Multitemporal multispectral drone imagery for water hyacinth mapping in Patancheru lake, Hyderabad, India

Abstract

Pontederia crassipes, commonly known as water hyacinth (WH), is a highly invasive aquatic weed and its infestation has caused significant adverse ecological and economic impact across the world. Remedial actions include manual monitoring and removal which are often time consuming and expensive. This paper proposes the use of multi-temporal multi-spectral drone imagery for WH mapping and monitoring in Patancheru Lake, located in the outskirts of the Indian city Hyderabad. The data collection was done in two steps: 1) multi-spectral drone based data acquisition by aerial survey and 2) ground level optical image capturing through an Android mobile application named MPRO. Data was collected in regular interval starting from January 2021. Spectral bands were used to produce the WH detection and mapping. We compare spectral signature of clean and infested water for five different sites inside the lake. Multitemporal water quality samples of these sites were also collected together with drone data to analyse the effect of WH infestation on those parameters. The multi-spectral data was processed using an unsupervised machine learning classifier named expectation maximisation (EM) clustering to create a segmentation map indicating WH, water and other regions.