield. KeywordsDigital twin, Data assimilation, WOFOST, Ensemble Kalman Filter, LAI