Expanding remote-sensing and machine-learning knowledge at the AGU Fall Meeting 2019

18 Feb 2020

Attending the American Geophysical Union (AGU) is a huge opportunity for a geospatial scientist. The annual AGU Fall Meeting is the largest international Earth and space science meeting where we can learn about new technology and build professional networks.

Having worked in Earth observation and geo-information science for the past eight years, I was quite excited to present my work – as part of a research team under the SERVIR Hindu Kush Himalaya (SERVIR-HKH) Initiative at ICIMOD – at the AGU Fall Meeting 2019 on 9–13 December in San Francisco, USA. Our work on analysing croplands annually using an automated methodology in Nepal was accepted for an oral presentation during the session titled “Use of Earth Observations in Mitigating Major Environmental Challenges I”.

The AGU Fall Meeting was a unique experience for me. On the first day, I realized that most of the guests at the hotel were there primarily for the conference. Everyone was excited and preparing for the five-day conference. The venue was bustling as participants rushed to attend the many sessions often running at the same time. It looked like no one wanted to miss out on the knowledge being shared. I was interested in attending several topics, particularly machine- learning capabilities and their applications in forest carbon estimation as well as land cover monitoring systems. Overall, the event gave me an opportunity to learn, update myself with the latest developments, and interact with international scientists. It motivated me to dive further into research so that I can contribute to the environment and society.

My session had a range of oral presentations on research efforts incorporating Earth observation information to generate tools, products, and services towards supporting environmental decision making by governmental institutions, non-governmental agencies, and the general public. Presenters shared their experiences from ongoing work in Afghanistan, the Amazon forest, Canada, Eastern and Southern Africa, the Mekong region, and Nepal, among others.

Sajana Maharjan at the AGU

I presented on progress and results from our research in Nepal, which aims to annually generate reliable data on croplands from 2000 to 2018 using remote-sensing and machine- learning techniques and analyse changes. We developed a cropland mapping system on the Google Earth Engine platform and used a random forest machine-learning algorithm to differentiate between crop and non-crop lands – a threshold level defined on local knowledge and collected samples. Our study results suggest that agricultural area in the Terai districts decreased between 2000 and 2018. Concurrently, agricultural area shrank in the middle mountain whereas the agricultural land cover was expanding during the same period in the Churia, High Mountains, and High Himal areas. This data provides valuable insights into physiographic changes in the country and could aid planning and policy decisions for food security, economic development, and social stability.

Apart from sessions at the AGU, I attended an event organized for the SERVIR network co- hosted by Mapbox and the University of San Francisco. At the event, I was able to meet colleagues from NASA, the University of Alabama, the University of San Francisco, and our counterparts from the different SERVIR hubs: SERVIR-Eastern and Southern Africa, SERVIR West Africa, SERVIR MEKONG, SERVIR Amazonia. I felt proud to be among wonderful, cooperative, dedicated colleagues working to improve the lives and livelihoods of people around the world.


Sajana Maharjan

GIS and Remote Sensing Analyst