A Recursive Update Model for Estimating High-Resolution LAI Based on the NARX Neural Network and MODIS Times Series

Abstract

Leaf area index (LAI) remote sensing data products with a high resolution (HR) and long time series are in demand in a wide variety of applications. Compared with long time series LAI products with 1 km resolution, LAI products with high spatial resolution are difficult to acquire because of the lack of remote sensing observations in long-term sequences and the lack of estimation methods applicable to highly variable land-cover types. To address these problems, we proposed a recursive update model to estimate 30 m resolution LAI based on the updated Nonlinear Auto-Regressive with Exogenous Inputs (NARX) neural network and MODIS time series. First, we used a variety of HR satellite remote sensing observations to produce HR datasets for recent years. Historical low spatial resolution MODIS products were employed as background information and used to calculate the initial parameters of the NARX neural network for each pixel. Subsequently, one year’s reflectance from the HR dataset was used as the new observation that was input into the NARX model to estimate the HR LAI of that year, and the background and HR data were then used for remodeling to update the NARX model parameters. This procedure was recursively repeated year by year until both MODIS background data and all HR data were involved in the modeling. Finally, we obtained an LAI time series with 30 m resolution. In the cropland study area in Hebei Province, China, the results were compared with LAI measurements from ground sites in 2013 and 2014. A high degree of similarity existed between the results for the two study years (RMSE2013=0.288 and RMSE2014=0.296). The HR LAI estimates showed favorable spatiotemporal continuity and were in good agreement with the multisample ground survey LAI measurements. The results indicated that for data with a rapid revisit cycle and high spatial resolution, the recursive update model based on the NARX neural network has excellent LAI estimation performance and fairly strong fault-tolerance capability.

Forest and Crop Leaf Area Index Estimation Using Remote Sensing: Research Trends and Future Directions

“…Verrelst et al [112] compared four machine learning regression algorithms (ANN, SVR, kernel ridge regression (KRR), and GPR) using Sentinel-2 and -3 data for estimating biophysical parameter retrieval and concluded that GPR had the best performance (relative RMSE ≈ 0.2) when compared to the other three methods. Wang et al [113] used a neural network approach to combine HJ-1 CCD, GF-1, Landsat TM, Landsat Enhanced Thematic Mapper Plus (ETM+), and MODIS data, which enabled them to fill in missing data in the Landsat images in order to generate continuous time series (8 days) crop LAI with 30 m resolution (RMSE = 0.30).…”

Evaluation of Prediction Performance of Vegetation Biomass Density for Two Different Case Study Areas in Turkey with Hybrid Wavelet and Artificial Neural Network Method

“…Shatnawi and Qdais applied NARX ANN model to simulate and estimate of LST values for the next 10 years using between 2000 and 2006 data (Shatnawi and Qdais 2019). The results of another study show that the NARX model is very successful and has very strong fault tolerance capability in LAI time series estimation based on MODIS data (Wang et al 2019). Although NARX ANN is a reliable and successful method for estimating nonlinear data, it is possible to increase the estimation success by developing hybrid methods (Alves and Lopes 2017).…”