A Kalman Filter-Based Method to Generate Continuous Time Series of Medium-Resolution NDVI Images

Fernando Sedano,Pieter Kempeneers,George Hurtt

doi:10.3390/rs61212381

Abstract

A data assimilation method to produce complete temporal sequences of synthetic medium-resolution images is presented. The method implements a Kalman filter recursive algorithm that integrates medium and moderate resolution imagery. To demonstrate the approach, time series of 30-m spatial resolution NDVI images at 16-day time steps were generated using Landsat NDVI images and MODIS NDVI products at four sites with different ecosystems and land cover-land use dynamics. The results show that the time series of synthetic NDVI images captured seasonal land surface dynamics and maintained the spatial structure of the landscape at higher spatial resolution. The time series of synthetic medium-resolution NDVI images were validated within a Monte Carlo simulation framework. Normalized residuals decreased as the number of available observations increased, ranging from 0.2 to below 0.1. Residuals were also significantly lower for time series of synthetic NDVI images generated at combined recursion (smoothing) than individually at forward and backward recursions (filtering). Conversely, the uncertainties of the synthetic images also decreased when the number of available observations increased and combined recursions were implemented.

Highlights

In the last 30 years, Earth observation satellites have played a central role in monitoring, understanding and quantifying land cover-land use dynamics and environmental processes
We propose a data assimilation method that simulates time series of medium-resolution synthetic images built from existing medium-resolution imagery and time series of moderate resolution imagery
This paper proposes a data assimilation approach based on a Kalman filter algorithm [52] simulates time series of medium-resolution synthetic images assimilating information from medium-resolution imagery and time series of moderate resolution imagery

Summary

Introduction

In the last 30 years, Earth observation satellites have played a central role in monitoring, understanding and quantifying land cover-land use dynamics and environmental processes. The Landsat satellite series, starting in 1972, has accumulated the oldest temporal record of space-based Earth observations. The opening of the Landsat archive [22] creates new opportunities for temporal studies of land surfaces at higher spatial resolution. Taking advantage of these opportunities and in response to a need to expose variations of land surfaces at finer spatial detail, a number of studies have used temporal series of Landsat data to analyze vegetation trends and the dynamics of phenology [23,24], as well as forest disturbance and recovery patterns [25,26,27,28]. Zhu and Woodcock [29]

Methods

Results

Discussion

Conclusion