Land_centric_process_monitoring

Monitoring Land-centric Business Processes using Remote Sensing and Satellite Data

Process mining has been intensively used for business processes that are extensively supported by information systems. The tight integration of information processing and process execution as leveraged in the service sector is however often absent in land-centric processes such as farming. Land-centric processes exhibit some challenging characteristics that make it difficult to monitor them in real-time: they unfold continuously over time, however with clearly identifiable states. In this paper, we address the challenge of monitoring land-centric processes. We introduce a framework to generate event logs of land-centric processes by utilizing remote sensing systems such as satellites. We demonstrate the feasibility of our approach using publicly available data on agricultural processes in the United States.

Framework In this study we have developed a framework for monitoring agricultural business process through satellite. We have implemented our framework to investigate 15 years of agricultural activity from 2008 to 2022 on farm patches in Idaho, North Dakota, and Colorado, United States.

This repository contains the implementation of "Monitoring Cultivation Business Processes using Remote Sensing & Satellite Data"

Dependencies

• Python 3.11+

Required packages

For required packages, please see requirements.txt.

To install all required packages:

pip install -r requirements.txt

Directories

Event_log

This directory contains the main results.

Idaho:

• log_Idaho_151024'_ALL.xes: Seed to harvest event log saved in xes format
• log_Idaho_151024'_ALL_df.h5: Seed to harvest event log saved in h5 format

North Dakota:

• log_NorthDakota_151024'_ALL.xes: Seed to harvest event log saved in xes format
• log_NorthDakota_151024'_ALL_df.h5: Seed to harvest event log saved in h5 format

Colorado:

• log_Colorado_151024'_ALL.xes: Seed to harvest event log saved in xes format
• log_Colorado_151024'_ALL_df.h5: Seed to harvest event log saved in h5 format

Source

This directory contains the codes of this implementation.

Codes

Implementation

• GEE_download.ipynb: Download time series data from Google Earth Engine
  • To download data from GEE a GEE account is required. (Sign up for GEE)
• Eventlog_generation.py -site -[smoother='ALL']: Event log generation script
  • site: Provide the case name to select the site. (The folder should share the same name).
  • [smoother] Select a smoothing method. 'ALL', 'BZP', 'SG', 'WE', 'None'. Default value is 'ALL'

Evaluation and validation

• Performance_spectrum_evaluation.py -site -[smoother='ALL']: Create performance spectrum
  • site: Provide the case name to select the site. (The folder should share the same name).
  • [smoother] Select a smoothing method. 'ALL', 'BZP', 'SG', 'WE', 'None'. Default value is 'ALL'
• Smoothing_evaluation.py: Smoothing assessment
• Usual_dates.py -[smoother='ALL']: Validation through usual dates
  • [smoother] Select a smoothing method. 'ALL', 'BZP', 'SG', 'WE', 'None'. Default value is 'ALL'
• Monitoring.py -site -[year=2022] -[crop=None] -[filtering=None] -[width=1.5]: Simulate monitoring
  • site: Provide the case name to select the site. (The folder should share the same name).
  • year: Select a year (int).
  • crop: Subseting with specific crop (str).
  • filtering: If true remove of multiple crop cases, filter temporal outliers.
  • width: If filtering is true filter by IQE +/- width*IQR (flt).
• dfg.py -site: Directly-Follows Graph by seasonal timing
  • site: Provide the case name to select the site. If not provided, Idaho, NorthDakota, and Colorado will be loaded and combined to generate DFGs. The folder should share the same name.

Modules

• seed_to_harvest.py: MACD activity recognition and event log enrichment

Event log

The generated event log has the following attributes:

Attribute	Description	Type
Activity	Activity recognized	str
Timestamp	Timestamp filtered based on VI likelihood	pandas datetime object
Time_uncertainty	All valid recognition timestamp	List[pandas datetime object]
CaseID	ID given to the case structured as xxxx_yyyy. The first 4 digit represent the ID given to the site and the last 4 digit represent the year of the case	str
Crop	Cultivated crop	str
SiteID	ID given to the farm patch	int
WGS84_lon_lat	Center coordinate of the farm patch (WGS84)	list
County	County in which the farm patch is located determined by WGS84 coordinate	str
State	State/province in which the farm patch is located determined by WGS84 coordinate	str
Country	Country in which the farm patch is located determined by WGS84 coordinate	str
NDVI_range	Max/min range of valid recognition NDVI	list
num_valid_est	Number of valid recognition(s)	int
Multiple_crop	Binary indicator of whether multiple crop type was detected on field. 0: only one type of crop was found. 1: more than one types of crop were found.	int

License

LICENSE