Characterising the multi-decadal evolution of highland ecosystems, Sibinacocha, Peru, using GoogleEarth Engine

The Sibinacocha catchment is located in the southern region of Peru, inside the Vilcanota Urubamba Basin (VUB) system, and provides a range of important ecosystem services that local people depend on in their daily lives. Mapping highland ecosystems such as these is challenging because of cloud cover, and thus large-scale mapping activities are frequently applied. For this reason, there is a lack of studies focused on annual-scale land cover changes that may reveal sudden changes, or expose the interaction of changes between ecosystems. In this study, we identify five different land covers comprising the Sibinacocha catchment, namely glaciers, water bodies, wetlands, pastures, and low-vegetation areas. The evolution of the land cover of these ecosystems is mapped using a Random Forest classification model, which is a supervised machine-learning algorithm developed in Google Earth Engine. We apply it to a 36 year-long stack of Landsat images (Landsat 5, 7, and 8) from 1984 to 2020, using five classification criteria such as different normalized indices and a slope discrimination criteria obtained from SRTM information. Overall results were validated using the Kappa coefficient (K; 0.97) and overall accuracy analysis (97%) both based on collected field data, highlighting a good performance of the Random Forest model at classifying highland ecosystems. The results of the land cover evolution from 1984 and 2020, show significant area changes mainly on glaciers (-35%), wetlands (-17%), and water bodies (+14%) with noticeable trends, and low changes in pastures (+2%) and low-vegetation areas (+8%). For the time period of analysis, we identify an increase less than +0.8ºC in the annual temperature and 20 mm in annual precipitation. Using simple linear regression and correlation analysis, the changes we observe can be explained by the ecosystem responding to a warming climate. As glaciers recede, they are replaced by water bodies and low-vegetation ecosystems, low-vegetation ecosystems have generally become wetter, and wetlands and pastures transition backward and forward depending on their management. With these results, it is possible to understand the ecosystem’s natural evolution, enhanced by external factors, and to observe that it is ultimately conditioned by accelerated glacier retreat in the catchment headwaters.