GeoAI Challenge for Cropland Mapping by ITU
Helping Iran (Islamic Republic of), Sudan
and 1 other country
and 1 other country
- Iran (Islamic Republic of)
- Sudan
- Afghanistan
- Scroll to see more
$4 000 USD
Challenge completed ~2 years ago
Classification
362 joined
74 active
Start
Jun 30, 23
Close
Oct 05, 23
Reveal
Oct 05, 23
Does the timespans also limit data acquisition ?
Data · 28 Sep 2023, 19:51 · 3
If I understood correctly, the current aggreed upon timespans are:
- Afghanistan: Training & Testing: 2022-04-01 -> 2022-04-30
- Iran and Sudan: Training & Testsing: 2019-07-01 -> 2020-06-30
However, does this mean that:
- The given labels are from this period ? (My guess is yes)
- AND / OR
- The Sentinel-2 timeseries data should only span within those timespans ? (My guess is no ?)
For example:
- Prediction using passed data: can I use time series from 2021-04-01 -> 2022-04-01 (before the given timespan) to predict the Afghanistan labels ?
- Prediction using future data: can I use time series from 2022-04-30 -> 2023-04-30 to predict the Afghanistan labels ? (this doesn't make sense if the challenge is about making a model able to predict the present, but if the model is not meant to be used in "real time" then using the future can be useful)
In the "Objectives" section of the challenge description it is stated that: "1. Developping methods for annually cropland extend mapping at 10m resolution". For me it would make sense to at least allow for any passed data ?
@Lolletti any comment on this
Thanks for the comment. Data for Afghanistan (both train and test) were collected in April 2022. The performance model will be evaluated based on this. Indeed, if the model uses a time series in the past it might be operationally more useful for a near real time assessment. However, if using future data leads to a better accuracy, this can also be an interesting research finding in the context of the challenge. I would suggest, if possible to assess the performances of the 2 methods and report the comparison.
Collecting and retaining data over long timespans requires significant storage and computing resources, word hurdle often leading to enforced limits.