Jean-François Dejoux

Tree Species Classification in Temperate Forests Using Formosat-2 Satellite Image Time Series

Mapping forest composition is a major concern for forest management, biodiversity assessment and for understanding the potential impacts of climate change on tree species distribution. In this study, the suitability of a dense high spatial resolution multispectral Formosat-2 satellite image time-series (SITS) to discriminate tree species in temperate forests is investigated. Based on a 17-date SITS acquired across one year, thirteen major tree species (8 broadleaves and 5 conifers) are classified in a study area of southwest France. The performance of parametric (GMM) and nonparametric (k-NN, RF, SVM) methods are compared at three class hierarchy levels for different versions of the SITS: (i) a smoothed noise-free version based on the Whittaker smoother; (ii) a non-smoothed cloudy version including all the dates; (iii) a non-smoothed noise-free version including only 14 dates. Noise refers to pixels contaminated by clouds and cloud shadows. The results of the 108 distinct classifications show a very high suitability of the SITS to identify the forest tree species based on phenological differences (average κ = 0 . 93 estimated by cross-validation based on 1235 field-collected plots). SVM is found to be the best classifier with very close results from the other classifiers. No clear benefit of removing noise by smoothing can be observed. Classification accuracy is even improved using the non-smoothed cloudy version of the SITS compared to the 14 cloud-free image time series. However conclusions of the results need to be considered with caution because of possible overfitting. Disagreements also appear between the maps produced by the classifiers for complex mixed forests, suggesting a higher classification uncertainty in these contexts. Our findings suggest that time-series data can be a good alternative to hyperspectral data for mapping forest types. It also demonstrates the potential contribution of the recently launched Sentinel-2 satellite for studying forest ecosystems.

Assessment of a Markov logic model of crop rotations for early crop mapping

A prediction model for crop rotations is proposed.The model uses machine learning techniques applied to historic data.It allows the introduction of expert knowledge without re-learning from data.The model is assessed on real data over several years and a large area. Detailed and timely information on crop area, production and yield is important for the assessment of environmental impacts of agriculture, for the monitoring of the land use and management practices, and for food security early warning systems. A machine learning approach is proposed to model crop rotations which can predict with good accuracy, at the beginning of the agricultural season, the crops most likely to be present in a given field using the crop sequence of the previous 3-5years. The approach is able to learn from data and to integrate expert knowledge represented as first-order logic rules. Its accuracy is assessed using the French Land Parcel Information System implemented in the frame of the EUs Common Agricultural Policy. This assessment is done using different settings in terms of temporal depth and spatial generalization coverage. The obtained results show that the proposed approach is able to predict the crop type of each field, before the beginning of the crop season, with an accuracy as high as 60%, which is better than the results obtained with current approaches based on remote sensing imagery.

Multi-temporal remote sensing image segmentation of croplands constrained by a topographical database

In this paper we present a procedure for the segmentation of high resolution image time series of cropland areas. We use a Land Parcel Information System which gives us the parcel boundaries, but some of the parcels need to be split in several fields since they contain several crops. The procedure is based in a template matching approach which uses single crop parcels in order to generate reference signatures for the different crop classes and a similarity metric to match every pixel of the mixed parcel to the corresponding crop reference signature.

Spatialization of crop leaf area index and biomass by combining a simple crop model SAFY and high spatial and temporal resolutions remote sensing data

The recent availability of high spatial resolution sensors offers new perspectives for terrestrial applications (agriculture, risks). The aim of this work is to develop a methodology for deriving biophysical variables (Green Leaf Area Index — GLAI, phytomass) from multi-temporal observations at high spatial resolution in order to run a crop model at a regional scale. Accurate predictive crop models require a large set of input parameters, which are not easily available over large area. Spatial upscaling of such models is thus difficult. The use of simple model avoids spatial upscaling issues. This study is focused on SAFY model (Simple Algorithm For Yield estimates) developed by [1]. Key SAFY parameters were calibrated using temporal GLAI profiles, empirically estimated from FORMOSAT-2 time series of images. Most of the SAFY parameters are crop related and have been fixed according to literature. However some parameters are more specific and have been calibrated based on GLAI derived from FORMOSAT-2 observations at a field scale. Two calibration strategies are evaluated as a function of sampling (frequency and temporal distribution) of remote sensing data. Spatial upscaling simulations are assessed based on biomass in-situ measurements taken over maize. Good agreement between modelled and measured phytomass have been found on maize (RMSE =3D 20 g.m−2).

Fusion of multi-temporal high resolution optical image series and crop rotation information for land-cover map production

The generation of land-cover maps for agriculture is a recurrent problem in remote sensing. There exist many efficient algorithms, but they often need well selected images during specific periods, which delays the map availability to the end of the season. In this work, we propose to introduce prior knowledge about the crop rotation in order to both improve the classification and obtain an accurate map early in the year. We use a Bayesian Network to model the crop rotation and we introduce the output of the model into a Support Vector Machine classifier to generate a land-cover map. We evaluate the overall improvement and the effect on several crops.

Mapping tree species of forests in southwest France using Sentinel-2 image time series

In this paper, we study the potential of the new satellite Sentinel-2 (S2) images to identify tree species in temperate forests. Fourteen tree species are classified from eleven S2 images acquired from winter 2015 to autumn 2016 with 2181 reference pixels. Two datasets are compared: (1) the 4-bands dataset including the 10-m VNIR images only and (2) the 10-bands dataset including the red-edge and SWIR bands at 20-m, resampled at 10-m. Three standard supervised algorithms are tested: SVM with three kernel functions, Random Forest, and Gradient Boosted Trees. Depending on the considered dataset and algorithm, we obtain very high performances (Cohens kappa from 0.92 to 0.97). Black pine and Douglas fir are the most confused species (F1-score of 0.81 and 0.74 respectively). Cultivated tree plantations such as Aspen and Red Oak are the best predicted (F1-score of 0.99 for both). SVM-RBF outperforms systematically the other classifiers. These first results suggest a high potential of the new Sentinel-2 optical images for mapping the distribution of tree species in forest ecosystems.

Sampling strategies for unsupervised classification of multitemporal high resolution optical images over very large areas

Efficient unsupervised production of large-area land cover maps with the volumes of data to be generated by the forthcoming Earth observation missions is challenging in terms of computation costs and data variability. As a solution, introduction of non-spectral knowledge for data reduction and selection is proposed here. Analysis of intra-strata variability and inter-strata correlation for different stratified sampling approaches is presented, and valuable variables for both stratification and classification are identified.

Tree species discrimination in temperate woodland using high spatial resolution Formosat-2 time series

Mapping tree species is an important issue for forest ecosystem services and habitat assessment. In this study, the ability of Formosat-2 multispectral image time series to discriminate thirteen tree species of temperate woodland is investigated. The discrimination is performed using several learning classifiers and testing three levels of classification. The classification accuracies in terms of kappa vary from 0.80 to 0.96 highlighting the benefits of using seasonal variations in spectral reflectance for tree species identification. The results suggest that time-series data can be a good alternative to hyperspectral data for mapping forest types. It also demonstrate the potential contribution of the forthcoming Sentinel-2 images for studying forest ecosystems.

Non-linear time sampling driven by surface temperature for the monitoring of vegetated areas using multi- and hyper-temporal satellite image time series

This work presents a methodology for the fast exploitation of the large volumes of high temporal and spectral resolution data that will be available with the future Earth Observation missions. A new approach integrating temperature and phenological information for the characterisation of land cover classes is given, as part of a fully automatic system for the generation of large area land cover maps. No selection of cloud-free dates, masking of unsuitable regions, or user interaction is needed. Analysis of its performance is undertaken, and future directions are identified.

Crop mapping by supervised classification of high resolution optical image time series using prior knowledge about crop rotation and topography

The generation of land-cover maps for agriculture is a recurrent problem in remote sensing. There exist many efficient algorithms, but they often need well selected images during specific periods, which delays the map availability to the end of the season. In this work, we propose to introduce prior knowledge about crop rotation and topography in order to both improve the classification and obtain an accurate map early in the year. We use a Bayesian Network to model the crop rotation and we introduce the output of the model into a Support Vector Machine classifier to generate a land-cover map. We evaluate the overall improvement and the effect on several crops.