A virtual hands-on training program on developing geospatial maps for supporting insurance products using Google Earth Engine and semi-automatic techniques was conducted for participants in Pakistan as part of the project “Strengthening Post-COVID-19 Food Security and Locust Attacks”.
The nine participants were from the PARC Agrotech company (PATCO) technical team and crop reporting service teams from Punjab and Sindh in Pakistan. They were introduced to remote sensing and its applications in agriculture. Hands-on training using Google Earth Engine (GEE), Image Processing Software – ERDAS 2015 and various automatic classification techniques was provided along with several applications for using these modern tools.
With the demand for crop insurance increasing in developing countries, it has become vital for stakeholders/policy makers to gain access to spatial information on crop type, crop health and stress at field level.
Crop insurance is an area that extensively uses crop type maps at various levels. Crop type and crop health maps generated using remote sensing technology are used to optimize the sampling of crop cutting experiment (CCE) locations. CCEs are the basis for crop yield estimation to determine crop loss. Crop acreage estimation at village/sub-district level helps in making informed decisions on claims of failed sowing and helps prevent ‘false insuring’ of more land than what is planted under a particular crop. The satellite-derived crop area statistics provide the basis for assessing national statistics and augment the decision-making and planning process by providing accurate information of even inaccessible areas. In addition to national-level assessments, the crop area and extent maps obtained from this classification technique can be used for village-level crop assessment for micro-level crop management and advisory.
The processing of high-resolution satellite imagery with the help of remote sensing techniques gives crop information, but takes time in processing the imagery manually. In order to get information in real time, machine-learning algorithms such as Random Forest and Support Vector Machines together with cloud computing platforms such as GEE provide updated high-resolution satellite imagery from Landsat and Sentinel-1&2. Processing the imagery using Google Earth Engine saves a lot of time.
This training included both automatic and semi-automatic techniques to monitor various themes such as crop type, irrigated vs. rainfed, crop stress, crop intensity and so on. To validate the importance of remote sensing, Dera Ismail Khan district was selected for a case study. Participants did the processing and crop classification using Sentinel-2 time series data and spectral matching techniques for the crop year 2020-21(Figure 1).
ICRISAT collaborated with PATCO and Research and Development Foundation, an NGO based in Hyderabad, Pakistan, for the training held from 26-29 July. The training was conducted by ICRISAT’s Geospatial Sciences and Big Data staff – Dr Murali Krishna Gumma, Cluster Leader, Mr Ismail Mohammed, Senior Officer – Data science and Mr Pranay Panjala, Scientific Officer-Remote Sensing.
Acknowledgments: We would like thank Ms Noriko Sato, Natural Resources Specialist ADB; Dr Takashi Yamano, Senior Economist, ADB for their support in organizing this training course.