Abstract:Expanding global and regional markets are driving the conversion of traditional subsistence agricultural and occupied non-agricultural lands to commercial-agricultural purposes. In many parts of mainland Southeast Asia rubber plantations are expanding rapidly into areas where the crop was not historically found. Over the last several decades more than one million hectares of land have been converted to rubber trees in areas of China, Laos, Thailand, Vietnam, Cambodia and Myanmar, where rubber trees were not traditionally grown. This expansion of rubber plantations has replaced ecologically important secondary forests and traditionally managed swidden fields and influenced local energy, water and carbon fluxes. Accurate and up-to-date monitoring and mapping of rubber tree growth is critical to understanding the implications of this changing ecosystem. Discriminating rubber trees from second-growth forests and fallow land has proven challenging. Previous experiments using machine-learning approaches with hard classifications on remotely sensed data, when faced with the realities of a heterogeneous plant-life mixture and high intra-class variance, have tended to overestimate the areas of rubber tree growth. Our current research sought to: 1) to investigate the potential of using a Mahalanobis typicality model to deal with mixed pixels; and 2) to explore the potential for combining MOderate Resolution Imaging Spectroradiometer (MODIS) imagery with sub-national statistical data on rubber tree areas to map the distribution of rubber tree growth across this mainland Southeast Asia landscape. Our study used time-series MODIS Terra 16-day composite 250 m Normalized Difference Vegetation Index (NDVI) products (MOD13Q1) acquired between March 2009 and May 2010. We used the Mahalanobis typicality method to identify pixels where rubber tree growth had the highest probability of occurring and sub-national statistical data on rubber tree growth to quantify the number of pixels of rubber tree growth mapped per administrative unit. We used Relative Operating Characteristic (ROC) and error matrix analysis, respectively, to assess the viability of Mahalanobis typicalities and to validate classification accuracy. High ROC values, over 0.8, were achieved with the Mahalanobis typicality images of both mature and young rubber trees. The proposed method greatly reduced the commission errors for the two types of rubber tree growth to 1.9% and 2.8%, respectively (corresponding to user’s accuracies of 98.1% and 97.2%, respectively). Results indicate that integrating Mahalanobis typicalities with MODIS time-series NDVI data and sub-national statistics can successfully overcome the earlier overestimation problem.

Automatic Rubber Tree Classification

Rubber Tree Recognition Based on UAV RGB Multi-Angle Imagery and Deep Learning

Individual Rubber Tree Segmentation Based on Ground-Based LiDAR Data and Faster R-CNN of Deep Learning

Prediction of South American Leaf Blight and Disease-Induced Photosynthetic Changes in Rubber Tree, Using Machine Learning Techniques on Leaf Hyperspectral Reflectance

Mapping rubber tree growth in mainland Southeast Asia using time-series MODIS 250 m NDVI and statistical data

Enhanced Rubber Yield Prediction in High-Density Plantation Areas Using a GIS and Machine Learning-Based Forest Classification and Regression Model

Automated plant identification using artificial neural network and support vector machine

Rubber Tree Distribution Mapping in Northeast Thailand

Fractional order modeling and recognition of nitrogen content level of rubber tree foliage

Mapping of Rubber Forest Growth Models Based on Point Cloud Data

Automatic Plant Identification Using Transfer Learning

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

Tapped area detection and new tapping line location for natural rubber trees based on improved mask region convolutional neural network

Handling the Challenges of Small-Scale Labeled Data and Class Imbalances in Classifying the N and K Statuses of Rubber Leaves Using Hyperspectroscopy Techniques

The development of an automatic rubber seed sowing system with machine vision assistance

Identification of Rubber Plantations in Southwestern China Based on Multi-Source Remote Sensing Data and Phenology Windows

Extracting distribution and expansion of rubber plantations from Landsat imagery using the C5.0 decision tree method

3D visualization technology for rubber tree forests based on a terrestrial photogrammetry system

Automated feature-specific tree species identification from natural images using deep semi-supervised learning

A Rubber-Tapping Robot Forest Navigation and Information Collection System Based on 2D LiDAR and a Gyroscope