Impacts of Data Collection on Flood Forecasting in the LDCs: Evidence from Nepal by Al Fahel-Dobler-Idrisu
This project explores the extent to which data collection improves flood prediction, mitigating the vulnerability of LDCs to climate change.
Many LDCs are impacted by climate change – a global problem that they did not contribute to create. Nepal is no exception: the country’s CO2 emission in 2014 was 0.298 metric ton per capita (WorldBank, 2019) and was ranked #109 on the 2017 list of biggest polluters (Global Carbon Atlas, 2017). Yet, climate change has a tremendous impact on this landlocked country of the Himalayan range.
Converging studies on the impact of climate change in Nepal have identified a dual phenomenon of increased temperatures and increased erratic rainfall pattern (Malla, 2009; Maharjan, 2013). Such a phenomenon is likely to disrupt the economic and social structures of a country mainly constituted of rural communities relying on natural resources (Maharjan, 2011).
In particular, some research indicates that hydrological reactivation of earthquake-triggered landslides may happen as a result of precipitation runoff or other mechanisms involving changes of supply of groundwater or ice melt having at least indirect links to climate change (Kargelet al. 2015). Resulting floods have become a major concern for the country.
In such a context, flood prediction can have several positive benefits and can effectively contribute to the development adaptation measures in Nepal. Thus, we propose to carry out a study comparing 3 different models for flood prediction that incorporate various combinations of remotely sensed data and data collected in situ and offer solutions to improve the existing network of weather stations.
Is this proposal for a practice or a project?
What actions do you propose?
1. Context and motivation:
As previously emphasized, Nepal is prone to various kinds of disasters and has experienced many tragedies in its recent history. Yet, (extreme) rainfall seems to play a role in many of these. It is, for instance, estimated that when daily precipitation exceeds 144 mm, the risk of landslides on the Himalayan slopes is high (Dahal, 2008). Meanwhile, a time-series study of the mean rainfall has revealed an annual increase in average precipitation by 13 mm, and in the number of rainy days by 0.8 days/year (Devkota, 2014). Additionally, studies have shown that the average temperature has increased consistently and continuously, at a rate of 0.05 ?C/year from 1971 to 2005 (DHM, 2008) leading to fast(er) glacier melt and enhanced summer river flow over time (Devkota, 2014).
The combined effect of precipitation (through increased number of landslide and runoff) and temperature warming (causing ice melts) increases the frequency and the magnitude of floods especially in the summer.
Globally, floods are the most widespread climate-related hazard causing over one third of total economic loss from natural disaster (Talchabhadel & Sharma, 2014). Damage to infrastructure, loss of life and livelihoods, loss of habitats and biodiversity, strain in water resources and reduced agricultural production are some of the effects of landslides and floods. In Nepal, it is estimated that floods and landslides cause on average 626 million NPR in damage and the death of 300 people.
For instance, on August 10, 2017, Nepal had one of its worst floods in decades. The flood affected over 75,000 homes and led to the death of 123 people. Thirty-Five (35) out of 75 districts were affected by floods, landslides and heavy rains. (United Nations Office of the Resident Coordinator, 2017). The floods also resulted in the loss of 64,000 hectares of crops and a fall in the agriculture growth rate from 5% in FY2017 to 2.8% in FY2018. Nepal’s economy is largely dependent on agriculture which makes up over 25% of the country’s GDP and employs over 75% of the workforce. (The World Bank, 2019)Beyond this, the floods also led to power cuts, damage to water and sanitation infrastructure and the closure of public health centers.(Gautam et al., 2017).
Given the potential damage that floods can cause, the efficiency of their forecasting has increasingly been under scrutiny and a topic of public policy. Indeed, forecasts improvements have demonstrated to enable flood-related damage reduction up to 35% (Talchabhadel & Sharma, 2014).In other words, real-time and forecasted flood early warning system together with flood protection schemes play a crucial role the development of LDCs like Nepal.
2. Problem Statement:
As emphasized above, improving flood forecast can help avoid dramatic flood-related damages. Therefore, we believe that the enhancement of weather data collection (especially on precipitation and on river flow) should occupy a prime rank in the list of issues to be addressed in an effort to “reshape development pathways into prioritizing proactive disaster risk reduction instead of reactive measures in the devastating aftermath of the event” (from prompt).
This issue is a critical importance in Nepal as the country suffers from a very peculiar paradox: In comparison to other LDCs, Nepal has a relatively extensive network of Weather/precipitation stations, but yet has a poor Index for Risk Management (INFORM), i.e. a High overall INFORM score (INFORM, 2019), even among its peer countries. In particular, the country presents a high score to natural hazard and exposure (especially for earthquake and flood exposure of respectively 9.9 and 6.8) but also a lack of coping capacity due essentially to the poor disaster governance and absence of Disaster Risk Reduction (DRR) programs.
Such a paradox interrogates the adequacy of the network of weather stations across the country and their ability to properly collect the data necessary to efficiently forecast or predict floods.
In an effort to reduce the vulnerability of the country, the government of Nepal has started the Nepal Risk Reduction Consortium (NRRC) in collaboration with development and humanitarian partners (Gaire et al., 2015). The objective of the consortium is to address the aforementioned paradox and to explore possibilities to ameliorate early warning system, and rainfall monitoring as the upward trend in the number of floods currently experienced is likely to continue to rise.
This proposed project will directly address methods for estimating flood risk via a combination of in situ data collected by weather stations and remotely sensing from publicly available satellite data.
3. Proposed method:
This project will encompass 4 key steps:
1. Assessment of the ability the existing network of stations to collect the data required for an adequate forecasting of floods events:
What is required from a network of stations is to forecast the presence or absence of flood at each location in space based on the collection of data about relevant features of the environment at given location (typically where the recording stations sit). In other words, proper modelling of the presence or absence of flood is obtained through a process of interpolation: the presence or absence of flood at location between stations is derived (is a function of) the observed presence or absence of flood at the location of the stations. So, to assess the efficiency of the flood forecast, it is necessary to determine whether the stations are strategically located in space so to enable interpolations results within a small margin of error to reality.
This assessment requires interpolating past data collected by the network of stations and to compare these result with known records of flood within that timeframe. This assessment will determine the need for improvement and the scope of that need. We will train an algorithm to recognize the presence or absence of flooding events at the station locations at different points in time based on the following input data: topographic data (height relative to surrounding areas, slope), amount of precipitation at the recorded time, amount of precipitation in the last 24 hours, land use category. The algorithm will then be tested on a different sample of records to determine its performance. Then, we will compare this model with a new model that does not include precipitation data to assess the importance of this data in flood prediction.
2.Comparing the performance of the algorithm vis a vis alternative models:
The goal is to develop an alternative prediction model that would utilize remote sensing (satellite imaging) to determine the amount of precipitation at every location. Remote sensing as a means to collect data offers the advantage that it covers large areas (like a country) simultaneously. The model developed here would be a multivariate regression to determine the presence or absence of flood at different times at any location simultaneously, that is, without interpolation. The Tropical Rainfall Measuring Mission (TRMM), was a joint mission of the NASA and the Japan Aerospace Exploration Agency which operated a satellite between the years 1997 and 2014 to collect precipitation data worldwide to study rainfall for weather and climate research. Satellite precipitation data for Nepal are available in the form of a 3-hour or 24-hour cumulative value at:
Comparing the performance of the prediction model based on remote sensing and that of the initial algorithm (based on ground recording of precipitation) will enable us to identify an upper limit to the goodness of fit of any forecasting model based on interpolation.
We note that the treatment of remote sensing data is time consuming and requires skill and financial resources that may not be available in LDCs. Moreover, satellite imaging does not enable real-time monitoring as weather station on the ground can.
3. Combining both approach to determine a middle ground model to predict flood events:
For the reasons mentioned above, neither of the two approaches is an ideal way to forecast flood events. In this proposal, we will combine both to develop an algorithm that would determine, in a similar fashion, the presence or absence of flood at different location and time. Moreover, the comparison of output prediction with known past flood records will allow us to identify ways to improve the existing network. We will be able to, for instance, identify areas of the country where the model efficiently predicts flood events and others where it performs poorly (for reasons varying from scarcity of recorded events in that region to outlier behavior). We will then determine an efficient way to complement the existing network with new recording devices to improve the overall goodness of fit of the model and hence the accuracy of the forecasts.
4. In situ experiment using portable sensors devices:
Rainfall driven floods occur when the soil can no longer absorb further rainfall. To predict such events with high accuracy ultimately requires data on flooding but especially, it requires the determination of what the soil’s response to the rainfall will be at a given time. In that regard, physically-based models are difficult to implement and pose simulation challenges.
Modeling these processes using physics creates a challenge from a simulation point of view. It would require details of the topography, soil composition, and land cover, along with meteorological conditions and hydrometeorological quantities such as soil moisture (Basha et al., 2008). Because such models are difficult to implement, in situ data collection is mandatory. Yet, neither satellite data nor interpolation processes (based on an established network of weather stations) has the ability to adequately monitor the different local specific weather conditions. An alternative would then be to identify those local areas where traditional models do not perform well and install there, sensors in a local network with an architecture similar to that developed by (Basha et al., 2008).
Local sensor networks have the advantage of decreasing the computational intensity of modeling procedures as they aim at local validity only. Moreover, local models self-calibrate as the influx of new data help adapt to the latest conditions. So a given modeling procedure can be applied at different locations and are suitable for on-site and real-time implementations.
In order to provide adequate data on rainfall, air temperature and water flow, the network architecture must meet the following requirements (Adapted from (Basha et al., 2008)):
- Monitor events in large geographic regions of approximately 10000 km2
- Provide realtime communication of measurements
- Survive longterm element exposure, including the flood events
- Minimize costs
- Distribute among nodes the significant computation needed for the prediction
Who will take these actions?
The ultimate objective of the proposed project is to determine the optimal mix between a centralized flood forecast modelling process (relying on an established network of station and satellite coverage) and a series of complementary decentralized local sensors networks.
This proposal outlines a pragmatic approach to the problem of data collection for forecasting purposes, which does not entail field operation before its implementation phase.
In other words, the proposing team will be conducting phase 1) through 3) as they can be conducted remotely since all of the necessary data is accessible online. However, these phases will necessitate recurrent contact with the agencies providing the data (NASA and Nepalese government) and surveying local experts to determine the accuracy of our data sources and request their help to obtain critical information such as an extensive record of past flood events involving different magnitudes.
These first three phases of the project will be conducted at the University of Delaware, in a research lab, by our team. This will ensure that we have a complete control over the process and will thus guarantee optimal output. Moreover, it will allow us to reserve most of our grant budget to the ultimate phase of the project which entail to experience in situ, alternative sensors network.
During this last phase, we will conduct ourselves preliminary experiments to determine locally (that is, in the US), the performance of competing sensors network architectures. Yet, in its final phase, the project will necessarily require on-site (in Nepal) experiment and testing of these network designs. Therefore, for that last step of the research project, we expect to rely on both our contact on site which will involve partner local Universities, local governmental agencies and members of the Nepal Risk Reduction Consortium. Among the current members of the consortium, two organizations would be particularly useful to provide us with on-site technical support: The World Bank for and the United Nations Development Program (UNDP).
However, we note that, through the proposed project, we envision the development of a forecasting modelling design that would provide enough flexibility to local conditions to be implementable in a diversity of environments with the idea that local authorities or organizations should ultimately be in charge of that implementation. To that end, we expect that training will necessarily take place in the medium run to educate local communities on the use of the technology. This could, again, be achieved through our partner universities which would have to be involved in the training process for obvious language and cultural barriers.
Where will these actions be taken?
As mentioned above, the first three phases of the project will be conducted at the University of Delaware. The last phase of the project will require deployment of a pilot sensor network and its testing on-site. We believe that our team should have control over the pilot network because it is critical to the successful completion and large-scale implementation of the project. As a result, it will be necessary to first test the pilot in the US.
However, the ultimate location of the sensors in Nepal will be determined by the results of the analyses that we will conduct during phases 1) to 3). In fact, we note that the targeted objective is to optimize flood forecasting by determining the optimal combination between centralized and decentralized modelling. Preliminary indications about what an optimal combination should be will arise from the assessment phases of the need for improving the existing station network. Very likely though, the need for local sensor networks will be felt in remote areas that are away from existing weather stations and in locations where satellite imaging may not lead to an optimal result: for example, forested areas or abrupt mountain sides where satellites cannot properly survey the amount of rainfall hitting the ground.
In addition, specify the country or countries where these actions will be taken.
No country selected
No country selected
No country selected
No country selected
What impact will these actions have on greenhouse gas emissions and/or adapting to climate change?
The nature of the proposed project is such that it is difficult to provide any estimate of the reduction of harm resulting from the implementation of the designed sensors network. In fact, there are three main reasons preventing us from determining a priori the positive externalities of the such an implementation.
First, the need for implementation is determined by the assessment of the existing network of weather stations, and thus cannot be anticipated. Moreover, this proposal was elaborated to ensure that our team maintains a control over the different phases of the project. Yet, the scaling up of our pilot network architecture is not part of the project and more importantly is expected to be of the responsibility of local agencies and authorities. We offer, through our project to identify a failure in the system, offer a fix and advocate for intervention by demonstrating that the offered fix effectively addresses the initial problem. Ultimately, our proposal is not a direct adaptive or mitigative action on the causes or consequences of climate change. As a forecasting tool, its main contribution to addressing climate change is to help assess risk and offer guidance on how to decrease vulnerability to that risk.
Nevertheless, it should be noted that Nepal approved a National Adaptation Programme of Action (NAPA) in 2010 that identifies priority activities needed to adapt to the impact of climate change. NAPA is a plan submitted by LDCs to the United Nations Framework Convention on Climate Change (UNFCCC) to access funding for the most immediate and urgent climate change adaptation needs. Some of the most pressing adaptation concerns include the development of early warning systems in disaster prone areas. This project provides a model for flood forecasting which is one of the important climate change risks faced by Nepal. A flood forecasting model provides the following benefits in line with National adaptation objectives:
- It reduces the risks of loss of lives and properties in the event of a flood.
- Early risk assessments can be made thereby preventing or reducing economic losses.
- Risk mapping of rain flood disaster prone areas can be developed and/or improved upon.
- Public health risks can also be minimized from adequate preparation.
What are other key benefits?
An important benefit of the proposed project is that it aims at developing an effective sensor network architecture (to improve data collection identified as a key factor to efficient adaptation measures) that presents enough flexibility to be implementable in a large variety of environments and geographies. This proposal initially offered to analyze the reasons and implications of what was coined as a Nepalese paradox (i.e. extensive but inadequate network of weather station). Yet, the output of the project is implementable in different countries especially in other LDCs also suffering from high vulnerability to floods.
A second benefit from our proposal should then be emphasized: there is very little cost associated with a potential generalization (that is, the replication in different environments) of our forecasting model. The relative simplicity of the architecture and the technology used, coupled with its portability, should ease the adoption of that technology in least developed areas where authorities are subject to serious financial constraints.
Finally, it could be envisaged that local populations become stakeholders of the diffusion of such a technology and that they become engaged in the monitoring of their own environment, which in turn could raise awareness for climate change and how it impacts their lives.
What are the proposal’s projected costs?
The proposal has been specifically designed to ensure optimal spending of the grant awarded. By keeping traveling costs to a minimum, the majority of the grant is to be dedicated to the purchase of the sensors necessary to the building of the desired flood forecasting system. In other words, most of the allowance will be dedicated to the final phase (the pilot phase) of the project as the work involved in the first three phases necessitates very little equipments and can be executed remotely (i.e. at our University). Furthermore, a successful pilot will allow for extensive replications of the forecasting tool at no cost given the aforementioned specificities of the technology utilized. In that regard, the proposal was specifically designed to guarantee the availability of fund over the entire project life.
We have not identified any negative side effect to our proposal. However, we do expect to face at least one of the following challenges:
- The quality of the assessment of the existing network of weather stations in phase 1 relies our ability to adequately train and test our algorithm and thus on the availability of data (especially data on past floods events). Given that some areas of the country are remote, it is possible that a significant enough number of past flood events were not recorded.
- Similarly, the satellite coverage of Nepal, on which relies the quality of the remote sensing data may be sparse. This could negatively affect our results from phases 2 and 3.
- Our desire to limit travel costs may complicate communication with our partners on site and we may encounter challenges associated with remotely training individuals to use the technology. This may result in delays in the last phase (pilot) of the project which is also the most critical one.
- As mentioned earlier, Nepal experiences some natural risk management governance issues which may result in extended delays in the last phase of the project, but more importantly, in a lack of interest of local authorities for our project and its outcomes. It will then be critical, for the success of the project, to create a strong relationship and an effective communication with our partners on site.
- one requirement for the efficient monitoring of rainfall and river flow is the ability of the sensor network to sustain harsh weather for extended periods of time. This may be a challenge in certain areas of the country. Due to its geographical location, Nepal is characterized very harsh weather conditions, especially during winter time, which could severely damage the equipment and compromise the project in its pilot phase.
The proposed project is expected to be completed within the following 2 year timeframe:
- 1 year will be allocated to the model building and models comparison (phases 1 through 3)
- 9 months will be dedicated to the design of alternative sensors networks, to their testing and to the selection of the most efficient one.
- 3 months will be necessary to ensure proper transfer and delivery of the technology to Nepal via our partners such as the World Bank.
Only after the completion of the project, the flood forecasting tool will be ready for use. However, the impact of the elaborated technology will heavily rely on its rate of adoption, first in Nepal, but potentially in other LDCs. Consequently, such an impact cannot be estimated before hand.
About the author(s)
Babatunde Idrisu is a consultant at the World Bank Group where he is conducting research under the Energy and Extractives Global Practice. He is a PhD candidate in Energy and Environmental Policy at the University of Delaware. His research focuses on renewable energy diffusion for sustainable development in the global south particularly in remote locations. He was a 2017 EDF Climate Corps fellow at the NYC Department of Education where he supported the deployment of rooftop solar PV in NYC schools. His background is the product of inter-disciplinary experiences spanning 10 years in finance, engineering, economics, and renewable energy policy teaching and research.
Nicolas Al Fahel is a PhD student in Energy and Environmental Policy at the University of Delaware (US) and a 2019 EDF Climate Corps Fellow at Dartmouth College (US) where he is in charge of developing an Electric Vehicle policy for the campus and the renewing of the university’s owned fleet. Nicolas has worked as a research fellow for over two years with projects focusing on offshore wind power and on developing renewable energy policy in developing countries. His work has involved data analysis and Machine Learning as tools to determine optimal policy implementation.
Prof. Gregory Dobler is an Assistant Professor at the University of Delaware’s Biden School for Public Policy and Administration, Department of Physics and Astronomy, and Data Science Institute. His expertise lies at the intersection of policy and technology with significant data analysis experience in computer vision, remote and in situ sensing, time series, machine learning, and statistics. He serves as the Director of the Urban Observatory facility, a multi-institutional observational platform for cities at the University of Delaware and New York University’s Center for Urban Science and Progress.
Basha, E. A., Ravela, S., & Rus, D. (2008, November). Model-based monitoring for early warning flood detection. In Proceedings of the 6th ACM conference on Embedded network sensor systems (pp. 295-308). ACM.
Dahal, R. K., & Hasegawa, S. (2008). Representative rainfall thresholds for landslides in the Nepal Himalaya. Geomorphology, 100(3-4), 429-443.
Devkota, R. P. (2014). Climate change: trends and people’s perception in Nepal. Journal of Environmental Protection, 5(04), 255.
Gaire, Surya, Rafael Castro Delgado, and Pedro Arcos González. "Disaster risk profile and existing legal framework of Nepal: floods and landslides." Risk management and healthcare policy 8 (2015): 139.
INFROM (2019). Index Risk management for Nepal. Retrieved from INFORM: http://www.inform-index.org/Countries/Country-profiles/iso3/NPL
Kargel, J. S., Leonard, G. J., Shugar, D. H., Haritashya, U. K., Bevington, A., Fielding, E. J., ... & Anderson, E. (2016). Geomorphic and geologic controls of geohazards induced by Nepal’s 2015 Gorkha earthquake. Science, 351(6269), aac8353.
Maharjan, S., Sigdel, E., Sthapit, B. and Regmi, B. (2011) Tharu Community’s Perception on Climate Changes and Their Adaptive Initiations to Withstand Its Impacts in Western Terai of Nepal. International NGO Journal,6, 35-42.
Malla, G. (2008). Climate change and its impact on Nepalese agriculture. Journal of agriculture and environment, 9, 62-71.
Maharjan, K. L., & Joshi, N. P. (2013). Effect of climate variables on yield of major food-crops in Nepal: A time-series analysis. In Climate Change, Agriculture and Rural Livelihoods in Developing Countries (pp. 127-137). Springer, Tokyo.
Talchabhadel, R., & Sharma, R. (2014). Real time data analysis of west Rapti River basin of Nepal. Journal of Geoscience and Environment Protection, 2(05), 1.
ADB. (2019). Poverty in Nepal. Retrieved from ADB: https://www.adb.org/countries/nepal/poverty
Gautam, O. P., Velleman, Y., Paudel, K. P., Curtis, V., & Dhimal, M. (2017, November). Water, sanitation, and hygiene interventions: an urgent requirement in post-flood Nepal. The Lancet Infectious Diseases, 17(11), 1118-1119.
The World Bank. (2019). The World Bank In Nepal overview. Retrieved from The World Bank: https://www.worldbank.org/en/country/nepal/overview
United Nations Office of the Resident Coordinator. (2017). Nepal: Flood 2017 Office of the Resident Coordinator Situation Report No. 2 (as of 16 August 2017). Retrieved from https://reliefweb.int/sites/reliefweb.int/files/resources/Nepal%20Flood%20Sitrep%2016%20August%202017.pdf