Analyzing Poverty Dynamics through Time Series: A Wavelet Transform Approach

Sammanfattning

Reducing global poverty is a central goal of the Sustainable Development Goals(SDGs), requiring timely, high-resolution data to monitor socioeconomic changesespeciallyin low- and middle-income countries where traditional survey data is sparse.Recent advances in machine learning (ML) and Earth observation (EO) data haveenabled new approaches to poverty estimation. However, many existing models relyon aggregated annual or multi-year imagery, overlooking intra-annual variation thatis particularly relevant in agriculturally driven economies. This study explores thepotential of integrating intra-annual vegetation dynamics captured through NormalizedDifference Vegetation Index (NDVI) with annual multi-spectral data to enhancepoverty prediction. An unsupervised approach using wavelet transforms is proposedto summarize temporal NDVI signals, allowing models to retain essential seasonalpatterns while reducing dimensionality and noise. Experiments were conducted ina simulated environment using nighttime light intensity as a proxy for wealth, andfurther evaluated against the Demographic and Health Survey (DHS) dataset acrossAfrica. The results show that incorporating selected wavelet-derived NDVI featuressignificantly improves model accuracy over baseline methods, even in the presenceof missing data. These findings highlight the critical role of intra-annual temporalinformation and demonstrate the value of wavelet-based summarization for scalable,robust poverty estimation using satellite imagery