Determine the strength of the relationship between the Normalized Difference Vegetation Index and several climatological variables using a Random Forest regression algorithm. This analysis was conducted for three different climate conditions in Kenya during the 2018 short rainy season.
Report, Study Areas Overview Map, Regression scatterplots and diagnostic reports comparing measured vs. predicted NDVI values for each study area, dataset showcase and discussion, Predicted vs. Measured NDVI comparison map.
Machine Learning (Random Forest Regression), Remote Sensing data acquisition and preprocessing (removing cloud cover, merging/clipping datasets), Imagery analysis (NDVI calculation), Statistical Analysis, Cartography, GIS (spatial joins, clipping, data model conversion, etc.), research, ArcGIS Pro
Used evapotranspiration (MODIS), precipitation (CHIRPS), soil moisture (SMAP) data and calculated NDVI (derived from multispectral Landsat 8 imagery) as inputs to the RF regression algorithm to predict NDVI values in three different climate regions in Kenya: highly productive agricultural zones (~50% agricultural area), arid/semi-arid lands (ASALs, ~30%), and arid land (~0.05%). Contrary to findings in similar research predictions were the strongest in the highly productive zone (R2 = 0.7) and were progressively weaker for the less productive study areas. This may have been do with increasing clustering of NDVI value distribution across study areas, potentially making it harder for the RF regression model to differentiate between subtle changes in measured NDVI.
Dataset Display and Description (pdf)
Report (pdf)
NDVI Prediction vs. Measured (pdf)
Study Areas Overview (pdf)
