Could you describe how your professional and/or personal experience relate to water? Where does your interest in space technology for water come from?
I have a solid understanding of the fundamentals of hydrologic and hydraulic engineering, which is relevant to water. I studied many courses in my undergraduate and postgraduate degrees where I learned how runoff in a watershed is generated from meteorological parameters including rainfall, evapotranspiration and infiltration. I also applied my theoretical knowledge to various projects. Using hydrologic tools and indices, I analyzed floods and droughts in different regions. Producing a reliable flood hazard map is indeed a challenging task in urban areas, especially when there is limited data. This lack of data is one of major reasons that motivates me to utilize space technology in quantifying such hydrologic extremes. Also, undertaking a hydrologic analysis demand many types of data, sometimes it is impossible to obtain these kinds of observations. For example, evapotranspiration is an essential factor in long-term hydrological applications and drought studies. However, it is not easily measurable. In this context, space technology plays an essential role by providing reliable evapotranspiration records. Additionally, measured records often contain long gaps and missing data, which can be filled by satellite and remotely sensed data. Space technology not only provides valuable input to hydrological analysis but it also serves as a reliable means of validating output of such studies. For instance, a flood hazard map produced from a hydraulic model could be validated through satellite measurements.
You recently completed your MSc degree with a thesis titled “Streamflow under climate change scenarios in Kelani River Basin, Sri Lanka”. Please tell us more about your research; expand on space technologies and data used, and the methods used to meet your research objectives.
In my thesis, I evaluated the changes streamflow under climate change scenarios at the basin scale, Kelani Basin in Sri Lanka. Using precipitation and temperature from CMIP6 projections, runoff was simulated over historical period (1985-2014), near future (2030-2060) and long-term future under SSPs 2-4.5 (business-as-usual scenario where no adaptation or mitigation exist with moderate radiative forcing) and 5-8.5 (worst case scenario driven by intense use of fossil fuels and unsustainable practices with highest radiative forcing) (O’Neill et.al 2016). The HEC-HMS was applied to simulate streamflow. To set up the model, several remotely sensed data were applied, including Shuttle Radar Topography Mission (SRTM) Digital Elevation Model (DEM) and ESRI global land use (Esri Releases New 2020 Global Land Cover Map - Esri).
What were the main insights and trends you got from this research?
Kelani Basin receives significant rainfall during monsoon periods, from May to September. As such, it experiences major flood events in this period. The downstream of the basin is also heavily urbanized coupled with lack of proper water management infrastructures that further exacerbate flood risk. My research finding suggested that this issue might become worse in the future under fossil fuel driven scenario, SSP5-8.5. SSP5-8.5 refers to combination of Shared Socioeconomic Pathway (SSP5) and Representative Concentration Pathway (RCP 8.5). SSP5 envisions a future driven by intense use of fossil fuels and unsustainable practices. RCP 8.5 is highest radiative forcing, which originate from the fifth phase of the Coupled Model Intercomparison Project (CMIP5) (O’Neill et. al 2016). The rainfall, temperature and discharge trends were increasing from reference period to future periods. The changes were much higher under SSP5-8.5, especially the long-term changes. Interestingly, significant increase in discharge was projected to happen in the monsoon season. Assessment of different streamflow regions also revealed high increaser for high flows. This means flood risk is likely to worsen in the coming decades.
As a hydrological modeling specialist, your work involves the application of big data for water resources management in cloud-based computing platforms such as the Google Earth Engine (GEE). In your view, how have cloud-based computing platforms revolutionized remote sensing and how is this benefiting the users?
Hydrologists often utilize big geospatial and weather data for relevant applications. From my experience, cloud computation platforms have made it much easier to access, compute and visualize these datasets. Without cloud-based platforms, first accessing a dataset is time-consuming, because users need to make an account and register in the platform. Then they have to read the data, which is stored in specific file format. Even after accessing data, there is a lot of processing needed to do desirable computation and analysis. This not only demands knowledge about workflows but it also creates additional room for error. Visualizing data is another challenging task, but not so when performed on cloud platforms. Working in the cloud is much faster than the traditional route. With one single run – within a few seconds, users can compute, apply algorithms and prepare an output with magnificent charts, graphs and maps.
An important advantage of cloud platforms is their capability to interact with users. For example, when an interested user wants to prepare a land use map from remotely sensed data in Google Earth Engine, the user can generate the land use map by retrieving and combining spectral bands and then applying a classification algorithm, supervised or unsupervised algorithm. When retrieving spectral bands of a dataset, there are several filtering criteria such as date, region of interest, cloud percentage are required. The intended user can change any of these parameters and then visualize the results in the map canvas to see whether satellite data of certain quality exist in the region of interest. This is also quite helpful to find out particular satellite dataset that has desired quality. Similarly, when applying supervised/supervised classification algorithms, the parameter could be changed on the fly, and its effect examined, which is similar to a trial-and-error process in which classification parameters are adjusted to fine-tune results. Managing all this workflow in a cloud-based platform like GEE can significantly minimize errors. Users who have limited skills and knowledge also benefit from cloud platforms. As they work and interact with a particular platform, they learn a great deal by comparing the changes of an output from the corresponding input.
Tell us more about the findings in your recent publication where you evaluated the performance of Coupled Model Intercomparison Phase 6 (CMIP6) in projecting temperature and precipitation over Afghanistan. What do these findings mean for water resources management and water-related disaster mitigation in the country?
This publication assessed the behaviour of rainfall and temperature under climate change scenarios – SSP2-4.5 and SSP5-8.5. We applied statistical and probabilistic approaches to quantify changes in rainfall and temperature from the reference period to the projected periods. Overall, the study showed a significant rise in rainfall and temperature towards the end of 21st century, with changes being more pronounced under SSP5-8.5. We also concluded that higher rainfall values increase more compared to medium and lower rainfalls under SSP5-8.5.
The study provided useful insights related to the variation of extreme rainfall, as it showed a significant increase in extreme rainfall events. Such a rise in extreme rainfall triggers serious floods and erosion issues. Given Afghanistan’s lack of proper water infrastructure, its impact could be exacerbated and result in serious damages to lives and properties of vulnerable people. In this context, the finding of the study will help stakeholders engaged in water resources management to consider hazards like floods an important factor in their decision- making and planning.
Due to their coarse spatial resolution, global climate model (GCM) outputs usually require downscaling for them to be able to inform regional and local adaptation. What are the main approaches to downscaling global climate model outputs?
Downscaling is an essential step for using GCM outputs in regional and local adaptation planning. Different approaches to downscaling have their strengths and limitations, and the choice of method depends on the specific application and available resources. To my knowledge statistical and dynamical downscaling are frequently used approaches.
Statistical downscaling involves developing statistical relationships between large-scale GCM outputs and local climate variables. The statistical relationships are then used to generate downscaled climate projections for specific locations. Examples of statistical downscaling methods include regression-based models, artificial neural networks, and decision trees. Bias correction can also be classified as a statistical downscaling approach. This approach involves using statistical methods to adjust GCM outputs to remove systematic biases. It is typically applied to correct the bias of GCM outputs that are used as input to RCMs or other applications.
Dynamical downscaling involves using regional climate models (RCMs) to simulate climate at a higher spatial resolution than GCMs. RCMs use the large-scale GCM outputs as boundary conditions and downscale them to produce detailed climate projections at the regional scale. Dynamical downscaling is computationally intensive and requires high-performance computing resources. Even a dynamically downscaled climate model has a systematic bias. For local applications, it is recommended to apply bias correction on these models as well.
Can artificial intelligence and data-driven machine learning models match or even replace process-driven hydrologic models for streamflow simulation?
In my view, artificial intelligence could not replace process-based models. Each approach has its strengths and weaknesses, which makes it suitable for a particular situation. Process-driven hydrologic models rely on a deep understanding of the physical processes that control the movement of water through the hydrologic cycle. They are designed to explicitly simulate the underlying physical processes and are typically calibrated using measured data to improve their accuracy. These models have been widely used for decades and are well-established in the hydrologic community. Developing a process-based model requires a significant workflow and skills. To set up a process-based hydrologic model, users need to prepare several input data in certain formats and then select methods/settings to run the model. Successively, it is required to calibrate and validate the model to optimise its performance for the particular datasets and selected methods. These models are computationally demanding and require significant time to run/calibrate them.
On the other hand, data-driven ML models establish a relationship between input and output by learning patterns that exist in the input dataset. They can capture complex and nonlinear relationships that may not be well understood or explicitly modelled in process-driven models. AI and ML are especially effective where the underlying physical processes are poorly understood or difficult to model. Compared to process-based models, they also require less resources and time to develop.
What would you say are the main challenges in applying machine learning models for hydrological inference?
While AI and ML models have great potential for simulating streamflow, there are still challenges and limitations. They require large amounts of data to train effectively, which can be a limitation in areas where data are scarce or unreliable. The other issue is so-called "black-box" models, meaning that it can be difficult to interpret the relationships the algorithms learn. They may also struggle to generalize to new situations or conditions that differ significantly from the data used to train them.
What do you need to innovate? Consider the working environment, people, circumstances, funding, what you need to be creative and what inspires you or contributes to your work on novel approaches to water management and hydrology.
For me being creative and curious about learning existing practices and challenging them would lead to innovation. Being open-minded, curious, and willing to take risks helps me to discover new approaches and solutions. Interacting and networking with colleagues, stakeholders, and experts in other fields can help generate new ideas, foster innovation, and facilitate the implementation of new approaches. Moreover, having access to reliable and comprehensive data, as well as appropriate resources and tools, can help facilitate the development and implementation of innovative approaches.
As a young professional applying geospatial techniques for hydrological modeling, how do you keep yourself updated with the fast advancements in the field?
Whenever I get a chance to attend relevant conferences and workshops in the field, I will certainly not miss it. It helps me tremendously to learn about the latest advancements and network with other professionals. This can provide opportunities to hear about cutting-edge research and technologies, as well as to connect with experts in the field.
Reading scientific publications is another approach that I use to stay up-to-date with the latest research and developments. By reading scientific publications, I get to learn about trending issues that other scientists are interested in.
In my free time, I like to take online courses or join webinars/online communities. This is a convenient way to learn new skills and stay on track with advancements in the field.
Which online resources to learn about EO data for water management and hydrology can you recommend? Do you also have any recommendations for learning about data-driven machine learning models for water?
I would recommend courses and webinars related to hydrological modeling and geospatial techniques on platforms such as Coursera, Udemy, or Esri. Coursera is an interesting choice as it has an abundant number of course and provides financial aid to complete courses.
When it comes to resources, there are several valuable ones for water-related applications. The most important resource in my view is Google Earth Engine since it has a vast number of datasets that could be applied in hydrology. The platform also has an interactive interface and users can rapidly access and process their desired data.
OPenDAP is a scientific archiving system with huge amounts of data including satellite measurements, observations, and climate model data. It is certainly useful for water professionals. The advantage of using this platform is that one can access data remotely through cloud-based computation platforms such as xarray in python, without downloading it and spending a lot of time on the web. Users can process their desirable data through this interface, which is much faster and more efficient.
What is your favourite aggregate state of water and why?
My favourite state of water is its frozen form. I am used to skidding on snow and ice from a very early age. It is immensely pleasing to jump in the air and to ride on the steep edge of mountains. Covered with a white and clean snowpack, mountains look much more attractive.