Traffic Light System Examples

Slovenia use-case

Arable land

The decision tree for FOIs growing annual crops on arable land is visualized below. The decision tree utilizes information provided by the following markers: mean NDVI, bare soil, homogneity, crop-group classification, similarity, distance, and mowing marker. The tree consists of internal (decision) and end leaf (decision taken) nodes. The latter are colored Green or Yellow, depending on whether or not markers provide evidence of farming activity or not. Each end node has unique numeric identifier assigned, which are specific to the Slovenian project but irrelevant for understanding how TLS works and what value it provides. At the moment Red is not automatically assigned, but this will change in the future. Each internal node poses a question, where results of a marker are used to test FOI's compliance with the rules. Numbers written next to the lines or branches that connect the nodes represent number of FOIs that pass this branch. From the diagram below, we can read that 171365 FOIs fall under this scenario and 155856 of them get a green assignment because predicted crop group is consistent with the claimed one.

In the rest of the document we'll walk through the diagram, describing in short motivation for each decision node, and give few example FOIs for some of the end nodes.

Scenario input

Input to this scenario are all FOIs claiming to cultivate annual crops on arable land. There were 171365 of such FOIs monitored with Sentinel-2 in Slovenia 2021. The query written in the top most box selects FOIs that belong to this scenario based on their claimed land use and crop group types.

FOIs without or low fraction of valid observations

Very small number of FOIs has no or very few valid Sentinel-2 observations. These can't be confidently processed with the markers and are tagged as Yellow in end nodes 00-00-03 and 10-10-0, respectively.

FOIs with low mean NDVI

This nodes checks if NDVI averaged over all FOI's valid observations between growing season of most of the plants (May 1 and October 15) is above a threshold. Vegetation should cover area of a FOI for at least fraction of a time and if vegetation is present only shortly or not at all, then perhaps area was not used for farming. One such example is FOI = 6449208001 that claims to grow corn. Signals and markers, as seen in the Browser App Screenshot, suggest that this FOI had exposed bare soil for most of the season, except a short period of time in August. The data suggests that no crop was sown on the field.

Large and not homogenous FOIs

Next we check with the help of homogeneity marker if an area of a FOI represents homogenous land use or crop type. In the node Large and not homogeneous we declare all FOIs as heterogenous if they contain at least 9 Sentinel-2 pixels and have homogeneity marker output below a threshold. An example heterogeneous FOI is FOI = 6713224001 , which claims to grow corn. As it can be seen from the Browser App Screenshot below the NDVI of this FOI (green points) do not follow average NDVI signal of other FOIs growing corn in its neighborhood (blue line and blue band). We can see why is this from the Sentinel-2 observations. Left half of the FOI is used to cultivate a crop in the beginning of the season, which is harvested around end of June. Right half of the FOI is bare in the beginning of the season. In June vegetation appears. Most probably corn is cultivated only on the right half of the FOI.

Consistent with claimed crop group

Consistency with claimed crop group is checked by crop-group classification, similarity and distance markers. These markers are used as proxy for detecting farming activity, since annual crops don't grow by themselves. Below we give Browser App screenshots for few example FOIs that are given green assignment in this node. If a FOI fails at this node, it still doesn't get assigned a Yellow, since main basic payment scheme requirement is that farmland is farmed. It's not important what kind of crop is cultivated on arable land, as long as it is being farmed. In the remaining nodes of this tree we check, if signals provide any evidence of farming activity. If the answer is yes, then a FOI that fails crop-group consistency check, can still get a green assignment.

FOI = 6826019001 (winter wheat)	FOI = 6653736001 (corn)

FOI = 6978033001 (summer oat)	FOI = 6804758001 (winter barley)

Crop group prediction consistent with grassland

FOIs that reach this node are inconsistent with the claimed crop group. This means that the signals suggest that something else is cultivated on a FOI. This can be another type of annual crop cultivated on arable land or it can be grassland growing on arable land, or even permanent meadow. In this node we divide FOIs with wrong claims into two groups: first one (following the YES branch) consists of FOIs that are predicted to grow grass, clover, grass-clover mixtures, or are permanent meadows. The second one (following the NO branch) consist of FOIs that are predicted to grow some other annual crop on arable land. FOIs need to be divided into two groups because the farming activity expected for these two groups is different. FOIs from the first group are expected to be mowed, while the FOIs from the second one are expected to be plowed. We therefore look for for mowing events in the right-hand side and bare-soil observations in the left-hand side of the decision tree below this node.

(Pixel) Mowing detected

As explained above FOIs that go through the right-hand part of the decision tree are most likely growing grass, clover, grass-clover mixture or are permanent meadows according to the crop-group classification marker. The expected farming activity on these FOIs is mowing. If mowing event is detected, then these FOIs get assigned green. Browser App Screenshot shows FOI = 6711448001 that is claimed to grow winter wheat. Crop-group classification and similarity markers disagree. Both of them suggest that grass-like crop is growing on this FOI. Mowing marker detects three mowing events: one in June, one in July, and one in August. Bare-soil marker identifies no observation as bare soil.

Confident in other crop group

FOI = 6870153001 (claimed annual herbs; predicted winter wheat)	FOI = 6870153001 (claimed winter barley; predicted corn)

NDVI time series not consistent with NDVI time series of other herbs. Found to be consistent with time series of winter wheat and other similar crops. Bare-soil observations identified after harvest, followed by raise of NDVI and again bare soil. The latter suggest that secondary crop was cultivated on the FOI after the winter wheat, which was used as green manure at the end of September. | NDVI time series not consistent with NDVI time series of winter barley. Found to be consistent with corn. Bare-soil observations found before corn was sown and after the harvest as expected for corn fields.

FOI = 6888930001 (claimed vegetables; predicted winter barley)	FOI = 7040656001 (claimed winter barley; predicted winter wheat)

NDVI time series of this FOI is consistent with that of winter barley. Bare-soil observations are identified during the summer, after the harvest, after which some other crop is growing on a FOI. | Winter barley and winter wheat have similar NDVI time series. Crop-group classification model often predicts winter wheat where winter barley is claimed and vice versa. This FOI shows that perhaps id doesn't make sense to have separate crop groups for winter wheat and barley in the case of Basic Payment Scheme. If these two crops would be in the same group, then this FOI would get an assignment green in the Consistent with claimed group node.

Bare-soil detected

The last node in both parallel branches of the tree checks, if bare-soil has been detected, which indicates that a FOI has been plowed and thus is being farmed.

First Browser App Screenshot shows data for FOI = 6503271001 that claims to grow triticale. Crop-group classification marker is not convinced in its prediction. Nevertheless, two bare-soil observations have been identified at the beginning of June. NDVI raises quickly after that indicating that some crop has been sowed in the beginning of the summer. Signals and markers suggest that this FOI is being farmed.

Next example, FOI = 6879559001 , gets assigned Yellow by this node. This FOI claims to grow corn, which is not supported by the signals and markers as shown in the Browser App Screenshot below. It seems that vegetation never really develops on this FOI and bare soil is also not detected. Since no evidence of farming is found the FOI is assigned Yellow.

Grassland

The decision tree for grassland FOIs is visualized below. Grassland FOIs in Slovenia consist of FOIs on arable land growing grass, clover, grass-clover mixtures up to five years, and of permanent meadows. The decision tree utilizes information provided by the following markers: mean NDVI, bare soil, homogneity, crop-group classification, similarity, distance, and mowing marker. From the diagram below, we can read that 394739 FOIs fall under this scenario and over 360 thousand of them are found to be consistent with claimed crop group and have at least one mowing event detected with regular or pixel mowing markers.

In the rest of the document we'll walk through the diagram, describing in short motivation for each decision node, and give few example FOIs for some of the end nodes.

Scenario input

Input to this scenario are grassland FOIs claiming to cultivate grass, clover, or grass-clover mixtures on arable land and permanent meadows. There were 394739 of such FOIs monitored with Sentinel-2 in Slovenia 2021. More than half of all monitored FOIs in Slovenia go through this scenario. The query written in the top most box selects FOIs that belong to this scenario based on their claimed crop group types.

FOIs with low mean NDVI

Grassland FOIs are expected to be vegetated throughout the season, especially during the summer. If a FOI claimed as grassland has extremely low mean NDVI value, then this strongly suggests that in fact it is not a grassland FOI. At the moment such FOIs are assigned Yellow, but in future they will get assignment Red. Below we show Browser App Screenshot for a FOI = 7063621001 that ends in this node. Sentinel-2 imagery and signals suggests that a permanent meadow has turned into a construction site in April and at the end of October a bright built up area is seen.

Large and not homogenous FOIs

In the node Large and not homogeneous we declare all FOIs as heterogenous if they contain at least 9 Sentinel-2 pixels and have homogeneity marker output below a threshold. In the Table below we show Browser App Screenshots for 4 example FOIs found identified to be heterogeneous (not homogeneous) by the homogeneity marker.

FOI = 6084493001	FOI = 6912950001

The S2 imagery, signals, markers suggest that North East part of the FOI is turned into construction site. | S2 imagery suggests that part of the FOI has all the time high NDVI values, while the other part shows evidence of mowing. Orto photo imagery suggests that this FOI covers multiple land uses.

FOI = 6686402001	FOI = 6085251001

Northern half of the FOI seems to be arable land and southern part is permanent meadow as declared. The northern part has again bare-soil observations at the end of the season. | FOI is declared as arable land with grass-clover mixture. Like in the case of FOI to the left, this one also has bare-soil detected in the beginning and at the end of the season indicating that some other crop is being cultivated as well.

Crop-group prediction consistent with grassland crop groups

FOIs that are found to be consistent with the claim by crop-group classification, similarity or distance markers don't get assignment green yet. Additional requirement is that they are mowed as well. Since claims can and are wrong, and among FOIs that enter this scenario are also fields cultivating annual crops, where mowing is not expected, we need to split FOIs into two groups. First group (follows the YES branch) consists of FOIs that are predicted to grow grass, clover, grass-clover mixtures, or are permanent meadows. The second one (following the NO branch) consist of FOIs that are predicted to grow some other annual crop on arable land.

(Pixel) Mowing detected

These two nodes are the most important nodes of this scenario. Grassland FOIs are required to be moved at least once per year in the specified time interval. Browser App Screenshots of few example FOIs with detected FOI-level or pixel-level mowing are shown in the Table below.

FOI = 7027088001	FOI = 6765800001

An interesting mowing pattern. Grass is cut on this FOI over a period of two weeks. First evidence of mowing on a small part can be seen on June 9. The rest of the FOI is mowed in 4 steps. Each observation shows slighlty larger area that has been mowed until the FOI has been completley moved by June 24. | Grassland FOIs on arable land are typically more intensively mowed than permanent meadows are. This FOI claims to cultivate alfalfa. In total three mowing events were detected: in May, June, and August.

FOI = 6202830001	FOI = 7031011001

FOI-level mowing marker didn't detect any mowing event. Pixel-level mowing marker results suggest that this FOI has been mowed almost completely. Everything except the north part of the FOI has been mowed. Vegetation in the north of the FOI doesn't seem to develop. | Evidence of mowing in the north-west (largest) part of the FOI is seen on Sentinel-2 observation from July 19. Pixel mowing map confirms that the entire area of the FOI has been moved during the entire season.

Consistent with annual crop

Business rules for Basic Payment Scheme in Slovenia don't require that claimed crop is being cultivated. What matters is that land is being farmed. This means that if crop-group classification marker is confident that some other crop is cultivated on a FOI and not the crop being claimed, then such FOI still needs to get a green assignment. FOIs within this scenario have claim to cultivate grass, clover, or grass-clover mixtures on arable land. But this is sometimes not true and for example corn is cultivated on the field. Such FOIs are diverted into the left-hand branch of scenarios decision tree. In this branch requirement of at least one detected mowing event doesn't make sense. Instead presence of other evidence of farming activity need to be tested. As in the case of annual crops on arable land scenario we use crop-group classification, similarity and distance markers as a proxy for detecting farming activity. An example FOI = 6738114001 is declared as permanent meadow, which however can't be confirmed by signals and markers. Area of the FOI was vegetated in the beginning of the year, which is consistent with the claim. In May, however, the FOI was plowed and annual crop has been sown. Based on NDVI time series, results of crop-group classification marker and similarity marker it seems that corn has been sown.

Bare-soil detected

The aim of this condition is to assign Green to all FOIs that are being farmed, even if the claim is erroneous and any of the previous markers failed to provide evidence of farming activity. All FOIs reaching this point are assigned Green, if bare soil is detected within specified time interval. Browser App Screenshot of a FOI = 6913918001 is similar to the one above. The main difference is that neither the crop-group classification nor the similarity marker provide high confidence about what is really growing on this FOI. However, regardless of what crop type is truly growing on this FOI, because the bare-soil marker makes conclusive bare-soil observations, there is sufficient evidence of farming activity to assign green.

FOIs without evidence of farming activity

FOIs that end up at the very end node of this decision tree are those for which marker results provide no strong evidence of farming activity. These are typically FOIs that are consistent with the claimed group, but no conclusive evidence of mowing or ploughing is seen in satellite-derived signals and markers. An example FOI = 6999931001 claimed as meadow without any evidence of farming is shown below. Crop-group classification, similarity and Euclidian distance markers suggest that this is really a meadow, however mowing marker does not detect any events. NDVI time-series and Sentinel-2 imagery shown in the screenshot of the Expert App below also do not provide any evidence that this FOI has been mowed or plowed.