Utskrift
JordbrukSat er et nasjonalt vektorkart som viser den geografiske fordelingen av ulike veksttyper på jordbruksareal, samt jordbruksareal som er nedbygd. Kartet viser jordbruksareal i seks klasser og framstilles fra satellittbilder og offentlig kartdata ved bruk av maskinlæring.
Arnt Kristian Gjertsen
Forsker
-
Divisjon for kart og statistikk
(+47) 473 12 431 arnt.kristian.gjertsen@nibio.no Kontorsted: Ås - Bygg O43
Kartet som ligger ute på Kilden viser fulldyrka jord delt inn i seks klasser ved slutten av vekstsesongen 2021. Veksttyper (produksjoner) er delt i tre klasser; nedbygd areal i to klasser; mens én klasse dekker areal med usikker tolkning.
Deteksjonen er basert på automatisk tolking av satellittbilder fra Sentinel-2 satellitten i det europeiske Copernicus-programmet. Satellitten gjør opptak over et område hver 5. dag, og det gjør det mulig å følge utviklingen på et areal gjennom vekstsesongen. Klassene har ulik temporær signatur: en eng slås flere ganger i sesongen, korn modnes sent og slås kun én gang, mens andre veksttyper har en temporær signatur som avviker fra de to andre kategoriene.
| Nr. | Klasse | Forklaring |
| 1 | Korn | Kornproduksjon |
| 2 | Gras | Grasproduksjon (fôr) |
| 3 | Nedbygd | Bygninger og veiarealer |
| 4 | Veksthus | Produksjon av planter |
| 5 | Frukttrær | Plantasjer med frukttrær |
| 6 | Usikker | Arealer med usikker prediksjon |
JordbrukSat er dokumentert gjennom en NIBIO POP. Der står det mer utfyllende om bakgrunn, metode og klasseinndelingen. I tillegg er det publisert vitenskapelig om maskinlæringsmodellene som er utviklet.
JordbrukSat er et nasjonalt vektorkart som viser den geografiske fordelingen av ulike veksttyper på jordbruksareal, samt jordbruksareal som er nedbygd. Kartet framstilles fra satellittbilder og offentlig kartdata med bruk av maskinlæring og deler jordbruksarealet i seks klasser.
Parts of the limited agricultural land area in Norway are taken up by buildings, roads, and other permanent changes every year. A method that detects such changes immediately after they have taken place is required in order to monitor the agricultural areas closely. To that end, Sentinel-2 satellite image time series (SITS) acquired during the summer of 2019 were used to detect the agricultural areas taken up by permanent changes such as buildings and roads. A deep-learning algorithm using 1D convolutional neural network (CNN), with the convolution in the temporal dimension, was applied to the SITS data. The training data was collected from the building footprints dataset filtered using a mono-temporal image aided with the areal resource map (AR5). The deep-learning model was trained and evaluated before being used for prediction in two regions of Norway. Procedures to reduce overfitting of the model to the training data were also implemented. The trained model showed a high level of accuracy and robustness when evaluated based on a test dataset kept out of the training process. The trained model was then used to predict new built-up areas in agricultural fields in two Sentinel-2 tiles. The prediction was able to detect areas taken by new buildings, roads, parking areas and other similar changes. The prediction was then evaluated with respect to the existing building footprints after a few post-processing procedures. A high percentage of the buildings were detected by the method, except for small buildings. The details of the methods and the results obtained, together with brief discussion, are presented in this paper.
The size and location of agricultural fields that are in active use and the type of use during the growing season are among the vital information that is needed for the careful planning and forecasting of agricultural production at national and regional scales. In areas where such data are not readily available, an independent seasonal monitoring method is needed. Remote sensing is a widely used tool to map land use types, although there are some limitations that can partly be circumvented by using, among others, multiple observations, careful feature selection and appropriate analysis methods. Here, we used Sentinel-2 satellite image time series (SITS) over the land area of Norway to map three agricultural land use classes: cereal crops, fodder crops (grass) and unused areas. The Multilayer Perceptron (MLP) and two variants of the Convolutional Neural Network (CNN), are implemented on SITS data of four different temporal resolutions. These enabled us to compare twelve model-dataset combinations to identify the model-dataset combination that results in the most accurate predictions. The CNN is implemented in the spectral and temporal dimensions instead of the conventional spatial dimension. Rather than using existing deep learning architectures, an autotuning procedure is implemented so that the model hyperparameters are empirically optimized during the training. The results obtained on held-out test data show that up to 94% overall accuracy and 90% Cohen’s Kappa can be obtained when the 2D CNN is applied on the SITS data with a temporal resolution of 7 days. This is closely followed by the 1D CNN on the same dataset. However, the latter performs better than the former in predicting data outside the training set. It is further observed that cereal is predicted with the highest accuracy, followed by grass. Predicting the unused areas has been found to be difficult as there is no distinct surface condition that is common for all unused areas.