In recent years, remote sensing technology has developed by leaps and bounds. Multiple remote sensing platforms such as aerospace, aviation, and near space have continued to increase. The spatial, temporal, and spectral resolution of data has continued to increase. The amount of data has soared. Remote sensing data has become more and more characteristic of big data. . The emergence of remote sensing big data provides unprecedented opportunities for related research, but how to deal with these data also poses great challenges. Traditional workstations and servers have been unable to meet the needs of large-scale, multi-scale massive remote sensing data processing.
Represented by Earth Engine (GEE) and PIE-Engine, online visualization computing and analysis cloud platforms for global-scale earth science data (especially satellite remote sensing data) are becoming more and more widely used. The GEE platform stores and synchronizes satellite images such as MODIS, Landsat and Sentinel, climate and weather, geophysics and other data sets that are currently commonly used in the field of remote sensing exceeding 80PB, and relies on millions of super servers around the world to provide sufficient computing power for these The data is processed. Compared with traditional remote sensing image processing tools such as ENVI, GEE has incomparable advantages in processing massive remote sensing data. On the one hand, it provides abundant computing resources, and on the other hand, its huge cloud storage saves a lot of data downloads for researchers. And the time of preprocessing, the calculation and analysis visualization of remote sensing data represent the most advanced level in this field in the world, and it is a revolution in the field of remote sensing.
In the early stage, we used Javascript and Python as programming languages, and successfully held several basic training courses on GEE remote sensing big data analysis and processing, which received extensive participation and support from students from various industries. In response to the requirements of the majority of scientific researchers, this course will focus on forestry, one of the most popular areas of remote sensing applications at present, and focus on combining typical application cases to comprehensively demonstrate the use skills and powerful functions of the GEE cloud platform, so as to enhance the participants' ability to solve practical problems . This course will be based on the JavaScript version of GEE. First, the basic knowledge of GEE will be introduced, followed by a series of lectures on key knowledge combined with micro-cases, and finally a comprehensive explanation combined with typical forestry application cases. In order to improve the teaching quality, this course will combine the most advanced AI natural language models such as ChatGPT and Wenxin Yiyan to assist teaching, assist students to answer doubts, provide targeted suggestions and guidance, not only allow students to have a deeper understanding of the course content, but also provide Future self-service learning provides an efficient and personalized learning experience.
1
Basic Practice
1.1
Platform and basic development platform
Introduction to GEE platform and typical application cases;
Introduction to GEE development environment and common data resources;
ChatGPT, Wenxin Yiyan and other GPT model introduction, account application and forestry remote sensing application
Introduction to JavaScript basics;
GEE remote sensing cloud important concepts and typical data analysis process;
GEE basic object introduction, vector and raster object visualization, attribute viewing, API query, basic debugging and other platforms to get started.
1.2
GEE basic knowledge interacts with AI models such as ChatGPT
Basic image operations and operations: mathematical operations, relational/conditional/Boolean operations, morphological filtering, texture feature extraction; image masking, cropping and mosaic, etc.;
Element basic operations and operations: geometric buffer, intersection, union, difference operations, etc.;
Collection object operations: loop iteration (map/iterate), merge Merge, union (Join);
Data integration Reduce: including image and image set integration, image synthesis, image area statistics and domain statistics, group integration and area neighborhood statistics, image set linear regression analysis, etc.;
Machine learning algorithms: including supervised (random forest, CART, SVM, decision tree, etc.) and unsupervised (wekaKMeans, wekaLVQ, etc.) classification algorithms, classification accuracy evaluation, etc.;
Data asset management: including local vector and raster data upload, cloud vector and raster data download, statistical result data export, etc.;
Drawing visualization: including bar charts, histograms, scatter plots, time series and other graphic drawing.
GPT model interaction: Combine the above basic knowledge points with AI tools such as ChatGPT and Wenxin Yiyan to conduct interactive demonstrations, including skills such as auxiliary question answering, code generation and correction.
2
Mini-case lectures on important knowledge points and GPT model interactive demonstration
1) Landsat, Sentinel-2 images automatically remove clouds and shadows in batches
2) Combine Landsat and Sentinel-2 to calculate vegetation index and annual synthesis in batches
3) Statistical analysis of the number of available images and the number of cloud-free observations in the study area
4) Synthesis of annual NDVI vegetation numbers in China and search for the greenest DOY time of the year
5) Moving window smoothing of time series optical image data
6) Stratified random sampling and sample export, sample local evaluation and data upload to the cloud
7) Trend analysis of rainfall in China in recent 40 years
8) Statistical analysis of annual forest loss in a certain area (based on Hansen forest products)
3
Typical case comprehensive drill
3.1
Case 1: Forest identification with joint multi-source remote sensing data
Introduce in detail the complete process of combining Landsat time series optical images and PALSAR-2 radar data, as well as the decision tree algorithm to achieve remote sensing classification of typical land types such as forests. Topics involved spatiotemporal filtering of image data, batch cloud masking and vegetation index calculation for optical images; stratified random sampling and sample export, local quality control and cloud upload, random segmentation of samples, separability analysis, classification algorithm construction and application, Classification post-processing and accuracy evaluation, thematic map drawing, etc.
3.2
Case 2: Long-term forest state monitoring
Use long-term MODIS or Landsat image data to conduct long-term monitoring of forest status and analyze forest vegetation greening or browning. Topics include time series image preprocessing, image set connection, image synthesis, non-parametric detection of changing trends, significance testing, quantification and grading of changing trends, spatial statistics and result visualization, and thematic map drawing, etc.
3.3
Case 3: Deforestation and Degradation Monitoring
Combining Landsat series images, spectral separation model and NDFI normalized difference score index to realize forest deforestation and degradation monitoring. Topics include image preprocessing, mixed pixel decomposition, NDFI index calculation, function encapsulation, change detection and intensity classification, result visualization, thematic map drawing, etc.
3.4
Case 4: Forest Fire Monitoring
This paper introduces in detail the use of Landsat and Sentinel-2 time series optical remote sensing images to monitor forest fire loss and achieve fire intensity classification. Topics include image filtering, Landsat and Sentinel-2 optical image cloud removal and other preprocessing, vegetation index calculation, image synthesis, fire area identification and disaster intensity classification, statistical analysis and visualization of results, etc.
3.5
Case 5: Long-term Scale Forest Disturbance Monitoring
Combining 30 years of Landsat and other optical images with the classic LandTrendr algorithm to monitor forest disturbances. Topics include preprocessing of long-term remote sensing images, batch calculation of vegetation indices, annual image synthesis, concept and usage of array images, LandTrendr algorithm principle and parameter settings, interpretation of forest disturbance results and spatial statistical analysis, visualization and thematic map drawing, etc.
3.6
Case 6: Key Physiological Parameters of Forests (Tree Height, Biomass/Carbon Stock)
Inversion combines GEDI lidar, Landsat/Sentinel-2 multispectral optical images, Sentinel-1/PALSAR-2 radar images, etc., and machine learning algorithms to retrieve key physical parameters of forests, such as tree height, biomass/carbon storage. Topics include introduction to GEDI lidar data, common optical and radar data processing, application of machine learning algorithms, inversion accuracy evaluation, variable importance analysis, and result visualization.
Original link: GEE remote sensing cloud big data forestry application typical case practice and GPT model application