Dr Muhammad Hasan

Postdoctoral Research Fellow

What do you do?

I am a Postdoctoral Research Fellow at LEEP, involved in the ADD-TREES project under professors from the University of Exeter and UCL. I hold a PhD in Mathematical Sciences from Durham University, and was awarded a prestigious Durham Doctoral Scholarship. Previously, I was a Lecturer and Assistant Professor of Statistics at the University of Dhaka, Bangladesh. My expertise lies in uncertainty quantification and decision-making, developing emulators with Deep Gaussian processes and Bayes linear methods, focusing on model calibration and high-dimensional data analysis.

How are you involved in this area of science? 

My research focuses on embedding Uncertainty Quantification (UQ) into crop models and farmer behaviour modelling, with an emphasis on automatic calibration and data assimilation.

To address the challenge of high-dimensional inputs in crop models, I modified an active subspace method, which identifies the linear combinations of inputs that most influence the output. This unique approach improves the handling of large input spaces, making it highly effective for emulation and uncertainty propagation. We have access to yield time series data from thousands of land parcels across the UK, along with some management information. However, critical details like fertilizer amounts, planting dates, and fertilization schedules are often missing, making it difficult to compare simulation results directly with real data.

To address these unknown management variables, we developed a hierarchical calibration framework. This framework integrates nested Bayesian optimization and linked deep Gaussian surrogates, which connect emulator models of different crops within a crop rotation, allowing uncertainties to propagate across multiple crops grown in sequence.

The framework also addresses the hierarchy of crop systems, which refers to the various scales of crop production—from individual field parcels to entire farms. By incorporating data from 3,500 farms and 2.5 million land parcels across the UK, this approach optimizes crop management practices more effectively. It enables the calibration of models at both local and larger scales, improving yield predictions, reducing food loss, and helping crop systems adapt to climate variability. Ultimately, this work will contribute to the UK’s net-zero goals by enhancing agricultural sustainability.

What do you love about this topic?

I love Uncertainty Quantification (UQ) because it adds precision and reliability to complex models across various important sectors. UQ helps us understand and manage the inherent variability and uncertainties in these systems, leading to more reliable predictions. It combines advanced math with real-world applications, solving practical problems. Improving decision-making and optimizing outcomes through UQ is not only intellectually stimulating but also deeply rewarding. It feels great to contribute to a field that has such a meaningful impact on diverse areas of life.

How does this work deliver real-world impact?

Uncertainty quantification enhances the accuracy of complex biophysical systems, enabling precise predictions and informed decisions. For example, our deep Gaussian process emulator significantly improves tree planting strategies. This supports the establishment of new woodlands in alignment with the UK Government Environment Act 2021. By planting half a million hectares of trees, our work contributes to substantial carbon capture, advancing the UK's strategy to achieve Net Zero by 2050. This methodological advancement ensures that our efforts have a real-world impact, helping to mitigate global climate change and promote sustainable land use.

Find out more:

• Read more about linked deep gaussian process emulation for model networks.
• Learn more about linked deep gaussian process emulation for stochastic imputation.
• Find out more about the ADD-TREES project.