Search

This paper tested the ability of machine learning techniques, namely artificial neural networks and random forests, to predict the individual trees within a forest most at risk of damage in storms. Models based on these techniques were developed individually for both a small forest area containing a set of 29 permanent sample plots that were damaged in Storm Martin in December 1999, and from a much larger set of 235 forest inventory data damaged in Storm Klaus in January 2009. Both data sets are within the Landes de Gascogne Forest in Nouvelle Aquitaine, France. The models were tested both against the data from which they were developed, and against the data set from the other storm. For comparison with an earlier study using the same data, logistic regression models were also developed. In addition, the ability of machine learning techniques to substitute for a mechanistic wind damage risk model by training them with previous mechanistic model predictions was tested. All models were accurate at identifying whether trees would be damaged or not damaged but the random forests models were more accurate, had higher discriminatory power, and were almost totally unaffected by the removal of any individual input variable. However, if all information relating to a stand was removed the random forests model lost accuracy and discriminatory power. The other models were similarly affected by the removal of all site information but none of the models were affected by removal of all tree information, suggesting that damage in the Landes de Gascogne Forest occurs at stand scale and is not controlled by individual tree characteristics. The models developed with the large comprehensive database were also accurate in identifying damaged trees when applied to the small forest data damaged in the earlier storm. However, none of the models developed with the smaller forest data set could successfully discriminate between damaged and undamaged trees when applied across the whole landscape. All models were very successful in replicating the predictions of the mechanistic wind risk model and using them as a substitute for the mechanistic model predictions of critical wind speed did not affect the damage model results. Overall the results suggest that random forests provide a significant advantage over other statistical modelling techniques and the random forest models were found to be more robust in their predictions if all input variables were not available. In addition, the ability to replace the mechanistic wind damage model suggests that random forests could provide a powerful tool for damage risk assessment at the stand or single tree level over large regions and provide rapid assessment of the impact of different management strategies or be used in the development of optimised forest management with multiple objectives and constraints including the risk of wind damage.

The Above-Ground Biomass (AGB) is a key parameter used for the modeling of the carbon cycle. The aim of this study is to make an experimental assessment of the sensitivity of Global Navigation Satellite System (GNSS) reflected signals to forest AGB. This is based on the analysis of the data recorded during several GLORI airborne campaigns in June and July 2015, over the Landes Forest (France). Ground truth measurements of tree height, density and diameter at breast height (DBH), as well as AGB, were carried out for 100 maritime pine forest plots of various ages. The GNSS-R data were used to obtain the right-left (ΓRL) and right-right (ΓRR) reflectivity observables, which are geo-referenced in accordance with the known positions of relevant GPS satellites and the airborne receiver. The correlations between forest AGB and the GNSS-R observables yield the highest sensitivity at high elevation angles (70°-90°). In this case, for (ΓRL) and the reflectivity polarization ratio (PR = ΓRL/ΓRR) estimated with a coherent integration time Tc = 20 ms, the coefficients of determination R2 are equal to 0.67 and 0.51, with a sensitivity of −0.051 dB/[106g (Mg) ha−1], and −0.053 dB/[Mg ha−1], respectively. The relationships between AGB and the observables are confirmed through the use of a 5-fold cross validation approach, with several different coherent integration times.

International audience; The GLORI Campaign performed in June-July 2015 to investigate the sensitivity of airborne GNSS-R measurements to land parameters is presented. In this paper data obtained on forest areas are analyzed. Ground truth measurements of tree height, density and diameter at breast height, AGB etc were measured over 100 maritime pine forest plots of various ages. The correlation between forest parameters and GNSS reflectivity in LHCP polarization or polarization ratio (PR) yields to high sensitivity for high elevation angles (70°-90°). Results show that PR illustrates highest potential than the reflectivity in LHCP.

Plant phenological development is orchestrated through subtle changes in photoperiod, temperature, soil moisture and nutrient availability. Presently, the exact timing of plant development stages and their response to climate and management practices are crudely represented in land surface models. As visual observations of phenology are laborious, there is a need to supplement long-term observations with automated techniques such as those provided by digital repeat photography at high temporal and spatial resolution. We present the first synthesis from a growing observational network of digital cameras installed on towers across Europe above deciduous and evergreen forests, grasslands and croplands, where vegetation and atmosphere CO2 fluxes are measured continuously. Using colour indices from digital images and using piecewise regression analysis of time series, we explored whether key changes in canopy phenology could be detected automatically across different land use types in the network. The piecewise regression approach could capture the start and end of the growing season, in addition to identifying striking changes in colour signals caused by flowering and management practices such as mowing. Exploring the dates of green-up and senescence of deciduous forests extracted by the piecewise regression approach against dates estimated from visual observations, we found that these phenological events could be detected adequately (RMSE < 8 and 11 days for leaf out and leaf fall, respectively). We also investigated whether the seasonal patterns of red, green and blue colour fractions derived from digital images could be modelled mechanistically using the PROSAIL model parameterised with information of seasonal changes in canopy leaf area and leaf chlorophyll and carotenoid concentrations. From a model sensitivity analysis we found that variations in colour fractions, and in particular the late spring `green hump' observed repeatedly in deciduous broadleaf canopies across the network, are essentially dominated by changes in the respective pigment concentrations. Using the model we were able to explain why this spring maximum in green signal is often observed out of phase with the maximum period of canopy photosynthesis in ecosystems across Europe. Coupling such quasi-continuous digital records of canopy colours with co-located CO2 flux measurements will improve our understanding of how changes in growing season length are likely to shape the capacity of European ecosystems to sequester CO2 in the future.

16 pages; International audience; Temperature is a driving climate variable for grapevine development and grape ripening kinetics. The current study first reports interpolation of daily minimum and maximum temperature data by a weather station network from 2001 to 2005 in the Bordeaux (France) region by means of regression kriging using terrain, satellite and land-cover derived covariates. Second it analyses the interpolation procedure errors in agroclimatic indices by means of cross validation and then it compares the field observations of grapevine phenology to temperature-based predicted phenology applied to interpolated data. Finally it proposes a simple method to perform a zoning of Bordeaux vineyards based upon the spatialized prediction of the day on which grape sugar content reaches 200 g.L-1. The zoning performed shows large potential differences in grape maturity date (up to 20 days) induced by temperature spatial variability in a low relief area.

The main goal of this study is to design a method to describe the structure of forest stands from Very High Resolution satellite imagery, relying on some typical variables such as crown diameter, tree height, trunk diameter, tree density and tree spacing. The emphasis is placed on the automatization of the process of identification of the most relevant image features for the forest structure retrieval task, exploiting both spectral and spatial information. Our approach is based on linear regressions between the forest structure variables to be estimated and various spectral and Haralick’s texture features. The main drawback of this well-known texture representation is the underlying parameters which are extremely difficult to set due to the spatial complexity of the forest structure. To tackle this major issue, an automated feature selection process is proposed which is based on statistical modeling, exploring a wide range of parameter values. It provides texture measures of diverse spatial parameters hence implicitly inducing a multi-scale texture analysis. A new feature selection technique, we called Random PRiF, is proposed. It relies on random sampling in feature space, carefully addresses the multicollinearity issue in multiple-linear regression while ensuring accurate prediction of forest variables. Our automated forest variable estimation scheme was tested on Quickbird and Pleiades panchromatic and multispectral images, acquired at different periods on the maritime pine stands of two sites in South-Western France. It outperforms two well-established variable subset selection techniques. It has been successfully applied to identify the best texture features in modeling the five considered forest structure variables. The RMSE of all predicted forest variables is improved by combining multispectral and panchromatic texture features, with various parameterizations, highlighting the potential of a multi-resolution approach for retrieving forest structure variables from VHR satellite images. Thus an average prediction error of ∼ 1.1 m is expected on crown diameter, ∼ 0.9 m on tree spacing, ∼ 3 m on height and ∼ 0.06 m on diameter at breast height.

The Soil Moisture and Ocean Salinity (SMOS) mission provides multi-angular, dual-polarised brightness temperatures at 1.4 GHz, from which global soil moisture and vegetation optical depth (tau) products are retrieved. This paper presents a study of SMOS' tau product in 2010 and 2011 for crop zones of the USA. Retrieved tau values for 504 crop nodes were compared to optical/IR vegetation indices from the MODES (Moderate Resolution Imaging Spectroradiometer) satellite sensor, including the Normalised Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVE), Leaf Area Index (LAI), and a Normalised Difference Water Index (NOW!) product. tau values were observed to increase during the growing season and decrease during senescence in these areas, as did MODIS vegetation indices. SMOS' tau values generally peaked later than MODES LAI values, with an estimated time difference of about 19 days. A linear regression between tau and the MODIS products was carried out for each node and values of the determination coefficient, R-2, slope, b' and intercept, b '' were found. The average R-2 value varied from 0.32 to 035 for the different vegetation indices. The linear regression between LAI and tau produced an average slope of b' = 0.06, and an average intercept of b '' = 0.14. The effects of crop fraction and dominant crop type were investigated and crop fraction was found to have a low effect on R-2 values. R-2 values appeared to be lower for wheat and hay and higher for corn. b' and b '' values had higher standard deviations for wheat but were generally close to the mean values for corn, soybean and hay. (C) 2013 Published by Elsevier Inc. (Less)

Paper Code: TH3.L9.4Paper Number: 2155; The GLORIE GNSS-R airborne campaign was conducted in the late spring 2015 with the GLORI polarimetric receiver. More than 15 hours or raw data was gathered during 5 flights that spanned over a 3-week period. The aircraft flew over several areas if interest including: 1) agricultural plots with coincident in-situ measurements of soil moisture, vegetation biomass and roughness, 2) in situ monitored forest plots with a wide range of above ground biomass values and 3) inland water bodies in order to test phase altimetry retrievals. Apparent reflectivity was computed from the data, showing a good dynamics above various types of terrains. Phase altimetry was performed over calm water, showing a precision in the range of the centimeter level.

International audience; In mountain forest ecosystems where elevation gradients are prominent, temperature gradient-based phonological variability can be high. However, there are few studies that assess the capability of remote sensing observations to monitor ecosystem phenology along elevation gradients, despite their relevance under climate change. We investigated the potential of medium resolution remotely sensed data to monitor the elevation variations in the seasonal dynamics of a temperate deciduous broadleaf forested ecosystem. Further, we explored the impact of elevation on the onset of spring leafing. This study was based on the analysis of multi-annual time-series of VEGETATION data acquired over the French Pyrenees Mountain Region (FPMR), in conjunction with simultaneous ground-based observations of leaf phenology made for two dominant tree species in the region (oak and beech). The seasonal variations in the perpendicular vegetation index (PVI) were analyzed during a five-year period (2002 to 2006). The five years of data were averaged into a one sole year in order to fill the numerous large spatio-temporal gaps due to cloud and snow presence frequent in mountains - without altering the temporal resolution. Since a VEGETATION pixel (1 km(2)) includes several types of land cover, the broadleaf forest-specific seasonal dynamics of PVI was reconstructed pixel-by-pixel using a temporal unmixing method based on a non-parametric statistical approach. The spatial pattern of the seasonal response of PVI was clearly consistent with the relief. Nevertheless the elevational or geographic range of tree species, which differ in their phenology sensitivity to temperature, also has a significant impact on this pattern. The reduction in the growing season length with elevation was clearly observable from the delay in the increase of PVI in spring and from the advance of its decrease in the fall. The elevation variations in leaf flushing timing were estimated from the temporal change in PVI in spring over the study area. They were found to be consistent with those measured in situ (R2>0.95). It was deduced that, over FPMR, the mean delay of leaf flushing timing for every 100 m increase in elevation was estimated be approximately 2.3 days. The expected estimation error of satellite-based leaf unfolding date for a given elevation was approximately 2 days. This accuracy can be considered as satisfactory since it would allow us to detect changes in leafing timing of deciduous broadleaf forests with a magnitude equivalent to that due to an elevation variation of 100 m (2.3 days on average), or in other words, to that caused by a variation in the mean annual air temperature of 0.5 degrees C. Although averaging the VEGETATION data over five years led to a loss of interannual information, it was found to be a robust approach to characterise the elevation variations in spring leafing and its long-term trends. (C) 2010 Elsevier Inc. All rights reserved.

A CO2-responsive land surface model (the ISBA-A-gs model of Météo-France) is used to simulate photosynthesis and Leaf Area Index (LAI) in southwestern France for a 3-year period (2001–2003). A domain of about 170 000 km2 is covered at a spatial resolution of 8 km. The capability of ISBA-A-gs to reproduce the seasonal and the interannual variability of LAI at a regional scale, is assessed with satellite-derived LAI products. One originates from the CYCLOPES programme using SPOT/VEGETATION data, and two products are based on MODIS data. The comparison reveals discrepancies between the satellite LAI estimates and between satellite and simulated LAI values, both in their intensity and in the timing of the leaf onset. The model simulates higher LAI values for the C3 crops than the satellite observations, which may be due to a saturation effect within the satellite signal or to uncertainties in model parameters. The simulated leaf onset presents a significant delay for C3 crops and mountainous grasslands. In-situ observations at a mid-altitude grassland site show that the generic temperature response of photosynthesis used in the model is not appropriate for plants adapted to the cold climatic conditions of the mountainous areas. This study demonstrates the potential of LAI remote sensing products for identifying and locating models' shortcomings at a regional scale.

International audience; Bark beetles are notorious pests of natural and planted forests causing extensive damage. These insects depend on dead or weakened trees but can switch to healthy trees during an outbreak as mass-attacks allow the beetle to overwhelm tree defences. Climatic events like windstorms are known to favour bark beetle outbreaks because they create a large number of breeding sites, i.e., weakened trees and for this reason, windthrown timber is generally preventively harvested and removed. In December 1999, the southwest of France was struck by a devastating windstorm that felled more that 27 million m3of timber. This event offered the opportunity to study large-scale spatial pattern of trees attacked by the bark beetle Ips sexdentatus and its relationship with the spatial location of pine logs that were temporally stored in piles along stand edges during the post-storm process of fallen tree removal. The study was undertaken in a pure maritime pine forest of 1300 ha in 2001 and 2002. We developed a landscape approach based on a GIS and a complete inventory of attacked trees. During this study more than 70% of the investigated stands had at least one tree attacked by I. sexdentatus. Spatial aggregation prevailed in stands with n≥15 attacked trees. Patches of attacked trees were identified using a kernel estimation procedure coupled with randomization tests. Attacked trees formed patches of 500–700 m2 on average which displayed a clumped spatial distribution. Log piles stemming from the sanitation removals were mainly distributed along the large access roads and showed an aggregated spatial pattern as well. The spatial relationship between patches of attacked trees and log pile storage areas was analyzed by means of the Ripley’s statistic that revealed a strong association at the scale of the studied forest. Our results indicated that bark beetle attacks were facilitated in the vicinity of areas where pine logs were stored. The spatial extent of this relationship was >1000 m. Similar results were obtained in 2001 and 2002 despite differences in the number and spatial distribution of attacked trees. The presence of a strong “facilitation effect” suggests that log piles should be removed quickly in order to prevent outbreaks of bark beetles.

International audience; Maritime pine (Pinus pinaster Ait.) forests in the Aquitaine region, south-west France, suffered catastrophic damage from Storms Martin (1999) and Klaus (2009), and more damage is expected in the future due to forest structural change and climate change. Thus, developing risk assessment methods is one of the keys to finding forest management strategies to reduce future damage. In this paper we evaluated two approaches to calculating wind damage risk to individual trees using data from different damage data sets from two storm events. Airflow models were coupled either with a mechanistic model (GALES) or a bias-reduced logistic regression model, in order to discriminate between damaged and undamaged trees. The mechanistic approach was found to successfully discriminate the trees for different storms, but only in locations with soil conditions similar to where the model parameters were obtained from previous field experiments. The statistical approach successfully discriminated the trees only when applied to similar data as that used for creating the models, but it did not work at an acceptable level for other data sets. One variable, decade of stand establishment, was a significant variable in all statistical models, suggesting that site preparation and tree establishment could be a key factor related to wind damage in this region.; Les forêts d’Aquitaine de Pin maritime (Pinus pinaster Ait.) du sud-ouest de la France ont souffert de deux tempêtes majeures, Martin (1999) et Klaus (2009) et de nouveaux dégâts sont attendus dans le futur en raison de changements dans la structure forestière et à cause du changement climatique. C’est pour cela que le développement d’une méthode d’évaluation des risques liés aux tempêtes est une des clés pour développer de nouvelles stratégies d’aménagement pour réduire les risques à venir. Dans ce papier, nous avons évalué deux approches pour calculer le risque au niveau des arbres individuels en utilisant deux jeux de données provenant des deux dernières tempêtes. Des modèles de simulation atmosphériques ont été couplés soit à un modèle mécaniste de prévision des dégâts (GALES) soit à un modèle de régression logistique avec réduction du biais de façon à distinguer les arbres endommagés ou non. L’approche mécaniste permet de distinguer avec succès les catégories d’arbres lors des deux tempêtes mais seulement pour les conditions pédologiques des données qui ont servi à la calibration des paramètres du modèle. L’approche statistique permet elle aussi de discriminer les catégories d’arbres avec succès uniquement pour les conditions de calibration mais elle ne fonctionne pas pour d’autres jeux de données. Une variable, la décade d’installation des peuplements, est significative pour tous les différents modèles statistiques, suggérant que les techniques de préparation du sol et de régénération pourraient faire parties des facteurs clés en relation avec les dégâts dus au vent dans cette région.

Chapitre 7; International audience; Optical imaging, which makes use of the reflective properties of short wavelength solar radiation (400 to 2,500 nm), is the oldest source of spatial data on which its applications in forestry have been developed.

International audience; Measuring turbulent fluxes with the eddy covariance method has become a widely accepted and powerful tool for the determination of long term data sets for the exchange of momentum, sensible and latent heat, and trace gases such as CO2 between the atmosphere and the underlying surface. Several flux networks developed continuous measurements above complex terrain, e.g. AmeriFlux and EUROFLUX, with a strong focus on the net exchange of CO2 between the atmosphere and the underlying surface. Under many conditions basic assumptions for the eddy covariance method in its simplified form, such as stationarity of the flow, homogeneity of the surface and fully developed turbulence of the flow field, are not fulfilled. To deal with non-ideal conditions which are common at many FLUXNET sites, quality tests have been developed to check if these basic theoretical assumptions are valid. In the framework of the CARBOEUROFLUX project, we combined quality tests described by Foken and Wichura (1996) with the analytical footprint model of Schmid (1997). The aim was to identify suitable wind sectors and meteorological conditions for flux measurements. These tools were used on data of 18 participating sites. Quality tests were applied on the fluxes of momentum, sensible and latent heat, and on the CO2-flux, respectively. The influence of the topography on the vertical wind component was also checked. At many sites the land use around the flux towers is not homogeneous or the fetch may not be large enough. So the relative contribution of the land use type intended to be measured was also investigated. Thus the developed tool allows comparative investigations of the measured turbulent fluxes at different sites if using the same technique and algorithms for the determination of the fluxes as well as analyses of potential problems caused by influences of the surrounding land use patterns.

This letter evaluates the potential of wavelet-based texture modeling for the classification of stand age in a managed maritime pine forest using very high resolution panchromatic and multispectral PLEIADES data. A cross-validation approach based on stand age reference data is used to compare classification performances obtained from different multivariate models (multivariate Gaussian, spherically invariant random vector (SIRV)-based models, and Gaussian copulas) and from co-occurrence matrices. Results show that the multivariate modeling of the spatial dependence of wavelet coefficients (particularly when using the Gaussian SIRV model) outperforms the use of features derived from co-occurrence matrices. Simultaneously adding features representing the color dependence and leveling the dominant orientation in anisotropic forest stands enhances the classification performances. These results confirm the ability of such wavelet-based multivariate models to efficiently capture the textural properties of very high resolution forest data and open up perspectives for their use in the mapping of monospecific forest structure variables.