*=Equal contributors
There is a correlation between temporal patterns of geospatial activity and land use type. A novel self-supervised approach is proposed for activity time-series-based landscape exploration, where the time-series signal is converted to the frequency domain and compressed by a nested self-encoder that preserves the cyclical temporal patterns observed in the time series. Inputs are input to the binary classification of the partition neural network. Experiments show that temporary embeddings are effective in classifying residential and commercial areas.