EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Physically Constrained Estimation of Dynamic Crop Coefficients Using Multisensor Satellite and Field Telemetry Data

Umer Tanveer, Hashim Ali, Jamal Ahmed, Abdu Salam, Farhan Amin, Isabel de la Torre, Lázaro Javier Hernández Rodríguez, Pablo Herrero García

Remote Sensing · 14.11.2026 · 18 (22) · 3917

DOI: 10.3390/rs18223917 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Remote Sensing, online erschienen am 14.11.2026. Verfasst von Umer Tanveer und 7 weiteren Autor:innen. Die Arbeit nennt 43 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Fundación Universidad Europea del Atlántico.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate estimation of crop evapotranspiration (ETc) plays an important role in sustainable agricultural water management, precision irrigation planning, and food security under a dynamically changing climate. Physics-based models such as the FAO-56 Penman–Monteith equation have a sound theoretical background but are limited in modeling real-time dynamics in vegetation, owing to their use of static crop coefficient (Kc) tables. Meanwhile, ML models trained between the satellite-derived vegetation indices and crop coefficients are prone to physical bound violations and poor spatial generalizability. In this study, the Physics-Informed Machine Learning (PIML) framework is introduced, which consists of connecting the data from Sentinel-2 multispectral reflectance with the data from MODIS gap-filled time-series and the thermal-based reference evapotranspiration by NASA ECOSTRESS within a hard-bounded residual structure, Kc = clip(Kc, prior + clip(fθ(x), −0.15, 0.15), 0.15, 1.20), where the neural network’s correction term resides between FAO-56 agronomic bounds. Although the held-out validation set (N = 200) was intentionally kept small due to the limited number of cloud-free ECOSTRESS overpasses, an operational telemetry system recorded hourly data from four crop fields at the Sustainable Agriculture Facility (RSRF) in California. Because the same four fields and 25 sectors are visited across both training and testing periods, the experimental evaluation represents an intra-site, short-term temporal forward-transfer test rather than a true spatial generalization. The limited sample size (five cloud-free dates) precluded true spatial generalization methods such as leave-one-field-out or field-blocked cross-validation, so we kept all four crops in the training set. Satellite data for the same field are highly spatially correlated; therefore, we used the ICC (intraclass correlation coefficient) to adjust for non-independence, reducing N = 200 to neff =12.0. Three models were benchmarked: an empirical baseline with an RMSE = 0.3801, a constrained PIML network (RMSE = 0.3263), and a global training-mean predictor (RMSE = 0.3257). Model B removed all bound violations in the physical domain (compared to 14% for Model A), reduced RMSE by 14.1% from the baseline, and produced results similar to the global mean; the design-effect-corrected p-value (0.5492) confirms the pilot comparison is statistically underpowered. When validating remote sensing applications with small, clustered samples, these results show that hard physical constraints eliminate physically implausible predictions during temporal forward projection, while highlighting the need to correct spatial autocorrelation and requiring dedicated field-blocked evaluations for future claims of spatial generalization.

Förderung

  • Fundación Universidad Europea del Atlántico

Zitieren

APA

Tanveer, U., Ali, H., Ahmed, J., Salam, A., Amin, F., Torre, I. D. L., Rodríguez, L. J. H., & García, P. H. (2026). Physically Constrained Estimation of Dynamic Crop Coefficients Using Multisensor Satellite and Field Telemetry Data. Remote Sensing, 18(22), 3917. https://doi.org/10.3390/rs18223917

Vancouver

Tanveer U, Ali H, Ahmed J, Salam A, Amin F, Torre IDL, et al. Physically Constrained Estimation of Dynamic Crop Coefficients Using Multisensor Satellite and Field Telemetry Data. Remote Sensing. 2026;18(22):3917. doi:10.3390/rs18223917

BibTeX
@article{Tanveer2026,
  author = {Tanveer, Umer and Ali, Hashim and Ahmed, Jamal and Salam, Abdu and Amin, Farhan and Torre, Isabel de la and Rodríguez, Lázaro Javier Hernández and García, Pablo Herrero},
  title = {Physically Constrained Estimation of Dynamic Crop Coefficients Using Multisensor Satellite and Field Telemetry Data},
  journal = {Remote Sensing},
  year = {2026},
  volume = {18},
  number = {22},
  pages = {3917},
  doi = {10.3390/rs18223917}
}
Kostenlos starten →