Grasslands are an important ecosystem globally for humans and wildlife populations, representing approximately a fifth of the earth’s land. However, monitoring grasslands across large spatial scales presents challenges to researchers and managers. Remote sensing offers a time- and cost-effective solution to large-scale grassland monitoring. Combining optical and radar satellite imagery offers a promising avenue to classify land cover because together they can differentiate variation in moisture, structure, and reflectance among land cover types. Here we used a combined approach using multi-date radar (PALSAR-2 and Sentinel-1) and optical (Sentinel-2) imagery with field data and visual interpretation of aerial imagery to create a land cover map of the Masai Mara National Reserve, Kenya with differentiated grass height categories using machine learning (Random Forests). We then compared the relative proportions of barren land and grasslands of different height classes between two regions of the Mara Reserve – one area where domestic livestock are restricted (the Triangle), and one where management was passive and the region has been heavily grazed by domestic livestock (the Talek region).
Results/Conclusions
These methods produced a highly accurate land cover map (86% overall accuracy) of the Masai Mara National Reserve that differentiated grassland cover into three height classes (i.e., short, medium, and tall grass user’s accuracies were 85%, 75%, and 95%, respectively). When comparing land cover among management entities, the Talek region of the Mara had higher proportions of short grass, medium grass, and barren land compared to the Triangle. Among the open grassland cover types, the Talek region was comprised of 14% short grass, 32% medium grass, 54% long grass, and 3% barren ground, compared to the Triangle which consisted of 7% short grass, 19% medium grass, 73% long grass, and 1% barren ground. Additionally, most barren ground and short grasslands in the Talek region were near the Reserve boundary, reflecting the increased grazing pressure on areas near the park border. These results demonstrate the utility of a combined radar and optical imagery approach to grassland monitoring and may provide an inexpensive and time efficient tool for monitoring grasslands over a variety of spatial and temporal scales.