A predictive framework formalising landscape-scale ecological memory as a measurable network property — applied to Zambia's miombo woodland and Zambezi floodplains.
Pull 22 years of vegetation, soil, and disturbance data for three Zambian ecosystems. All data sources are free and public.
Time-lagged cross-correlations between proxy signals form a directed multiplex network. The EMN memory metric is the weighted eigenvector centrality of disturbance-related nodes across all lag steps.
An LSTM trained on EMN memory metrics + disturbance history, compared against a climate-only baseline. If EMN-LSTM wins, the theory holds.
EMN-LSTM vs climate-only baseline across all sites and forecast horizons. A positive Δ R² means ecological memory adds predictive power beyond climate signals.
This software implements the Ecological Memory Networks (EMN) theory — a framework formalising how ecosystems encode disturbance history across three measurable proxy signals: vegetation spectral indices, soil organic carbon, and atmospheric carbon flux anomalies.
Zambia sits at the intersection of miombo woodland, Zambezi floodplains, and agricultural frontiers — one of the most understudied ecological transition zones on Earth. The carbon dynamics here are actively unknown. This project fills that gap.