As a PhD student delving into the wide world of evapotranspiration (ET) mapping for Ireland, my journey led me to the centre of Vienna, Austria, for the European Geosciences Union (EGU) General Assembly 2024 . The assembly, a gathering of 20,979 brilliant minds from 116 countries and cutting-edge research, offered a platform to explore the latest advancements in remote sensing, machine learning, and various methods pertaining to ET estimation and validation.
Under the auspices of a Met Éireann-Teagasc co-funded Walsh Ph.D. Scholarship, my research endeavours are focused on producing high-resolution gridded datasets of both Actual and Potential ET for Ireland. This ambitious objective aims to bridge critical gaps in our understanding of water flux dynamics, essential for hydrological modelling and agricultural decision-making in Ireland.
My first conference experience into this academic journey came in the form of a poster presentation in session HS6.3, titled "Evapotranspiration estimation using remote sensing and in-situ methods." Here, the stage was set to engage with fellow researchers, exchange ideas, and gain insights crucial for the progression of my doctoral journey.
As I reflect on my experience at EGU 2024, I am reminded of the insightful impact this gathering has had on better understanding the profound significance of my research. The insights gained from sessions covering a wide range of ET estimation and validation methodologies have not only enriched my understanding, but have also sparked new avenues of inquiry within my own work.
In the following sections of this blog post, I will discuss the key takeaways from the oral and poster sessions I attended, highlighting their relevance to my research objectives and the implications for my ongoing endeavours. From the evaluation of satellite-derived ET products to the integration of machine learning methods, join me on a journey to explore the challenges researchers face in ET mapping.
Session Reflection
Within the busy halls of the EGU 2024 General Assembly, one particular session resonated deeply with the core objectives of my research: HS6.3 - Evapotranspiration estimation using remote sensing and in-situ methods. This session, a union of multidisciplinary expertise, represented the collective endeavour to uncover the complexities of ET estimation across varying spatial scales.
The session discussed the complex nature of ET estimation, acknowledging its crucial role in addressing pressing challenges such as climate change, drought, precision agriculture, and sustainable water management practices. With a keen focus on leveraging cutting-edge technologies and methodologies, the session aimed to explore the diverse range of approaches employed in quantifying ET from point-scale measurements to large-scale derivations using remote sensing.
Key themes in this sessions were:
- Methodological Innovation: From artificial intelligence and machine learning to data fusion and sharpening algorithms, the session emphasised the pivotal role of technological innovation in advancing ET estimation methodologies. The fusion of physical- and process-based models with empirical/statistical methods showcased the potential for bridging different scales while accounting for method-specific uncertainties.
- Scale Dependencies and Uncertainty Management: The session provided a platform for reflecting on the scale dependencies inherent in various ET estimation approaches and strategies to mitigate uncertainties and systematic biases. By facilitating cross-scale comparisons of remote sensing, modelled, and ground-based derived ET, the session highlighted the importance of robust validation, calibration, and upscaling techniques.
- Remote Sensing Applications: Remote sensing emerged as a keystone of ET estimation, offering unparalleled insights into spatiotemporal patterns and trends. From analysing trends in ET data to exploring the physical consistency of ET estimates over diverse landscapes, the session showcased the transformative potential of satellite-based observations in informing evidence-based decision-making and sustainable water management practices.
Abstracts of Relevance
Offering a unique perspective on machine learning applications in ET estimation, this study by Yucong Hu et al. presents a fundamental framework that combines unsupervised clustering methods and artificial neural networks to resemble calculations in process-based models. By discerning aerodynamics and energy processes and providing a space for potential underrepresented processes, the framework offers a novel approach to hybridising machine learning approaches and mechanisms. With a focus on interpretability and accuracy, this methodology holds promise for validating known and exploring unknown knowledge in ET estimation.
Similarly, Qingchen Xu et al., introduced a novel approach to generating high-resolution and long-term ET datasets by integrating diverse machine learning techniques and multiple ET products. By combining direct site observations and various ET datasets including remote sensing, machine learning outputs, and land surface models, the study produces fused datasets with improved spatiotemporal resolution and extended temporal coverage. This innovative data integration framework addresses the limitations of existing ET datasets, offering enhanced accuracy and applicability for global water, energy, and carbon cycle applications.
Matěj Orság et al., delve into the complexities of evaluating remote sensing-derived ET products for hydrological modelling applications. By employing a round-robin experiment and point-scale downscaling benchmarking criteria, the study aims to identify the most suitable ET products for integration into hydrological models. The exploration of different energy balance closure scenarios and the utilisation of the triple collocation method add depth to the evaluation process, highlighting the challenges and opportunities in selecting optimal ET products for assimilation approaches.
In another effort to evaluate remote sensing products, Rojin Nejad et al., investigated the spatial and temporal variability of ET estimates over Madagascar using remote sensing data. By evaluating five popular ET products and analysing their utility and accuracy across different climatic zones, the study provides valuable insights into the factors contributing to differences in ET estimates. The Budyko Curve analysis used in this research highlights the strengths and limitations of each dataset, offering a deeper understanding of ET estimation in regions with limited climate monitoring and diverse climatic conditions.
As estimating ET from remotely sensed data comes with its uncertainties, Albert Olioso et al., are focusing on the uncertainty quantification of ET estimation from remote sensing data, in their talk they introduced a multi-model, multi-data framework (EVASPA) for mapping ET and its uncertainty. By analysing uncertainties related to input variables and model formulations, the study provides insights into the sources of uncertainty in ET estimation. The proposed framework offers a systematic approach to characterising ET uncertainties, with implications for improving the accuracy and reliability of ET estimates in hydrological and climate studies.
Suhyb Salama also focused on the validation of satellite-based hydrological products and offered insights into the various statistical metrics commonly used by researchers to assess the goodness-of-fit (GoF) between satellite products' estimates and in-situ measurements. By developing a universal methodology for quantifying GoF, his study proposes a holistic measure that integrates multiple validation metrics, providing a comprehensive assessment of the accuracy of satellite-derived hydrological products. This approach addresses the challenges associated with comparing and interpreting disparate validation metrics, offering a unified framework for evaluating the fidelity of satellite-based records.
Implications for the ET4I Project
One of the highlights of my conference experience was presenting my poster in session HS6.3. The work presented in this poster is part of the broader EvapoTranspiration for Ireland (ET4I) project aiming to enhance ET modelling and develop high-resolution daily ET maps for Ireland. It focused on evaluating seven different satellite-derived ET products for Ireland by comparing them to ground-based measurements from lysimeters in four different locations in Ireland. The systematic evaluation highlighted the challenges of applying these products in regions with significant cloud cover, such as Ireland, and proposed avenues for advancing ET mapping techniques. This includes integrating machine learning with an ensemble of well-established satellite-based ET products and in-situ measurements from a recently installed extensive flux tower network (NASCO) and rescued historic lysimeter observations, to develop a high spatial and temporal resolution ET gridded dataset for Ireland.
The insights gained from the EGU 2024 sessions are positioned to have a positive impact on the approaches considered for the ET4I project. To begin with, the state-of-the-art machine learning approaches that were introduced have great potential to be utilised for the ET4I project. For instance, I could employ Yucong Hu et al.’s framework, which combines spectral clustering and artificial neural network regressions in machine learning, to resemble calculations in process-based models in the ET4I project. Alternatively, I could explore Qingchen Xu et al.’s framework, which integrates various machine learning approaches such as Automated Machine Learning, Deep Neural Network, Light Gradient Boosting Machine, and Random Forest with different available ET products to generate high-resolution, long-term ET estimates, and apply it to estimate ET in Ireland.
Additionally, I will incorporate a couple of ET products that were introduced to me by Matěj Orság et al., such as HOLAPS and GLDAS-NOAH. Furthermore, new validation tools like the Budyko framework, utilised by Rojin Najad et al., to provide insights into the validation of ET estimates, particularly in comparing observed ET values with estimates derived from models or remote sensing data, could be of great use in catchment areas. Another valuable validation tool I could utilise for deciding which ET products to include in the ensemble method is the unique GoF developed by Suhayb Salama, which integrates all validation metrics used and provides a collective measure of accuracy for satellite-derived hydrological products. Finally, in consideration of the ensemble approach, the EVASPA framework by Albert Olioso et al. could be employed to extend the ensemble modelling to a multi-model – multi-data framework, providing ET estimations along with an estimation uncertainty.
In conclusion, my participation in EGU 2024 has not only broadened my perspective on ET estimation and validation methodologies, but has also motivated me to advance my understanding of water-vegetation-atmosphere interactions in Ireland. Equipped with new insights and inspiration from the conference, I am eager to embark on the next phase of my research journey, confident in the positive impact it will have on our understanding of ET dynamics and its implications for hydrological modelling, agricultural decision-making, and sustainable water management practices in Ireland.
Stay tuned for more updates as I navigate the challenging landscape of evapotranspiration research, fueled by the insights gained from EGU 2024.
Haneen
References
Muhammad, H., Finkele, K., Flattery, P., Jarmain, C., Lanigan, G., and Sweeney, C.: Evapotranspiration for Ireland (ET4I): Ground-Truthing Satellite-Driven Evapotranspiration Products., EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-4530, https://doi.org/10.5194/egusphere-egu24-4530, 2024.
Hu, Y. and Jiang, Y.: Interpretably reconstruct physical processes with combined machine learning approaches, a case study of evapotranspiration, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-20579, https://doi.org/10.5194/egusphere-egu24-20579, 2024.
Xu, Q., Li, L., and Wei, Z.: A high-resolution (1d, 9km) and long-term (1950-2022) gridded evapotranspiration dataset, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-4278, https://doi.org/10.5194/egusphere-egu24-4278, 2024.
Orság, M., Fischer, M., García-García, A., Peng, J., Samaniego, L., and Trnka, M.: Evaluation of remote sensing actual evapotranspiration products for hydrological modeling applications, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-19308, https://doi.org/10.5194/egusphere-egu24-19308, 2024.
alimohammad nejad, R., D. Carrière, S., Olioso, A., and Oudin, L.: Exploring the physical consistency of evapotranspiration estimates over Madagascar using remote sensing., EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-3899, https://doi.org/10.5194/egusphere-egu24-3899, 2024.
Olioso, A., Mwangi, S., Desrutins, H., Sobrino, J., Skoković, D., Carrière, S., Farhani, N., Etchanchu, J., Demarty, J., Hu, T., Mallick, K., Jia, A., Buis, S., Weiss, M., Ollivier, C., and Boulet, G.: Multimodel – multidata simulations for mapping evapotranspiration and its uncertainty of estimation from remote sensing data, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-13113, https://doi.org/10.5194/egusphere-egu24-13113, 2024.
Salama, S.: Validation of satellite hydrological products: which goodness-of-fit to use?, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-15155, https://doi.org/10.5194/egusphere-egu24-15155, 2024.
