University of Twente Student Theses

Login

Estimating micronutrient concentrations in maize grains with Sentinel-1 and -2 images

Gohil, Jaykumar Harishbhai (2023) Estimating micronutrient concentrations in maize grains with Sentinel-1 and -2 images.

[img] PDF
2MB
Abstract:Hunger remains a major problem for low and middle-income countries. Moreover, Micronutrient deficiencies(MNDs) or hidden hunger are fatal and prevalent in these regions. Billions, including children, suffer from MNDs. In trace amounts, micronutrients like Selenium, Calcium, Iron, and others are found in the human body. Measurement of these micronutrients is vital, but the current methods available for testing and measuring are time-consuming and expensive. Hence, a method tackling both setbacks from prior techniques is the need of the hour. Several scientists have taken the help of field and crop samples to spatial map these micronutrition concentrations using spatial statistics. This study aimed to develop a machine learning method comprising remote sensor data from Sentinel 1 and Sentinel 2 and other ancillary information like topographic, climatic, and soil characteristics data for micronutrient concentration in Calcium, Iron, Magnesium, and Zinc. To do this, we take the help of GeoNutrition Surveys as our reference data and use the Random Forest model for building the model. The study area for the research is Malawi. We use the GEEMAP library and GEE Python API to fetch all the datasets but the Sentinel 2 L1C, which we used Copernicus hub. L2A Sen2Cor algorithm was used for the atmospheric correction. After filtering many data points with inconsistent data or NaN values. After several combinations of the models, R2 accuracy for each model was Calcium(0.25), Iron(0.25), Magnesium(0.29), and Zinc(0.23). Finally, partial dependence plots indicated that topographic and climatic features have the highest correlation with micronutrient concentrations. SWIR band depicted most dependency among the spectral features. Cloud cover and the consequences of these missing data were major limitations. From these findings, this study is a foundation for using spectral and polarimetric features with machine learning techniques for micronutrient concentrations in tropical settings. Moreover, further development in terms of improving models via deep learning could be done. Using UAVs for acquiring spectral features is an experiment for future use.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:https://purl.utwente.nl/essays/97338
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page