Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples...
Transcript of Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples...
![Page 1: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/1.jpg)
Cropland mapping using big earth data and crowd-source
samples
Miao Zhang
Aerospace Information Research Institute,
Chinese Academy of Sciences
November 2019, Tashkent, UZB
![Page 2: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/2.jpg)
Outline
• Introduction
• Study area and data
• Method
• Result and discussion
• Conclusion and future plans
![Page 3: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/3.jpg)
Source: World Food Programme
![Page 4: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/4.jpg)
What about the future?
![Page 5: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/5.jpg)
Introduction
Food security is a serious issue for Africa and might be more problematic;
Understanding the extent of cropland and the changes are the basis to fight against hunger and make agriculture plans;
There are several global/regional land cover / cropland extent datasets;
But with some uncertainty and great discrepancies exist among these products
![Page 6: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/6.jpg)
Large discrepancies between four different datasets
62
8392 96
93 2 1
No agreement Partialagreement
Highagreement
Full agreement
Accuracy% Com. ERR
Mohsen Nabil, et al., draft manuscript, 2018
![Page 7: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/7.jpg)
Outline
• Introduction
• Study area and data
• Method
• Result and discussion
• Conclusion and future plans
![Page 8: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/8.jpg)
Study area
• Zambezi River Basin• Rain-fed
• Single cropping
• Low productivity
• Vulnerable to Agro-climatic conditions
![Page 9: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/9.jpg)
Data
SATELLITE DATA CROWDSOURCE DATA IN-SITUMEASUREMENTS
Sentinel-2 top of canopy (TOC) reflectance
Data from GEOWIKI, NASA, and local experts
Joint field data collection in 2016-2018
![Page 10: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/10.jpg)
Sentinel-2 Source: NASA
![Page 11: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/11.jpg)
Samples includes GEOWIKI points, joint field surveys,
etc
Data collected using GVG
![Page 12: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/12.jpg)
Outline
• Introduction
• Study area and data
• Method
• Result and discussion
• Conclusion and future plans
![Page 13: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/13.jpg)
Methods
Data composite Training the classifier
Mapping Validation
![Page 14: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/14.jpg)
Flowchart of procedures
Sentinel 2 image
collection
Multi source
samples
Field
measurements
![Page 15: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/15.jpg)
Data composite via GEE
• Composite for rainy season / dry season
• Quality Mosaic based on Vis (qualityMosaic)
• Seasonal median composite (median)
• Composite by different percentile (ee.Reducer.percentile)
Dry season 2015 May to October
Rainy season 2015 October to May 2016
![Page 16: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/16.jpg)
Classifiers and training
• Random forest
• Support vector machine
• First applied for six major classes
• Results were grouped into two classes, cropland & non-cropland
![Page 17: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/17.jpg)
Validation
Kappa Coefficient
Confusion Matrix
User accuracy
Producer accuracy
Overall accuracy
![Page 18: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/18.jpg)
Outline
• Introduction
• Study area and data
• Method
• Result and discussion
• Conclusion and future plans
![Page 19: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/19.jpg)
Data composite
Median Max reflectance Quality Mosaic based on NDVI
Inputs for classification: MaxNDVI, MedianNDVI, SRmax, Srmedian, Viredge, P5, P25, P50, P75, P95
![Page 20: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/20.jpg)
Mapping on Google Earth
EngineWithout downloading raw data
![Page 21: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/21.jpg)
Samples Random forest SVM
T V T V
1 0.82 0.65 0.68 0.64
2 0.83 0.54 0.67 0.65
3 0.83 0.65 0.76 0.74
4 0.84 0.69 0.78 0.76
5 0.82 0.56 0.80 0.69
Average 0.83 0.62 0.74 0.70
Accuracy (5 repeat, 70% randomly as training and rest as validation)
Samples Random forest SVM
T V T V
1 0.77 0.67 0.76 0.76
2 0.80 0.70 0.76 0.86
3 0.82 0.92 0.93 0.93
4 0.78 0.78 0.72 0.82
5 0.76 0.66 0.94 0.84
Average 0.79 0.75 0.82 0.84
2015-2016 2016-2017
SVM out performs that of Random forest, but more time consuming
![Page 22: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/22.jpg)
Cropland maps
Cropland 2015-2016 Cropland 2016-2017
![Page 23: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/23.jpg)
Cultivated cropland area expanded by 27%
![Page 24: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/24.jpg)
Inter-annual cultivated cropland
2015-2016 cropland
Expanded cropland in 2016-2017
Uncultivated in 2016-2017
![Page 25: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/25.jpg)
Limitations
Pixel based methods result in ‘salt and pepper’ effect
Will try to include object-based classification methods
![Page 26: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/26.jpg)
Outline
• Introduction
• Study area and data
• Method
• Result and discussion
• Conclusion and future plans
![Page 27: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/27.jpg)
Conclusion and future plans
Taking advantage of GEE cloud platform, cropland mapping over large area is possible and promising if training data are sufficient
Data composite based on data over a single year still have gaps during rainy season, Two or three years data will be better
Training samples have significant impacts on the classification and accuracy; More samples will be derived from outdated datasets for early years
Expanded to early years using Landsat series to detect the inter-annual changes / hotspots, and driving forces will be analyzed
![Page 28: Cropland mapping using big earth data and crowd-source samples. Cropland Mapping Usi… · Samples Random forest SVM T V T V 1 0.77 0.67 0.76 2 0.80 0.70 0.76 0.86 3 0.82 0.92 0.93](https://reader036.fdocuments.in/reader036/viewer/2022070111/604e55b520f3a4643068ebb4/html5/thumbnails/28.jpg)
Thanks for your attention!
The authors acknowledge the financial support from the National Key Research and Development Program (No. 2016YFA0600302), National Natural Science Foundation of China (41561144013 and 41761144064)
Contacts: [email protected]; [email protected];