EVAonline: An open-source web platform for global reference evapotranspiration estimation via multi-source data fusion
Authors/Creators
-
Cunha Soares, Ângela S M
(Data manager)1, 2
-
Ribeiro, Vitor
(Project member)3
-
Duarte, Sergio
(Project member)4, 2
-
PERRELLA BALESTIERI, JOSE ANTONIO
(Related person)5
-
Padovani, Carlos Roberto
(Project member)6
- Bordignon, Álex Júnior Zanchet (Project member)
-
Maciel, Carlos
(Supervisor)3
-
Marques, Patricia
(Supervisor)1, 2
- 1. Universidade de São Paulo Escola Superior de Agricultura Luiz de Queiroz
-
2.
Universidade de São Paulo
-
3.
Universidade Estadual Paulista (Unesp)
- 4. Universidade de São Paulo Escola Superior de Agricultura Luiz de Queiroz, Universidade de São Paulo
- 5. Universidade Estadual Paulista Júlio de Mesquita Filho - Câmpus de Guaratinguetá
-
6.
Brazilian Agricultural Research Corporation
Description
This dataset provides the complete validation results and reproducibility package for EVAonline, an open-source system designed for high-accuracy reference evapotranspiration (ETo) estimation using multi-source data fusion. This version updates the deposit with fully referenced data sources, an expanded four-source validation, robustness (sensitivity) analyses, and software test/reproducibility metadata.
📦 What's Included
- 186,286 daily ETo observations spanning 30 years (1991–2020) across 17 Brazilian cities (MATOPIBA region: Maranhão, Tocantins, Piauí, Bahia — plus Piracicaba/SP as an out-of-region control site). 17 cities × 10,958 days = 186,286 daily values.
- 1,304,002 daily meteorological records per reanalysis source (7 variables × 17 cities × 10,958 days) for both NASA POWER and Open-Meteo, with 0.00% missing data.
- Four ETo sources compared against the reference:
- Reference (ground truth): Xavier et al. BR-DWGD (0.1° ≈ 10 km resolution).
- NASA POWER ETo, computed in-house with the FAO-56 Penman-Monteith engine (underlying reanalysis: MERRA-2, ~0.5° × 0.625°).
- Open-Meteo ETo, computed in-house with FAO-56 PM (underlying reanalysis: ERA5-Land, ~9 km).
- Open-Meteo API ETo, pre-computed by the Open-Meteo service (ERA5-Land), used as an independent benchmark.
- EVAonline — Adaptive Kalman Fusion product ⭐.
📂 Repository Structure
- 📁 data/ — raw inputs, intermediate files, descriptive statistics, per-source ETo, per-city validations, the four-source comparison, sensitivity analyses, and final figures.
- 📁 notebooks/ — demonstration notebooks for each provider (NASA POWER, Open-Meteo Archive/Forecast, MET Norway, NWS Forecast, NWS Stations) plus utility examples.
- 📁 scripts/ — API adapters (NASA POWER, Open-Meteo, MET Norway, NWS, OpenTopoData), the Kalman fusion core, the FAO-56 Penman-Monteith engine, and the sequential pipeline (scripts 1–7) to reproduce every manuscript result, including the two sensitivity scripts.
- 📁 validation/ — supplementary, unit-tested validation analyses reported in the article: temporal-split + leave-one-site-out cross-validation, site-wise metrics (median/IQR/min–max across the 17 sites), a component ablation (variable/component importance), and the error-boxplot figure. A single tested
metrics.pyprovides one definition of KGE/NSE/R²/MAE/RMSE/PBIAS. - 📄 Root files —
complete_validation_analysis.ipynb(main entry point, all 17 cities),tutorial_full_pipeline.ipynb(step-by-step single-city guide using Piracicaba/SP),environment.ymlandrequirements.txt(Python 3.12 environment).
🗄️ Data Sources & References
All data sources used to build and validate EVAonline are documented below with their original scientific and data citations. Full bibliographic entries are provided in the References section.
| Source | Role | Native resolution | Underlying dataset | Primary citation |
|---|---|---|---|---|
| Xavier BR-DWGD | Reference / ground truth | 0.1° (~10 km) | Brazilian Daily Weather Gridded Data | Xavier et al. (2022), Int. J. Climatol. — doi:10.1002/joc.7731 |
| NASA POWER | Global reanalysis (validation) | ~0.5° × 0.625° | MERRA-2 | NASA LaRC POWER Project; Gelaro et al. (2017) — doi:10.1175/JCLI-D-16-0758.1 |
| Open-Meteo Archive | High-res reanalysis (validation + benchmark) | 0.1° (~9 km) | ERA5-Land / ERA5 | Zippenfenig (2023) — doi:10.5281/zenodo.7970649; Muñoz-Sabater et al. (2021) — doi:10.5194/essd-13-4349-2021 |
| Open-Meteo Forecast | Global forecast (platform) | 0.1° (~9 km) | ECMWF IFS / ERA5 | Zippenfenig (2023) — doi:10.5281/zenodo.7970649 |
| MET Norway | Regional forecast (platform, Europe) | ~1 km | Locationforecast 2.0 | MET Norway (NLOD / CC BY 4.0) — api.met.no |
| NWS (USA) | Regional forecast/stations (platform) | Station/grid | NWS/NOAA API | National Weather Service (NOAA), public domain — weather.gov |
| Open Topo Data | Station elevation | ~30 m | SRTM / ASTER GDEM v3 | Nisbet (2019); Farr et al. (2007) — doi:10.1029/2005RG000183; NASA/METI ASTER GDEM v3 — doi:10.5067/ASTER/ASTGTM.003 |
Data attribution required by providers (following each source's official referencing/licensing page):
- NASA POWER (https://power.larc.nasa.gov/docs/referencing/): include both references. Project reference: "The data was obtained from National Aeronautics and Space Administration (NASA) Langley Research Center's Prediction Of Worldwide Energy Resources (POWER) project funded through the NASA Earth Science Division." Data reference: "The data was obtained from the POWER Project's Daily API (data version v10.2.1) on 2025/11/20." Data under CC BY 4.0.
- Open-Meteo (https://open-meteo.com/en/licence): cite the API software as Zippenfenig (2023), doi:10.5281/zenodo.7970649 (CC BY 4.0); underlying data from Copernicus C3S ERA5-Land (CC BY 4.0).
- MET Norway (https://www.met.no/en/free-meteorological-data/Licensing-and-crediting): licensed under NLOD and CC BY 4.0; credit "Based on data from MET Norway".
- NWS / NOAA (https://www.weather.gov/disclaimer): information produced by the U.S. National Weather Service is in the public domain (U.S. Government work).
- Elevation: point queries served by the Open Topo Data API (Nisbet, 2019) over the SRTM (Farr et al., 2007) and ASTER GDEM v3 (NASA/METI, 2019) digital elevation models (~30 m).
- Xavier BR-DWGD: Brazilian daily weather gridded product; cite Xavier et al. (2022).
🔬 Methodology Notes
Input vs. output distinction:
nasa_power_rawandopen_meteo_rawcontain raw meteorological variables (T2M_MAX, T2M_MIN, T2M, RH2M, WS2M, ALLSKY_SFC_SW_DWN, PRECTOTCORR) used as inputs to EVAonline's FAO-56 PM engine.eto_open_meteocontains ETo values pre-calculated by the Open-Meteo service, used for independent benchmarking (sourceOPENMETEO_API).
Reference evapotranspiration (FAO-56 Penman-Monteith): ETo is computed with the standard FAO-56 PM equation (Allen et al., 1998). Wind-speed handling is critical: NASA POWER provides native 2 m wind, whereas Open-Meteo provides 10 m wind that must be converted to 2 m using FAO-56 Eq. 47 (omitting this conversion causes ~15% ETo overestimation). Station elevation is fetched from OpenTopoData.
Adaptive Kalman Fusion: EVAonline combines NASA POWER and Open-Meteo estimates through an adaptive Kalman filter (Kalman, 1960) with (i) seasonal process noise derived from Xavier monthly variability, (ii) measurement noise reflecting relative source uncertainty (R_NASA ≈ 0.3, R_Open-Meteo ≈ 0.4), and (iii) bias correction anchored to Xavier BR-DWGD climatology, producing a fused ETo plus per-day uncertainty.
Validation scope: focuses on the mathematical core (Kalman filter, weighted fusion, FAO-56 engine) — not the full web platform (FastAPI, Celery, Redis).
Climatological rigor: the 30-year period (1991–2020) follows WMO guidelines for climate normals.
Evaluation metrics: coefficient of determination (R²), Kling-Gupta Efficiency (KGE; Gupta et al., 2009), Nash-Sutcliffe Efficiency (NSE; Nash & Sutcliffe, 1970), mean absolute error (MAE), root mean squared error (RMSE), and percent bias (PBIAS; Moriasi et al., 2007).
📈 Scientific Performance
Mean performance across all 17 cities, evaluated against the Xavier BR-DWGD reference (n = 10,958 days per city):
| Metric | EVAonline (Fusion) | Open-Meteo API (ERA5-Land) | NASA POWER (FAO-56) | Open-Meteo (FAO-56) |
|---|---|---|---|---|
| R² | 0.694 | 0.649 | 0.740 | 0.636 |
| KGE | 0.814 | 0.584 | 0.411 | 0.432 |
| NSE | 0.676 | 0.216 | −0.363 | −0.547 |
| MAE (mm/d) | 0.423 | 0.690 | 0.845 | 0.859 |
| RMSE (mm/d) | 0.566 | 0.860 | 1.117 | 1.097 |
| PBIAS (%) | +0.71 | +8.27 | +15.78 | +13.02 |
Key improvements of EVAonline over individual global sources:
- MAE reduction of 38.7–50.8% (0.423 mm/d vs. 0.690–0.859 mm/d).
- Bias reduction of 91.4–95.5% (PBIAS +0.71% vs. +8.27% to +15.78%).
- Highest KGE (0.814) — 39% higher than the best individual source (Open-Meteo API) and up to 98% higher than NASA POWER.
- Only source with positive NSE across all cities, and the most consistent performance (smallest inter-city metric spread: KGE SD = 0.05 vs. 0.19–0.41 for individual sources).
Robustness / sensitivity analyses (data/analysis_results/):
- Fusion weights (
sensitivity_weights.csv): perturbing the source weights by ±20% changes mean RMSE and MAE by less than ~4%, confirming the fusion is not over-tuned. - Kalman parameters (
sensitivity_kalman_params.csv): sweeping the process/measurement-noise ratio, base measurement variance, and clipping factors leaves KGE within ~0.75–0.85, confirming stability of the filter configuration.
🧪 Reproducibility & Testing
- Deterministic pipeline: scripts 1→7 regenerate every table, figure, and CSV in this deposit from the raw inputs; the two notebooks reproduce the single-city tutorial and the full 17-city study end-to-end.
- Environment: pinned Python 3.12 environment via
environment.yml(conda) andrequirements.txt(pip). - Automated software tests: the EVAonline codebase is covered by a
pytestsuite (unit + integration) with 86.5% coverage (as reported in the associated article) across the core modules (FAO-56 engine, Kalman ensemble, API adapters, preprocessing); coverage reports (coverage.xml) are version-controlled in the source repository. The supplementaryvalidation/package adds its own unit tests for the metric definitions. - Data integrity: primary datasets carry MD5 checksums in the Zenodo deposit; raw datasets contain 0.00% missing values over 1991–2020.
📚 References
Data sources
- Xavier, A. C., Scanlon, B. R., King, C. W., & Alves, A. I. (2022). New improved Brazilian daily weather gridded data (1961–2020). International Journal of Climatology, 42(16), 8390–8404. https://doi.org/10.1002/joc.7731
- Xavier, A. C., King, C. W., & Scanlon, B. R. (2016). Daily gridded meteorological variables in Brazil (1980–2013). International Journal of Climatology, 36(6), 2644–2659. https://doi.org/10.1002/joc.4518
- Gelaro, R., McCarty, W., Suárez, M. J., et al. (2017). The Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2). Journal of Climate, 30(14), 5419–5454. https://doi.org/10.1175/JCLI-D-16-0758.1
- NASA Langley Research Center, Prediction Of Worldwide Energy Resources (POWER) project, funded through the NASA Earth Science Division. Daily API (data version v10.2.1), accessed 2025/11/20. https://power.larc.nasa.gov/ (referencing guide: https://power.larc.nasa.gov/docs/referencing/)
- Zippenfenig, P. (2023). Open-Meteo.com Weather API (Computer software). Zenodo. https://doi.org/10.5281/zenodo.7970649 (CC BY 4.0; https://open-meteo.com/en/licence)
- Muñoz-Sabater, J., Dutra, E., Agustí-Panareda, A., et al. (2021). ERA5-Land: a state-of-the-art global reanalysis dataset for land applications. Earth System Science Data, 13(9), 4349–4383. https://doi.org/10.5194/essd-13-4349-2021
- Hersbach, H., Bell, B., Berrisford, P., et al. (2020). The ERA5 global reanalysis. Quarterly Journal of the Royal Meteorological Society, 146(730), 1999–2049. https://doi.org/10.1002/qj.3803
- Norwegian Meteorological Institute (MET Norway) (2026). Locationforecast API — Free meteorological data. https://api.met.no/ (NLOD / CC BY 4.0; https://www.met.no/en/free-meteorological-data/Licensing-and-crediting)
- National Weather Service (NOAA), U.S. Department of Commerce (2026). National Weather Service API. https://www.weather.gov/documentation/services-web-api (public domain; https://www.weather.gov/disclaimer)
- Nisbet, A. (2019). Open Topo Data: An elevation API. https://www.opentopodata.org/ (accessed 2026-08-03)
- Farr, T. G., Rosen, P. A., Caro, E., et al. (2007). The Shuttle Radar Topography Mission. Reviews of Geophysics, 45(2), RG2004. https://doi.org/10.1029/2005RG000183
- NASA/METI/AIST/Japan Spacesystems, and U.S./Japan ASTER Science Team (2019). ASTER Global Digital Elevation Model V003. NASA EOSDIS Land Processes DAAC. https://doi.org/10.5067/ASTER/ASTGTM.003
Methods & metrics
- Allen, R. G., Pereira, L. S., Raes, D., & Smith, M. (1998). Crop evapotranspiration — Guidelines for computing crop water requirements. FAO Irrigation and Drainage Paper 56. FAO, Rome.
- Kalman, R. E. (1960). A new approach to linear filtering and prediction problems. Journal of Basic Engineering, 82(1), 35–45. https://doi.org/10.1115/1.3662552
- Gupta, H. V., Kling, H., Yilmaz, K. K., & Martinez, G. F. (2009). Decomposition of the mean squared error and NSE performance criteria: Implications for improving hydrological modelling. Journal of Hydrology, 377(1–2), 80–91. https://doi.org/10.1016/j.jhydrol.2009.08.003
- Nash, J. E., & Sutcliffe, J. V. (1970). River flow forecasting through conceptual models part I — A discussion of principles. Journal of Hydrology, 10(3), 282–290. https://doi.org/10.1016/0022-1694(70)90255-6
- Moriasi, D. N., Arnold, J. G., Van Liew, M. W., et al. (2007). Model evaluation guidelines for systematic quantification of accuracy in watershed simulations. Transactions of the ASABE, 50(3), 885–900. https://doi.org/10.13031/2013.23153
📄 License and Citation
- Data license: Creative Commons Attribution 4.0 International (CC BY 4.0).
- Software license: GNU AGPL-3.0 (see GitHub repository).
Recommended citation:
Soares, A.S.M.C., Ribeiro, V.P., Duarte, S.N., Balestieri, J.A.P., Padovani, C.R., Bordignon, Á.J.Z., Maciel, C.D., & Marques, P.A.A. EVAonline: An open-source web platform for global reference evapotranspiration estimation via multi-source data fusion (1.0.1) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.19339254
Associated article:
Soares, A.S.M.C., Ribeiro, V.P., Duarte, S.N., Balestieri, J.A.P., Padovani, C.R., Bordignon, Á.J.Z., Maciel, C.D., & Marques, P.A.A. (2026). EVAonline: An open-source web platform for global reference evapotranspiration estimation via multi-source data fusion. Environmental Modelling & Software, 204, 107113. https://doi.org/10.1016/j.envsoft.2026.107113
🔗 Links
- Live repository: https://github.com/angela-cunha-soares/EVAONLINE
- Associated article (DOI): https://doi.org/10.1016/j.envsoft.2026.107113
- Reference data (BR-DWGD): https://doi.org/10.1002/joc.7731
- NASA POWER: https://power.larc.nasa.gov/
- Open-Meteo: https://open-meteo.com/ · https://doi.org/10.5281/zenodo.7970649
- ERA5-Land: https://doi.org/10.5194/essd-13-4349-2021
Files
EVAonline_validation.zip
Additional details
Identifiers
Funding
- Coordenação de Aperfeiçoamento de Pessoal de Nível Superior
- PROEX – Finance Code 001 88887.804895/2023-00
- Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP)
- 2021/02464-0
- Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP)
- 2019/07665-4
Dates
- Issued
-
2026-03-30Version 1.0.2 - Updated all data-source references and attributions (Xavier BR-DWGD, NASA POWER/MERRA-2, Open-Meteo/ERA5-Land, MET Norway, NWS, Open Topo Data/SRTM/ASTER) to their official citation formats; added a supplementary, unit-tested validation/ package (temporal-split + leave-one-site-out cross-validation, site-wise metrics, component ablation, and the error-boxplot figure); removed unnecessary files. No changes to the underlying data.
Software
- Repository URL
- https://github.com/angela-cunha-soares/EVAONLINE
- Programming language
- Python
- Development Status
- Active
References
- Xavier, A. C., Scanlon, B. R., King, C. W., & Alves, A. I. (2022). New improved Brazilian daily weather gridded data (1961–2020). International Journal of Climatology, 42(16), 8390–8404. https://doi.org/10.1002/joc.7731
- Xavier, A. C., King, C. W., & Scanlon, B. R. (2016). Daily gridded meteorological variables in Brazil (1980–2013). International Journal of Climatology, 36(6), 2644–2659. https://doi.org/10.1002/joc.4518
- Gelaro, R., McCarty, W., Suárez, M. J., et al. (2017). The Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2). Journal of Climate, 30(14), 5419–5454. https://doi.org/10.1175/JCLI-D-16-0758.1
- NASA Langley Research Center POWER Project. Prediction Of Worldwide Energy Resources (POWER), Daily API (data version v10.2.1), accessed 2025/11/20. https://power.larc.nasa.gov/
- Zippenfenig, P. (2023). Open-Meteo.com Weather API. Zenodo. https://doi.org/10.5281/zenodo.7970649
- Muñoz-Sabater, J., Dutra, E., Agustí-Panareda, A., et al. (2021). ERA5-Land: a state-of-the-art global reanalysis dataset for land applications. Earth System Science Data, 13(9), 4349–4383. https://doi.org/10.5194/essd-13-4349-2021
- Hersbach, H., Bell, B., Berrisford, P., et al. (2020). The ERA5 global reanalysis. Quarterly Journal of the Royal Meteorological Society, 146(730), 1999–2049. https://doi.org/10.1002/qj.3803
- Norwegian Meteorological Institute (MET Norway). Locationforecast API. https://api.met.no/ (Licensed under NLOD and CC BY 4.0; "Based on data from MET Norway").
- National Weather Service (NOAA), U.S. Department of Commerce. National Weather Service API. https://www.weather.gov/documentation/services-web-api (public domain).
- Nisbet, A. (2019). Open Topo Data: An elevation API. https://www.opentopodata.org/
- Farr, T. G., Rosen, P. A., Caro, E., et al. (2007). The Shuttle Radar Topography Mission. Reviews of Geophysics, 45(2), RG2004. https://doi.org/10.1029/2005RG000183
- NASA/METI/AIST/Japan Spacesystems, & U.S./Japan ASTER Science Team (2019). ASTER Global Digital Elevation Model V003. NASA EOSDIS Land Processes DAAC. https://doi.org/10.5067/ASTER/ASTGTM.003
- Allen, R. G., Pereira, L. S., Raes, D., & Smith, M. (1998). Crop evapotranspiration — Guidelines for computing crop water requirements. FAO Irrigation and Drainage Paper 56. FAO, Rome.
- Kalman, R. E. (1960). A new approach to linear filtering and prediction problems. Journal of Basic Engineering, 82(1), 35–45. https://doi.org/10.1115/1.3662552
- Gupta, H. V., Kling, H., Yilmaz, K. K., & Martinez, G. F. (2009). Decomposition of the mean squared error and NSE performance criteria: Implications for improving hydrological modelling. Journal of Hydrology, 377(1–2), 80–91. https://doi.org/10.1016/j.jhydrol.2009.08.003
- Nash, J. E., & Sutcliffe, J. V. (1970). River flow forecasting through conceptual models part I — A discussion of principles. Journal of Hydrology, 10(3), 282–290. https://doi.org/10.1016/0022-1694(70)90255-6
- Moriasi, D. N., Arnold, J. G., Van Liew, M. W., et al. (2007). Model evaluation guidelines for systematic quantification of accuracy in watershed simulations. Transactions of the ASABE, 50(3), 885–900. https://doi.org/10.13031/2013.23153