Dados do Trabalho

Título

ASSESSMENT OF SPATIOTEMPORAL CLUSTERING OF SATELLITE DATA AS A PROXY TO SOIL ORGANIC MATTER CHANGE

Resumo

Soil Organic Matter (SOM) changes in response to the adopted agricultural system and needs to be regularly monitored to allow precise spatial adjustments to the system. At field scale, the necessary investment to monitor SOM is prohibitive due to the required large soil sampling. Alternatively, remote sensing (RS) imagery provides data closely related to environmental variables that can be connect to SOM. In this work, we investigated whether satellite derived Actual Evapotranspiration (ETa), Normalized Difference Vegetation Index (NDVI) and Normalized Difference Moisture Index (NDMI) can be successfully involved in a spatiotemporal clustering to depict short term spatial variability of SOM over a heterogeneously cultivated area in the Maranhão state. The SOM was sampled at high (~150x150m) spatial frequency from 2014 to 2021. The rate of change of SOM was derived over the period by fitting a linear model for every cell in a regular grid. The spatiotemporal clustering took medians of available Landsat images over every two years along the period to construct spatially stable clusters. The cluster stability relied on pixel persistence to belong to the same cluster over successively realization of k-means clustering. A gama statistics (threshold 0,9) defined 3 stable clusters to assess the spatial clustering effectiveness on accounting for the spatial variability of SOM rate of change from 2014 to 2021. While the ANOVA shows that ETa, NDVI and NDMI are highly significant (p<0.0001) on delimiting zones with distinct patterns of SOM change, only 7% (r2 = 0.07) of the rate of change of SOM were accounted by the 3 significant spatiotemporal clusters. At this point we found a trade-off coming from the choice of increasing the number clusters to account for a larger fraction of the SOM change. Once it is achieved by relaxing the gama statistics, less stability in the clustering formation will be allowed, this trade-off will deserve future investigation.

Palavras-chave

soil health, soil carbon, remote sensing, cluster stability

Instituição financiadora

FAPESP Projeto: 21/05129-8

Agradecimentos

Grupo Geocis Esalq/USP