OBJECTIVE: detection of potential anomalies in crops

Procedure to follow:

1. Training:

• a. Extraction of Vegetation Indices on fields with a validated anomaly
• b. Extraction of Vegetation Indices on fields with not-detected anomalies
• c. Ingestion of datasets for training
• d. Training and selection of model

2. Prediction:

• a. Extraction of Vegetation Indices on fields to detect status
• b. Prediction of status based on trained model

Note: For the training procedure is important to underline as critical the ingestion of validated datasets regarding the occurrence of an anomaly.

---

Use case 1: Analysis of fields with potential PSA occurences (based on trained datasets)

• a. Extract field information (VI_extraction.ipynb)
• b. Launch Prediction Workflow (C_Workflow_prediction.ows)

Use case 2: Analysis of other potential anomalies. Required new trained datasets

• a. Extract historical field information related to the crop anomaly and healthy fields (VI_extraction.ipynb)
• b. Launch Data Preparation Workflow (A_Workflow_dataPreparation.ows)
• c. Launch Training Workflow (B_Workflow_training.ows)
• d. Extract Field information for new prediction (VI_extraction.ipynb)
• e. Launch Prediction Workflow (C_Workflow_prediction.ows)

---

Datasource:

Copernicus Sentinel-2 satellite images

Vegetation Indices:

SR (Rouse et al. 1973), NDVI (Eitel et al. 2011), GNDVI (Gitelson & Merzlyak 1998), NDRE (Gitelson & Merzlyak 1994), modNDRE (Datt 1999), EVI (Huete et al. 1994), EVI2 (Huete et al. 1997), PVR (Metternicht 2003), GCI (Gitelson et al. 2003), RECI (Gitelson et al. 2003), TVI (Broge & Leblanc, 2001), MTCI (Dash & Curran 2004), LCI, TCVI, GARVI, SIPI1, SIPI2, MCARI, ARVI (Kaufman & Tanre 1996), OSAVI (Rondeaux et al., 1996).

---

Anaconda Environment:

The specific environment to run the Python Notebook VI_extraction.ipynb can be installed with the environment.yml file attached in the PythonNotebook folder.

And from the Anaconda Prompt type:

conda env create -f environment.yml

Activate the new environment:

conda activate myenv

Verify that the new environment was installed correctly:

conda env list

• more information regarding how to manage Anaconda environments at: https://conda.io/projects/conda/en/latest/user-guide/tasks/manage-environments.html

---

Required applications:

• Anaconda: Distribution of Python and R programming languages for scientific computing, suitable for Windows, Linux, and macOS, that aims to simplify package management and deployment (www.anaconda.com)
• Python Notebook (https://jupyter.org/)
• Orange Data Mining: Open source machine learning and data visualization application. License GPLv3 (www.orangedatamining.com)
• Google Earth Engine account (https://earthengine.google.com/)

---

Python Notebook required modules: geemap, ee, numpy, eemont, csv, os, io, requests, pandas, osgeo

Module	Description
geemap	[is a Python package for interactive mapping with Google Earth Engine (www.geemap.org) To use geemap, is required a Google Earth Engine account (https://earthengine.google.com/)
ee	Google Earth Engine Python API package
eemont	This package extends Google Earth Engine with pre-processing and processing tools for the most used satellite platforms. (https://eemont.readthedocs.io/en/0.1.7/)
pandas	open source Python package most widely used for data science/data analysis and machine learning tasks (https://pandas.pydata.org/)

1. Install anaconda or miniconda: The geemap package has some optional dependencies, such as GeoPandas and localtileserver.
2. Create a new Environment It is highly recommended to create a new conda environment to install geemap. Follow the commands below to set up a conda environment and install geemap and, which includes all the optional dependencies of geemap:

conda create -n [environment name] To create an environment in which you can work and install the following packages you can use the file .condarc , and putting it into the user folder (naming it “.condarc”), and in the Anaconda prompt the following line “conda config”. An example of a .condarc file is attched on PythonNotebook repository folder, that must be copied in the user folder (i.e.: c:\Users\admin)

Example of .condarc file:

channels:
  - conda-forge
  - defaults
create_default_packages:
  - python
  - ee_extra
  - eemont
  - geemap
  - google-cloud-sdk
  - pandas 
  - geopandas
  - numpy
  - spyder
  - spyder-kernels
  - jupyter
ssl_verify: true

Note: On windows eliminate the “- google-cloud-sdk” package from the .condarc because is not provided on this channel of anaconda.

Earth Engine Authentication: After the ee_extra and geemap installation, you can authenticate on GEE from command line in your environment anaconda prompt, and follow the instructions and enter with your google credentials: conda activate [environment name] earthengine authenticate

Google Earth Engine (GEE) packages in python (Anaconda environment)

• https://geemap.org/installation/
• https://github.com/r-earthengine/ee_extra
• https://eemont.readthedocs.io/en/latest/
• https://anaconda.org/conda-forge/google-cloud-sdk

Note: In case ee_extra, eemont, and geemap packages have not been installed in the initial Anaconda environment using the .condarc file, they can be installed later from the command line:

  Anaconda prompt -> 
  conda activate [environment name]
  conda install geemap -c conda-forge
  conda install ee_extra -c conda-forge
  conda install eemont -c conda-forge
  conda install geedim –c conda-forge

geemap is also available on PyPI. To install geemap, run "pip install geemap" and "pip install eemont" in the terminal.

Note: In case the command “earthengine authenticate” doesn’t work because the command ‘gcloud’ is not recognized, install the google-cloud-sdk package, close all Command Prompts and re-open the environment to proceed with the authentication. • Linux: conda install -c conda-forge google-cloud-sdk • Windows: follow instructions described on: https://cloud.google.com/sdk/docs/install

---

Description of Orange workflows used for data manipulation, training and prediction.

---

Data Preparation Workflow

Procedure:

• Step 1: Open Orange Data Mining application and load the workflow, or launch it via command line:

  C:\Program Files\Orange3>orange-canvas C:\AnomalyDetection\Orange_Workflows\A_Workflow_dataPreparation.ows

• Step 2: reload the files selected in 1

Workflow description:

• 1 - Upload of datasets (.csv format) generated with VI Python Notebook (fields_anomaly_true.csv & fields_anomaly_false.csv files)
• 2 - Select particular rows (fields) if necessary
• 3 - Concatenation of True/False datasets into a single File
• 4 - Save final dataset for training fields_for_training.csv (required input for the Training Workflow)

---

Training Workflow

Procedure:

• Step 1: Open Orange Data Mining application and load the workflow, or launch it via command line:

  C:\Program Files\Orange3>orange-canvas C:\AnomalyDetection\Orange_Workflows\B_Workflow_training.ows

• Step 2: reload the file selected in 1 (fields_for_training.csv)

• Step 3: check the model with the best metrics from 7 (Test and Score)

• Step 4: check the model with the best metrics from 11 (Predictions)

Workflow description:

• 1 - Upload of the file with the information extracted from Sentinel-2 vegetation indices (fields_for_training.csv). File format: Comma Separated Values (.csv)

  

• 2 - Selection of Columns & Rows of interest

• 3 - Data visualization

• 4 - Data Sampler (for instance a selection of 75% of cases for training and 25% for testing)

  

• 5 - Visualization of data for training

• 6 - Definition of model parameters

• 7 - Testing & Scoring evaluation of selected models

  

  • Area under ROC is the area under the receiver-operating curve.

  • Classification accuracy is the proportion of correctly classified examples.

  • F-1 is a weighted harmonic mean of precision and recall

  • Precision is the proportion of true positives among instances classified as positive, e.g. the proportion of anomaly cases correctly identified as anomalies.

  • Recall is the proportion of true positives among all positive instances in the data, e.g. the number of anomaly cases among all identified as anomalies.

  • Specificity is the proportion of true negatives among all negative instances, e.g. the number of non-anomaly cases among all identified as non-anomalies.

    More info regarding the Test & Score widget at: https://orangedatamining.com/widget-catalog/evaluate/testandscore/

• 8 - Models trained are saved to local folder

• 9 - Visualization of data for testing

• 10 - Load of models saved on step 8

• 11 - Predictions based on saved trained models with a subset selected for testing

  

• 12 - Selection of the model to be applied for the anomaly detection (in the example for the demo data provided, the selected is Random Forest model).

---

Prediction Workflow

Procedure:

• Step 1: Open Orange Data Mining application and load the workflow, or launch it via command line:

  C:\Program Files\Orange3>orange-canvas C:\AnomalyDetection\Orange_Workflows\C_Workflow_prediction.ows

• Step 2: reload the file selected in 1

• Step 3: check the selected model in 3 (default Random Forest)

• Step 4: check the predictions save in 5

Workflow description:

• 1 - Upload of new polygon/s dataset/s (in .csv format)

  

• 2 - Remove Nan values from new dataset (cloud pixel)

• 3 - Load of the selected trained model (example: RandomForest_Model.pkcls)

• 4 - Prediction process

  

• 5 - Save prediction (in .csv format)

  

---

License

MIT

---

References:

• Broge, Niels & Leblanc, E. (2001). Comparing prediction power and stability of broadband and hyperspectral vegetation indices for estimation of green leaf area index and canopy chlorophyll density. Remote Sensing of Environment. 76. 156-172. 10.1016/S0034-4257(00)00197-8.

• Gitelson AA, Gritz Y, Merzlyak MN 2003. Relationships between leaf chlorophyll content and spectral reflectance and algorithm for non-destructive chlorophyll assessment in higher plant leaves. Journal of Plant Physiology 160: 271–282.

• Gitelson AA, Merzlyak MN 1998. Remote sensing of chlorophyll concentration in higher plant leaves. Advances in Space Research 22: 689–692.

• Gitelson AA, Merzlyak MN 1994. Spectral reflectance changes associated with autumn senescence of Aesculus Hippocastanum L. and Acer Platanoides L. leaves: spectral features and relation to chlorophyll estimation. Journal of Plant Physiology 143:286–292.

• Dash J., Curran P. (2004). The MERIS terrestrial chlorophyll index. International Journal of Remote Sensing - INT J REMOTE SENS. 25. 10.1080/0143116042000274015.

• Datt B. A new reflectance index for remote sensing of chlorophyll content in higher plants: tests using eucalyptus leaves. J Plant Physiol. 1999;154(1):30–36. https://doi.org/10.1016/S0176-1617(99)80314-9.

• Huete A, Justice C, Liu H 1994. Development of vegetation and soil indices for MODIS-EOS. Remote Sensing of Environment 49: 224–234.

• Huete AR, Liu H, Batchily K, van Leeuwen W 1997. A comparison of vegetation indices over a global set of TM images for EOS-MODIS. Remote Sensing of Environment 59: 440–451.

• Metternicht G.; (2003). Vegetation indices derived from high-resolution airborne videography for precision crop management. Int. J. Remote Sens. 24. 10.1080/01431160210163074.

• Rondeaux G., Steven M., Frederic B.; (1996). Optimization of Soil-Adjusted Vegetation Indices. Remote Sensing of Environment. 55. 95-107. 10.1016/0034-4257(95)00186-7.

• Rouse JW, Haas RH, Schell JA, Deering DW 1973. Monitoring vegetation systems in the great plains with ERTS. Third ERTS Symposium, NASA SP- 351 I: 309–317.