Reconstruction of a Monthly 1 km NDVI Time Series Product in China Using Random Forest Methodology

Abstract

The normalized difference vegetation index (NDVI) is one of the most common metrics used to describe vegetation dynamics. Unfortunately, low-quality pixels resulting from contamination (by features including clouds, snow, aerosols, and mixed factors) have impeded NDVI products’ widespread application. Researchers have thought of several ways to improve NDVI quality when contamination occurs. However, most of these algorithms are based on the noise-negative deviation principle, which aligns low-value NDVI products to an upper line but ignores cases where absolute surface values are low. Consequently, to fill in these research gaps, in this article, we use the random forest model to produce a set of high-quality NDVI products to represent actual surface characteristics more accurately and naturally. Climate and geographical products are used as model inputs to describe environmental factors. They represent the random forest (RF) model that establishes relationships between MODIS NDVI products and meteorological products in high-quality areas. In addition, auxiliary data and empirical knowledge are employed to meet filling requirements. Notably, the random forest (RF) algorithm exhibits a mean absolute error (MAE) of 0.024 and a root mean squared error (RMSE) of 0.034, in addition to a coefficient of determination (R2) value of 0.974. Furthermore, the MAE and RMSE of the RF-based method decreased by 0.014 and 0.019, respectively, when compared to those of the STSG (spatial–temporal Savitzky–Golay) plan and by 0.013 and 0.015, respectively, when compared to the LSTM (long short-term memory) method. R2 increased by 0.039 and 0.027, respectively, compared to the STSG and LSTM methods. We introduced a novel series of NDVI products that demonstrated consistent spatial and temporal connectivity. The novel product exhibits enhanced adaptability to intricate environmental conditions and promises the potential for utilization in investigating vegetation dynamics within the Chinese region.

Development of long‐term spatiotemporal continuous NDVI products for alpine grassland from 1982 to 2020 in the Qinghai–Tibet Plateau, China

“…Alpine meadow types are densely distributed in the region, NDVI is easily saturated during the growing season (Wunderle et al, 2004), is insensitive to changes in high biomass conditions (Gitelson, 2004;Sakamoto et al, 2010;Zeng et al, 2016), and NDVI no longer continues to increase with vegetation growth, making it difficult to distinguish seasonal variations in vegetation greenness (May et al, 2018). Furthermore, the spatial distribution of NDVI in alpine meadow types is subject to a combination of climatic factors and human activities, and with the instability of global climate change and the increasing intensity of human activities, this has led to highly abrupt spatial changes in grasslands and significant changes in NDVI (Sun, Gong, et al, 2023;Sun, Li, et al, 2023), which may also contribute to the high bias in the downscaled product. On the contrary, the Enhanced Vegetation Index (EVI), also used to monitor vegetation growth, performs well in vegetation types where NDVI saturates, and can clearly reflect the seasonal characteristics of vegetation growth (Bai, 2021;Lin et al, 2008;Ma, Xie, et al, 2022).…”

“…With the continuous evolution of remote sensing image processing technologies, machine learning algorithms such as Random Forest (RF), Support Vector Machine (SVM), Gaussian Process Regression, Artificial Neural Networks, and so forth have shown excellent performance in scale transformation and spatiotemporal modeling of remote sensing data (Ali et al, 2021;Huang et al, 2023;Karbalaye Ghorbanpour et al, 2021;Liu et al, 2020;Sdraka et al, 2022;Yan et al, 2021). Among these, RF performs better in reconstructing NDVI time-series products than the Long Short-Term Memory Artificial Neural Networks (Sun, Gong, et al, 2023;Sun, Li, et al, 2023). Based on the RF model, the fusion of AMSR-E and MODIS data can effectively avoid excessive smoothing, providing land surface temperature (LST) estimates that better match real-world conditions (Zhang, He, et al, 2022); when downscaling evapotranspiration using three machine learning methods (RF, SVM, Cubist), the RF model produces the smallest error, demonstrating its potential in scale transformation (Ke et al, 2016).…”

Remote Sensing Monitoring of Drought in Southwest China Using Random Forest and eXtreme Gradient Boosting Methods

“…The experimental results indicated that the algorithm had the capacity to enhance the estimation of MODIS LST products in terms of accuracy and data availability. Sun et al (2023) [70] used the RF model to produce a set of high-quality NDVI products to represent actual surface characteristics more accurately and naturally. Notably, the RF algorithm exhibited a MAE of 0.024 and a RMSE of 0.034, besides a R 2 value of 0.974.…”