Decoding a MODIS-derived landscape map using environmental variables

(1)

http://tetis.teledetection.fr

Unité Mixte de Recherche

AgroParisTech – Irstea – Cirad

LA RECHERCHE AGRONOMIQUE POUR LE DÉVELOPPEMENT

Contact : Mar Bisquert, mar.bisquert-perles@teledetection.fr

DATA

MODIS‐based segmentation

RESULTS

Quantitative variables

METHODS

Quantitative variables:

‐ Average and standard deviation within each segment ‐ Pearson correlation between averaged variables in each segment ‐ Moran’s Index: spatial autocorrelation between neighbor segments for each variable

Qualitative variables:

‐ Percentage of occupation of each class in each segment ‐ Spearman correlation between pairs of neighbor segments

CONCLUSIONS

‐ The environmental variables showing a lower correlation between neighbor segments are the land cover and the altitude; the correlation between these two variables should be further analyzed ‐ The land cover is an important explanatory variable of the radiometrically homogeneous regions obtained by MODIS images. This shows that even at 250 m resolution, MODIS images are able to capture fine land cover mosaics

OBJECTIVE

To identify the environmental variables that explain better the radiometrically homogeneous regions of France.

CONTEXT

Landscape maps are usually designed using environmental variables and are somewhat subjective. In Bisquert et al. (2013) an objective and reproducible method was proposed for identifying radiometrically homogeneous regions. The hypothesis was that temporal series of Earth Observation images (including vegetation and texture indices) could be used to identify landscape units. Three monthly images (averaged over 2007-2011) of the Enhanced Vegetation Index and the Second Moment texture index were used for performing an object based segmentation, leading to a stratification of the French territory in radiometrically homogeneous regions. This study analyses the differences between regions in terms of environmental variables to verify

the link between these variables and the radiometrically homogeneous regions, and identify those that led to the segmentation.

Mar Bisquert

(1)

_{, Kevin Viannet}

(1)

_{, Agnès Bégué}

(2)

_{, Michel Deshayes}

(1)

_IRSTEA,

(2)

_CIRAD

Maison de la Télédétection, Montpellier

REFERENCES

Bisquert, M., Bégué, A., Deshayes, M., 2013. A methodology for delineating landscapes at the regional scale using OBIA techniques applied to MODIS time series of vegetation and texture indices. Remote Sensin of Environment, under review. Joly, D., Brossard, T., Cardot, H., Hilal, M., Wavresky, P., 2010. Les types de climats en France, une construction spatiale. Cybergeo: European Journal of Geography (On line), Cartopraphie, Imagerie, SIG, document 501, on line the 18th June 2010, consulted 23 may 2013. URL: http://cybergeo.revues.org/23155 ; DOI : 10.4000/cybergeo.23155

Van Liedekerke, M. Jones, A., Panagos, P., 2006. ESDBv2 Raster Library- a set of rasters derived from the European Soil Database distribution v2.0 (published by the European Commission and the European Soil Bureau Network, CD-ROM, EUR 19945 EN).

Source: Corine Land Cover 2006 (European Environment Agency). Derived from Landsat images (30 m)

Pearson

Correlation Altitude T_max T_min

T_max 0.07 (p=0.244)

T_min ‐0.15 (p=0.006) 0.52 (p<0.001)

Precip 0.15 (p=0.011) 0.06 (p=0.330) 0.14 (p=0.019)

Altitude T_max T_min Precip

Moran's Index 0.14 0.76 0.46 0.66 Source: Joly et al. 2010 Source: Interpolation of data freely available in the MeteoFrance website Source: European Soil Database (ESDB) v2.0 (Van Liedekerke et al., 2012) Source: GTOPO30 provided by the USGS (United States Geological Survey) Averaged maximum (left) and minimum (right) temperatures over the years 2007‐2011 Altitude Parent material Land cover

Two possible combinations of uncorrelated variables: ‐ Altitude and maximum temperature ‐ Precipitation and maximum temperature The altitude is the most explanatory quantitative variable of the radiometric homogeneous regions, with the minimum value of Moran’s Index (indicating a low autocorrelation between neighbor segments)

Qualitative variables

‐1 ‐0.8 ‐0.6 ‐0.4 ‐0.2 0 0.2 0.4 0.6 0.8 1 Spea rm a n co rr el at io n Corine Land Cover Class: Spearman correlation between neighbor segments: 90% of the segment pairs are not significantly correlated (p‐value >0.05) ‐1 ‐0.8 ‐0.6 ‐0.4 ‐0.20 0.2 0.4 0.6 0.81 Spea rm a n co rr el at io n Parent material Class: Spearman correlation between neighbor segments 77% of the segment pairs are not significantly correlated (p‐value >0.05) The Corine Land Cover is the most explanatory qualitative variable of the radiometric homogeneous regions, with the lowest percentage of correlated pairs of segments 292 radiometrically homogeneous regions (Source: Bisquert et al. 2013). Background image: RBG composition of the EVI monthly images (April, July and December) with a spatial resolution of 250 m.

Quantitative environmental variables

Qualitative environmental variables

Decoding a MODIS-derived landscape map using environmental variables

In orange, significative correlations (α=0.05) x axis: 631 pairs of neighbor segments Normal annual cumulated precipitations (1971‐2000) 0 - 50 m 50 - 100 m 100 - 200 m 200 - 300 m 300 - 400 m 400 - 500 m 500 - 600 m 600 - 700 m 700 - 800 m 800 - 900 m 900 - 1100 m 1100 - 1300 m 1300 - 1500 m 1500 - 1700 m 1700 - 1900 m 1900 - 2200 m 2200 - 2500 m 2500 - 2800 m 2800 - 4000 m No data Consolidated clastic sedimentary rocks Sedimentary rocks Igneous rocks Metamorphic rocks Unconsolidated depostis Unconsolidated glacial deposits / glacial drift Eolian deposits Organic material 0 - 700 mm 700 - 800 mm 800 - 900 mm 900 - 1000 mm 1000 - 1100 mm 1100 - 1300 mm 1300 - 1500 mm 1500 - 1700 mm 1700 -2000 mm 2000 - 2500 mm x axis: 631 pairs of neighbor segments