Rubber Identification Based on Blended High Spatio-Temporal Resolution Optical Remote Sensing Data: A Case Study in Xishuangbanna

Abstract

As an important economic resource, rubber has rapidly grown in Xishuangbanna of Yunnan Province, China, since the 1990s. Tropical rainforests have been replaced by extensive rubber plantations, which has resulted in ecological problems such as the loss of biodiversity and local water shortages. It is vitally important to accurately map the rubber plantations in this region. Although several rubber mapping methods have been proposed, few studies have investigated methods based on optical remote sensing time series data with high spatio-temporal resolution due to the cloudy and foggy weather conditions in this area. This study presented a rubber plantation identification method that used spatio-temporal optical remote sensing data fusion technology to obtain vegetation index data at high spatio-temporal resolution within the optical remote sensing window in Xishuangbanna. The analysis of the proposed method shows that (1) fused optical remote sensing data with high spatio-temporal resolution could map the rubber distribution with high accuracy (overall accuracy of up to 89.51% and kappa of 0.86). (2) Fused indices have high R2 (R2 greater than 0.8, where R is the correlation coefficient) with the indices that were derived from the Landsat observed data, which indicates that fusion results are dependable. However, the fusion accuracy is affected by terrain factors including elevation, slope, and slope aspects. These factors have obvious negative effects on the fusion accuracy of high spatio-temporal resolution optical remote sensing data: the highest fusion accuracy occurred in areas with elevations between 1201 and 1400 m.a.s.l., and the lowest accuracy occurred in areas with elevations less than 600 m.a.s.l. For the 5 fused time series indices (normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), normalized difference moisture index (NDMI), normalized burn ratio (NBR), and tasseled cap angle (TCA)), the fusion accuracy decreased with increasing slope, and increasing slope had the least impact on the EVI, but the greatest negative impact on the NDVI; the slope aspect had a limited influence on the fusion accuracies of the 5 time series indices, but fusion accuracy was lowest on the northwest slope. (3) EVI had the highest accuracy of rubber plantation classification among the 5 time series indices, and the overall classification accuracies of the time series EVI for the four different years (2000, 2005, 2010, and 2015) reached 87.20% (kappa 0.82), 86.91% (kappa 0.81), 88.85% (kappa 0.84), and 89.51% (kappa 0.86), respectively. The results indicate that the method is a promising approach for rubber plantation mapping and the detection of changes in rubber plantations in this tropical area.

Assessment of the Impact of Rubber Plantation Expansion on Regional Carbon Storage Based on Time Series Remote Sensing and the InVEST Model

“…However, most rubber forests are planted at the expense of the primary tropical rainforests or secondary forests, which inevitably leads to a certain loss of carbon sinks [7,16,17]. In addition, the deforestation of the tropical rainforest caused by rubber forest plantations will inevitably lead to the continuous reduction in the tropical rainforest-based biological habitat, the gradual reduction of soil and water conservation capacity, regional environmental degradation, and serious damage to biodiversity and the ecological environment [18][19][20][21]. Therefore, it is of great significance to understand the quantitative impact of rubber forests on carbon storage for the rational development of rubber plantations, the protection of forest ecosystems, the maintenance of the carbon sink balance, and the stability of climate change.…”

Application of Optical Remote Sensing in Rubber Plantations: A Systematic Review

“…Remote sensing technology provides spatial and temporal information over large extents of an observed area [30][31][32][33][34][35]. It has played a vital role in mapping rubber trees at local and regional scales [18,[36][37][38] and has facilitated understanding of changes in spatial patterns of rubber plantations over time [39].…”

Integrating Phenological and Geographical Information with Artificial Intelligence Algorithm to Map Rubber Plantations in Xishuangbanna

“…To date, multiple classification approaches have been used in mapping rubber plantations by combining phenological information, among which the maximum likelihood (ML) classification method, QUEST decision tree (QDT) classification method [12,18,25], object-based classification method [15,20], the random forest method (RF) [17,26], and other machine learning-based classification methods [26] have been widely used. However, limited information exists on how these classifiers perform when phenological information is incorporated.…”