Extraction Of Xinyang Tea Planting Area Based On Random Forest Algorithm And Its Temporal And Spatial Variation Analysis

Globally,tea,as a beverage,serves as one significant part of human daily life.Meanwhile,tea-related industries play an important role in national and regional economic development.Therefore,it is necessary to accurately grasp the relevant information such as tea planting areas and tea yield.Among them,obtaining the spatial distribution and areas of tea plantations in a timely and accurate manner,often appears as the prerequisite to understanding tea-related studies.Conventional methods rely heavily on manual field surveys or visual interpretation to pinpoint the spatial distribution of tea plantations,which is time-consuming,laborintensive,and often difficult to update effectively.With the development of remote sensing（RS）technology,the multi-spectral band information inherent in RS images contributes to distinguishing tea trees from other land covers,and at the same time,the unique characteristics of RS’s capabilities in the repeated,continuous,and large-scale observation,also help to quickly identify tea planting areas from local to regional scales,which is expected to accurately acquire crucial information on the spatial distribution patterns and planting areas of tea plantations.As a result,applying RS techniques to identify tea plantations using satellite images has become one important study in RS communities.With the emergence and rapid development of cloud-computing platforms such as Google Earth Engine（GEE）the traditional remote sensing processing method based on client-side mode is undergoing unprecedented changes.The GEE cloud-computing platform has massive storage capacity,high-efficiency computing power,and a wide variety of remote sensing image data,which facilitates researchers to conduct research in a more efficient manner that helps to diminish tedious preprocessing such as image download.Based on this,using the time series remote sensing image data on the GEE platform to accurately extract the tea planting area,and analyze its inter-annual dynamic change law,has become the main motivation and topic of this paper.In this paper,the multi-temporal Sentinel-2 and Landsat-5/7/8 imageries on GEE are used as input.Through adopting the random forest algorithm,this study proposes a novel method to extract tea planting areas,which is compared with other popular classification methods.In order to illustrate the effectiveness of the proposed method,this paper takes Shihe District,Xinyang City,Henan Province as the research area,and has carried out multi-year extraction of tea planting areas.The experimental results of this paper show that:（1）The adopted and developed random forest model outperforms other machine learning models such as Support Vector Machines and Decision Tree.Applying the modified RF classifier to Sentinel-2 time series images can achieve high-precision extraction of tea planting areas.（2）Using spatiotemporal analysis methods to the mapping results of extracted tea planting areas in Xinyang,it was found that during the 11 years from 2010 to 2020,the changes in tea planting in Shihe District,Xinyang City were relatively stable,mainly concentrated in the central mountainous area of the study area.（3）This paper conducts a correlation analysis on the driving factors of tea planting areas in the study area,and concludes that population is the most important driving factor,followed by slope and aspect factors among topographic factors.In addition,temperature,precipitation,soil,water system and elevation factors also have a certain influence on tea planting.