Monitoring seasonal changes of a mixed temperate forest using ERS SAR observations

Christophe Proisy, Eric Mougin Centre d’Etudes Spatiales de la Biosphère CNES / CNRS / UPS Bpi 2801 18 avenue E. Belin 31401 Toulouse Cedex 4 - France Fax: (33) 5.61.55.85.00 e-mail: [email protected]

Eric Dufrêne, Valérie Le Dantec Laboratoire d’Ecologie Végétale Université de Paris XI 91405 Orsay Cedex France

Transactions on Geoscience and Remote Sensing, 38(1) : 540-552, January 2000

ABSTRACT

Temporal variations of ERS-1/2 backscattering coefficients acquired over a mixed deciduous forest are analyzed aimed at relating the observed radiometric variations to changes either in the vegetation seasonal cycle or in the structural parameters. Overall, the results are somewhat pessimistic. Temporal σo plots show chaotic variations which are difficult to relate to the seasonal changes of forest parameters and particularly to the foliage dynamics. Furthermore, no distinction between stand types or between deciduous species is found to be possible and nearly identical temporal plots are observed for all the stands suggesting that the radar signatures are partly under the influence of nonforest parameters. Besides, the effect of meteorological events are difficult to evaluate. Discrimination between deciduous stands and conifers is nevertheless possible since the radiometric difference between the two species is about 1 dB. With an overall sensitivity to standing biomass of about 0.1 dB / 50 tons per hectare, ERS SARs can be considered as almost insensitive to biomass variations. For the young stands, the C-band response is found to be dominated by stand structure whereas the backscattering coefficient saturates for biomass values higher than 50 and 80 t DM ha-1 for deciduous and conifers, respectively.

Transactions on Geoscience and Remote Sensing, 38(1) : 540-552, January 2000

3

I.

INTRODUCTION

Recent studies, based on measurements of atmospheric CO2 concentrations and on the use of atmospheric transport model, have suggested that temperate ecosystems, particularly forests, might presently act as major carbon sinks [1-2]. Besides the forest area expansion at these latitudes, it is hypothesized that large quantities of carbon absorbed by terrestrial ecosystems would result from a human-induced effect of CO2 fertilization on forest growth. However, large uncertainties still exist for determining whether carbon sink or source occurs in temperate forests [3]. Forest models, combined with remote sensing observations, represent suitable tools for analyzing the different processes (photosynthesis and respiration) that influence the carbon budget. Remote sensing data are required a) to supply forest models with input parameters estimated over large areas, b) to validate forest model predictions. Simulations made with the forest process model CASTANEA [4], enabled two key parameters to be identified, namely the phenology of the forest canopy and the wood biomass, that might be estimated by remote sensing sensors. The phenology of a forest canopy can be described by the variation of its Leaf Area Index or LAI. In the case of a deciduous forest, this vegetation cycle is characterized by the length of the leafy period (from leaf-on to leaf-fall) and by the maximum LAI reached by the canopy. The objective for determining wood biomass is twofold; firstly, the amount of living woody tissues is required for calculating the respiration term of the carbon budget. Second, it can be a measure of the primary productivity of the forest when the biomass is estimated at two different dates.

Spaceborne Synthetic Aperture Radars (SARs) present considerable potentialities for monitoring forest ecosystems. Their faculty to penetrate natural canopies and their sensitivity

4

to the geometrical properties of vegetation allow forest structural attributes to be estimated [56]. Moreover, their sensitivity to the dielectric properties of vegetation and soil provides information about the moisture status of the ecosystem [7-8]. Monitoring temporal variations of the radar backscatter from forest canopies has become feasible over long periods of time with the launches of C-band SARs on board ERS-1, ERS-2 and RADARSAT in 1991, 1994 and 1995, respectively. The continuity of C-band observations from space is already planned with the foreseen launch of ENVISAT in 2000 [9]. Besides the obvious limitations of a single-frequency, single polarization and single-illumination instrument, the ERS series provides regular observations of forests on a monthly basis, thus enabling seasonal variations to be monitored. The potential of these temporal signatures for studying forest ecosystems was demonstrated in the case of boreal forests where frozen / thawing conditions could be detected [10-11]. On the other hand, past studies have also pointed out the sensitivity of the radar response to environmental parameters including temperature and rainfall, thus affecting the temporal signature of the backscattering coefficient σo [12-15]. Up to now, little attention has been given to the study and to the understanding of temporal signatures provided by spaceborne SARs, particularly for temperate deciduous forests apart from a few studies [1617].

The present study aims to evaluate the relevance of C-band VV polarized spaceborne SAR instruments for monitoring temporal changes in a mixed temperate forest ecosystem. Firstly, we want to examine the magnitude of variations in the radar backscattering in response to changing phenological, structural and environmental conditions. Second, we investigate the possibility to relate the observed radiometric variations to changes in the vegetation seasonal cycle or in the structural parameters. In this context, we analyze three years of ERS-1/2 multitemporal data acquired over a mixed deciduous-coniferous forest located near Paris

5

(France). The second section of this article describes the forest site and the ground data collection program. The analysis of the ERS-1/2 time series is presented in the third section. The fourth section deals with the relationships between the backscattering coefficient and forest parameters.

II.

SITE DESCRIPTION AND GROUND DATA COLLECTION

A.

The study site

The Fontainebleau forest, located south-east of Paris (48°25’N, 2°40’E), is a large mixed deciduous-coniferous forest extending over 17 000 ha (Fig. 1). The mean altitude is 120 meters. The region is characterized by a temperate climate with a mean annual temperature of 10.2°C. The mean air temperature of the coldest month (January) is 2.2°C whereas the warmest is 18.2°C (July). Freezing temperatures and snow cover can be observed in JanuaryFebruary. The mean annual precipitation of 720 mm is fairly well distributed throughout the year. Dominant species consist of oaks (Quercus petraea and Quercus robur), beech (Fagus sylvatica) and Scots pine (Pinus sylvestris). A few young plantations of Austrian pine (Pinus nigra) are also present. Co-dominant species are hornbeam (Carpinus betulus) and birch (Betula pendula). Most of the deciduous stands are located on a flat topography while coniferous species are generally found on the hilly parts of the forest and on rocky soils. Deciduous trees exhibit a well pronounced seasonality throughout the year, characterized by the leaf-on in April and the leaf-off in November, with a growing period of about 6 months. Maximum Leaf Area Index (LAI) is reached 3-4 weeks after leaf-on and ranges from about 1

6

to 8, depending on stand development. For coniferous trees, LAI shows minimum values in winter and maximum values in July, ranging from 1 to 8.

The forest is actively managed by the French National Office for Forestry (ONF). Silvicultural practices through thinning regularly modify the structure and species composition of the forest stands. Accordingly, the Fontainebleau forest shows a variety of stands, usually exceeding 10 ha, including the successive stages of stand development: seedlings, thickets, sapling stands, pole stands, mature forests and seed trees stands (see Fig. 2 for the deciduous stands). Furthermore, there also exist a variety of species compositions. Single-species stands are found as well as mixed deciduous and mixed deciduous-coniferous stands. The Fontainebleau forest therefore captures major characteristics of a managed temperate forest. Forest and soil maps including general tree parameters and a soil description in terms of soil texture are available. Besides, the Fontainebleau forest was selected as a test site within the frame of the European Multisensor Airborne Campaign (EMAC) organized by the European Space Agency (ESA) in 1994 [18].

B.

Ground data collection

Prior to SAR acquisitions, color-infrared aerial photographs followed by a field check, are used for the delineation of 56 test stands within the area covered by EMAC sensors. Among them, 21 are oak-dominant stands, 12 are beech-dominant stands, 18 are pine stands and 5 are mixed deciduous trees (oak and beech). Mean surface of the stands is about 13 ha. These stands are located on a flat terrain and they are homogeneous from a structural point of view (height, tree diameter, canopy closure). This sample of different stands represents the main types of single-species stands in terms of stage development, tree density, LAI and biomass. In

7

addition, the 5 mixed deciduous trees are included for comparison purposes as well as 2 clearcuts and one grassland.

Two kinds of ground measurements, namely inventory measurements and intensive measurements, were achieved during the period April 94 - February 97. Inventory measurements consisted of collecting structural forest parameters that can be considered as constant during the study period. This inventory work was performed during the 94-95 winter. Tables Ia-Ib show the forest parameters under consideration with their measured range of variation. Total basal area of each stand is derived from the measure of the circumference at breast height, C130. These structural parameters are used to estimate the standing biomass of trunks and branches from empirical equations used by foresters. For each stand, two models are used : one for the dominant species and another one for the understorey [19].

1)

Model for the deciduous stands : For the dominant species, the dry biomass of trunks

Bt in kg of dry matter per hectare (kg DM ha-1) is given by the product of the total volume of trunks Vt (m3 ha-1) and the dry density of wood ρs (kg m-3) : Bt = ρs Vt = ρs α G hdom

(1)

where G and hdom denote the basal area (m2 ha-1) of the dominant species and the stand height (m), respectively. The empirical parameter α slightly varies with species, stand age and silvicultural practices. Here, a value of 0.44 is retained. The dry density of wood ρs is equal to 570 and 550 kg m-3 for oak and beech, respectively. Based on published data [20], the biomass of branches Bbr (kg DM ha-1) is then estimated as: i= N

Bbr =0.002

∑ i =1

 C130 i   π 

3.265

(2)

8

where C130 (cm) is the circumference at breast height of the dominant species and N is the number of stems. Total dry biomass of the understorey Bund (kg DM ha-1) is given as the sum of the trunk biomass Bt and the branch biomass Bbr with :

i= N

Bt = 0.0762

∑ i =1

 C130 i   π 

2. 523

(3)

and with the biomass of branches Bbr calculated using equation (2). The same model is applied to the youngest stands (with C130 < 11 cm).

Finally, the total dry wood biomass of a given deciduous stand, BM, is given as the sum of the different wood components: BM = Bt + Bbr + Bund

2)

(4)

Model for the coniferous stands : Similarly to equation (1) for deciduous stands and

based on published data [21], total trunk biomass of coniferous trees Bt (kg DM ha-1) is given as : Bt = ρs α G hdom

(5)

with ρs and α equal to 460 kg m-3 and 0.42, respectively. The biomass of branches Bbr (kg DM ha-1) is derived from the following equation: Bbr = 2.658 exp[0.0027 C130]

(6)

9

Besides, total biomass of the understorey is estimated using equations (2) and (3). Similarly to equation (4), the total dry wood biomass of a given coniferous stand, BM, is given as the sum of the different wood components.

The associated uncertainties on the means of biomass are about 15% [19]. Overall, total standing biomass ranges from about 10 to 470 t DM ha-1 for the deciduous stands and from about 10 to 195 t DM ha-1 for the pine stands. Linear coefficients of determination r2 between structural parameters are given in Table IIa and IIb. As expected, a strong intercorrelation between forest parameters is observed, particularly for the coniferous trees. For these latter, the highest correlations are found between total biomass and basal area (r2 = 0.95) or tree height (r2 = 0.83). On the other hand, a slight correlation is found between C130 and basal area which may be due to the use of a simple linear model.

Intensive measurements were performed during the EMAC campaign (from April to June 1994) and during the 1996 growing season on 3 reference stands, labeled C08 (mature oak stand), H13 (pole beech stand) and P08 (mature pine stand). Measurements consisted of the determination of temporally varying forest parameters like the Plant Area Index (PAI) which is the sum of Leaf Area Index (LAI) and Wood Area Index (WAI). Table III illustrates the measured parameters and their range of variation for the beech stand H13. The PAI is estimated with the Licor LAI meter. For deciduous trees, it provides a good estimation of LAI when the foliage is fully developed [22]. For conifers, the LAI is obtained by the product of the PAI and an aggregation factor. Here, a value of 1.72 is used [23]. Total dry foliage biomass Bf is obtained from the product of the LAI (m2 m-2) and the Leaf Mass per Area or LMA (g m-2).

10

For the 1994 and 1995 seasons, only maximum PAI were measured in June. The daily evolution of PAI as well as the daily variation of the soil moisture content is performed using a forest process model [4]. Input meteorological data, including air temperature, wind speed and precipitation, are provided by the meteorological station of Fontainebleau.

III.

DESCRIPTION OF ERS-1/2 DATA

The ERS SARs provide measurements of the backscattering coefficient σo at 5.3 GHz (Cband) with VV polarization and at a mean incidence angle of 23°. ERS-1/2 Precision Images (PRI) data have been acquired on a monthly basis from April 94 to February 97. Measurements acquired during ascending and descending passes are not considered separately since they do not show significant differences. Finally, a total of 45 SAR PRI images have been acquired over the site and coregistered using the forest map. The backscattering coefficient σo of a stand under consideration is then derived from the pixel values following the calibration algorithm described in [24]. Resulting absolute calibration accuracy is about 0.3 dB. Stands smaller than 4 ha are not considered in the statistical analysis. Finally, 49 stands including 2 clearcuts are retained. For each stand, the number of pixels used to derive the mean associated backscattering coefficient is larger than 240 ensuring a radiometric resolution of ±0.5dB with a 90% confidence level. For a given date, the whole dynamic range of σo values within the entire forest does not exceed 4 dB. As well, during the study period, the σo dynamic range for a given development stage is about 5 dB compared with the 17 dB and 8 dB variations observed for the surrounding agricultural fields and the grassland, respectively (Fig. 3).

11

IV.

ANALYSIS OF ERS-1/2 TIME SERIES

The aim of this section is to analyze the ERS-1/2 temporal observations in relation with both the seasonal change of forest parameters and the variation of environmental factors.

A.

Influence of tree species

Temporal plots of σo values for the three main species are given in Fig. 4. Backscattering data correspond to σo values averaged over all the stands of each species. These plots are compared with that observed for the grassland. Several observations can be made. Firstly, the forest plots show chaotic temporal variations characterized by a succession of dips and jumps. The highest backscattered values of deciduous trees are usually observed during wintertime but a few high values are also found in summer (e.g. August 95). Nearly identical backscattering responses between deciduous species are observed with a mean value of about -8.0 dB. Conifers always show a lower backscatter of about 1 dB with the smallest differences between deciduous trees and pines found in summer. Moreover, the range of backscatter observed within all the stands does not exhibit any species and seasonal dependence. For both species, the largest dynamic range reaches about 5 dB. Second, similar temporal behaviors are observed between the deciduous species (r2 = 0.94) and between deciduous and conifers (r2 = 0.79). Surprisingly, there are also close similarities between the temporal plots of forest stands and that of the grassland (r2 = 0.80). Particularly, the highest backscattering values are coincident.

12

B.

Influence of development stage

Fig. 5 illustrates the σo temporal behavior of the main development stages for the oakdominant stands. On the whole, similar profiles are observed for all the stand types with coefficients of determination larger than 0.80 between the various stands. Mature stands exhibit the highest response whereas the lowest σo values are reached by the seedlings in winter or by the seed trees in summer. Maximum σo changes range between 1.8 dB (sapling / pole stands) and 3.3 dB (mature stands). Discrimination among mature stands and seedlings is high during the leafless period.

C.

Influence of environmental factors

Rainfall data, air temperature and wind speed recorded prior and during the acquisition of the SAR images, are compared with the corresponding backscattering coefficients. On the whole, there is no clear indication of any influence of the climatic parameters on the σo responses. Particularly, this is the case for rainfall (Fig. 6). High backscatter values are sometimes associated to a rain event (e.g. August 95). In other cases, a rain shower does not induce any backscatter enhancement (e.g. April 95). Also, precipitation can lead to a decrease in the backscattering (e.g. March 95). Accordingly, the coefficient of determination between rainfall data cumulated over different periods of time and σo is always low, usually smaller than 0.4. A weak correlation is nevertheless observed between seasonally cumulated rainfall and σo (Fig.7). Particularly, the dryness period occurring in spring 96 is well reflected in the σo values which exhibit a drop of about 1 dB. Moreover, no correlation is found either with wind

13

speed (r2 = 0.30) or air temperature (r2 = 0.12). For this latter, the absence of correlation is not surprising since no temperature below -2°C was recorded during the acquisition time of ERS images.

D.

Influence of temporally-varying forest parameters

At a seasonal scale, deciduous stands exhibit a pronounced phenological cycle characterized by the foliage expansion in spring reaching maximum LAI values by the end of May and followed by leaf-fall in autumn. As well, during the active season, leaf characteristics such as leaf water content, leaf dimensions and leaf mass per area are subject to large changes (Table III). On the opposite, conifers are evergreen trees and therefore do not show a marked seasonal cycle. New needles form in spring whereas old needles fall throughout all the seasons. Maximum LAI occurs in July. Differences as large as 50% are observed between LAI values measured during winter and summer. The difference in the needle water content is also large between young and old needles. Accordingly, strong changes in the dielectric properties of the forest components including understorey and soil surface occur throughout the different seasons. However, from the previous figures, there is no clear correspondence between the σo temporal plots and the variation of the main canopy parameters. For the beech dominant stand H13, Fig.8 displays the 1996 temporal plots of the backscattering coefficient (dB), the mean foliage water content (tons H2O ha-1), the mean branch water content (tons H2O ha-1) and the mean soil volumetric water content (%). For this later parameter, we have no measurement for the two winter acquisitions in January-February. Precipitation occurred the day before these dates and it can be reasonably assumed that the water content of the soil surface was higher than that we measured in March.

14

In January, the radar response is at a maximum of -7.1 dB. On the whole, from January to March, the decrease in the σ° plot follows the decrease of the soil water content. As this later reaches a maximum value during winter, high radar responses are always observed during this period. From April to May, the slight increase in the σ° plot can be related both to the increase in the branch water content and to the leaf emergence. From the end of April, the increase in σ° is coincident with the foliage development. During the leafy season, σ° slightly increases up to a maximum in June - July and then decreases till October. During this period, the overall σ° variation is of the order of 1 dB. However, the σ° change cannot be related to the variation of foliage water content which exhibits a continuous decrease from May. Finally, the drying and the fall of the leaves are followed in December by an increase in the σ° values.

E.

Interpretation of σ° seasonal changes

We use the Karam et al’s backscattering model [25] to simulate the temporal variation of the radar response observed over the beech canopy H13. The forest stand is modeled as a 3-layer medium above a rough surface. The upper layer contains all the leaves and 50% of the twigs. The medium layer contains the woody components (twigs, branches and the upper part of the trunk). The lower part of the trunk forms the lowest layer. The branches and trunks are treated as randomly oriented finite cylinders. Leaves are modeled as randomly oriented elliptic discs whose orientation is described by the Eularian angles (α, β, γ). All forest scatterers are assumed to be uniformly distributed in the azimuth direction. The probability density functions of the angle β in the zenith direction are given by the following equation [25]:

15

π  β − β m   P (β) = Acosn    2  β0 − β m  0

for β1 ≤ β ≤ β2

(7)

otherwise

where A is the normalization factor and where β0 , β1 , β2, βm and n are parameters defined in [25]. The third Eularian angle γ is set to 0°. Numerical simulations are performed using input parameters measured during the 1996 growing season. These parameters are given in Table IV. The dielectric constants for woody and leafy parts are estimated using the Ulaby and El Rayes’ model [26]. The Hallikainen et al’s model [27] is used to compute the soil dielectric constant.

Fig. 9 shows the comparison between simulated and experimental data. Individual contributions of the leaves, branches and forest floor are also indicated. On the whole, there is a good agreement between model predictions and experimental data. Particularly, the overall backscattering level is well simulated throughout the whole year. Differences between the model and the data lies between 0.1 and 1.3 dB. The analysis of the different contributions show that the backscatter from the soil and branches are comparable during wintertime when foliage is absent. As suggested above, the decrease in the backscattering from January to March is mainly due to the drying of the soil surface. However, uncertainties still exist for the winter period since no measurement of the soil moisture content was performed. It is suspected that the moisture values used in the simulation are slightly underestimated. Leaf emergence leads to an increase of about 2 dB during a short period when the leaves still contain a high percentage of water. Meanwhile, the contribution of the soil drops due to the high attenuation by the foliage. During summer, the slight decrease of the backscattering results from the slow drying of leaves. However, the highest values observed in June - July cannot be explained either by the changes in the forest parameters or by rainfall events. At C-

16

band, numerical simulations show that the yearly σo variations remain necessarily small and the contribution from foliage is therefore difficult to observe. It is mainly because the different contributions offset each other.

V.

RELATIONSHIPS BETWEEN σo AND FOREST PARAMETERS

In this section, we analyze the dependence of the backscattering coefficient σo on forest structural parameters, particularly standing biomass.

A.

Radar backscatter versus total biomass

Radar values are extracted from ERS-1/2 images and averaged within the 49 test stands. Fig. 10 illustrates the 3 - year averaged σo, as a function of total biomass. On the whole, there is a slightly positive relationship (r2 = 0.42) between σo and total biomass. The backscatter dynamic range is about 3.5 dB. There is also a marked difference between deciduous trees and coniferous trees. For the highest biomasses, the difference is about 1 dB between the two species. Fig. 11a and Fig. 11b display the response of the different deciduous stands and coniferous stands, respectively. For each species, the σo dynamic range is 2.5 dB. When the clearcuts are not considered, the dynamic range is only 1.3 dB for the deciduous trees, giving an overall sensitivity of about 0.1 dB / 50 t DM ha-1 ! Moreover, there is no experimental evidence for a greater sensitivity of the backscatter for the youngest deciduous stands due to the large scatter observed in the σo responses (Fig. 11a). Furthermore, no distinction between deciduous species (oak, beech or mixed) or stand types (e.g. seed tree and mature stands) can

17

be noticed. The same general observations can be made for the coniferous stands (Fig. 11b). Particularly, for the lowest biomasses, the large scatter in the σo response is attributed to effects of stand structure like the presence and dimensions of rows, which are related to forest practices. This effect is reduced when biomass increases.

The temporal variation of the σo / biomass relationship is illustrated in Fig. 12a and Fig. 12b for deciduous trees and for conifers, respectively. The indicated tendency lines and the associated equations correspond to means of σo responses acquired in winter, spring, summer and autumn. The clearcuts i.e. the nonforest stands are not taken into account in this analysis. For deciduous trees, the highest correlation is found in winter (r2 = 0.57) when the leaves are absent. In this case, the dynamic range estimated from the regression curve is at a maximum of about 1.3 dB. For the other seasons, the dynamic range is only 0.5 dB and the regression lines do not differ significantly. For conifers, the largest dynamic range of about 1.8 dB is observed in summer and in spring. Associated coefficients of determination are about 0.500.53. Temporal variations of the regression lines are more pronounced than those found for the deciduous stands.

B.

Radar backscatter versus forest parameters

Linear regression techniques are used to examine the relationships between σ° and the logarithm of the forest parameters. Table V summarizes, for deciduous and coniferous trees, the determination coefficients r2 and the standard error s of the different regression relationships. Here, only the significant relationships are shown, i.e. the winter relationship for

18

deciduous trees and the summer relationship for pine trees. Overall, the correlations are stronger for pine trees than for deciduous trees, with in all cases, large associated standard errors. When using these empirical relationships to estimate the forest parameters, the resulting mean errors are of the order of 50-60% for total biomass. Additionally, as expected, no correlation is found between σo and foliage biomass.

VI.

CONCLUSION

Temporal variations of ERS-1/2 backscattering coefficients acquired during 3 years over a mixed deciduous forest are analyzed aimed at relating the observed radiometric variations to changes either in the vegetation seasonal cycle or in the structural parameters. Overall, the results are somewhat pessimistic. Observed backscattering coefficients exhibit small changes throughout the year of the order of 2-3 dB compared with 17 dB found over agricultural fields. Furthermore, temporal σo plots show chaotic variations that are difficult to relate to the seasonal changes of forest parameters and particularly to the foliage dynamics. This is partly due to the strong scattering response of branches which probably masks the beginning and the end of the leafy cycle. As well, the different contributions (soil, branches and foliage) offset each other leading to small temporal variations in the backscattering level. During wintertime, the high backscattering values can be certainly related to a strong soil contribution. The discrimination between deciduous stands and conifers appears possible since the radiometric difference between the two species is about 1 dB. On the other hand, no distinction between stand types or between deciduous species is found to be possible and nearly identical temporal plots are observed for all the stands suggesting that the radar signatures are partly under the influence of nonforest parameters. If we exclude any

19

calibration problem of the instrument, seasonally averaged σo seems to be correlated to seasonally cumulated rainfalls. Particularly, the dryness period occurring in 1996 is well detected. However, the effect of meteorological events is difficult to evaluate. With an overall sensitivity to standing biomass of about 0.1 dB / 50 t DM ha-1 , ERS SARs can be considered as almost insensitive to biomass variations. For the young stands, the Cband response is found to be dominated by stand structure whereas the backscattering coefficient saturates for biomass values higher than 50 and 80 t DM ha-1 for deciduous and conifers, respectively.

This study demonstrates that C-band VV polarized signatures cannot be advantageously used for monitoring seasonal or structural changes of a managed mixed deciduous forest. Particularly, the foliage seasonal cycle cannot be detected as well as long term structural changes including the temporal variations of standing biomass. In the near future, the capabilities offered by multi-polarization and multi-incidence spaceborne SARs like the Advanced Synthetic Aperture Radar (ASAR) on board ENVISAT must be assessed. The observation of forest ecosystems should also benefit from the new possibilities offered by spaceborne interferometry techniques [28].

ACKNOWLEDGMENTS We would like to thank Myriam Legay from the French National Office for Forestry (ONF) for her help and for providing us with the Fontainebleau forest map. We also wish to acknowledge the numerous people who collected the ground data. The ground data collection was supported by CNES within the frame of the Preparatory Program for Earth Observation with SAR, by the French National Program for Remote Sensing (PNTS) and by the French Program for Environment (IGBP Ecosystèmes - Forêts tempérées and SEAH). The ERS-1 SAR images were provided by ESA within the frame of the EMAC-94 campaign and under the ERS-1/2 project (E. Mougin, AO2.F121). We thank Dr. M.A. Karam for providing us with his theoretical model. We are very grateful to Mike Wooding and Paul Mason from RSC for their co-operation.

20

REFERENCES

[1] P.P. Tans, I.Y. Fung, and T. Takahashi,’’ Observational constraints on the global atmospheric CO2 budget,’’ Science; vol. 247, pp. 1431-1438, 1990. [2] P. Ciais, P. P. Tans, M. Trolier, J.W.C. White, and R.J. Francey,‘’ A large northern hemisphere terrestrial CO2 sink indicated by the 13C/12C ratio of atmospheric CO2,‘’ Science, vol. 269, pp. 1098-1102, 1995. [3] R.K. Dixon, S. Brown, R.A. Houghton, A.M. Solomon, M.C. Trexler, and J. Wisniewski,’’ Carbon pools and flux of global forest ecosystems,’’ Science, vol. 263, pp. 185190, 1994. [4] E. Dufrêne, F. Badeck, D. Epron, A. Granier, V. Le Dantec, and B. Saugier,’’ Testing a mechanistic carbon and water balance model against CO2 and H2O fluxes measured by eddy covariance above a beech forest,’’ to be submitted to Agricultural and Forest Meteorology, 1999. [5] E. Kasischke, N.L. Christensen, and L.L. Bourgeau-Chavez,’’ Correlating radar backscatter with components of biomass in Loblolly pine forests,’’ IEEE Trans. Geosci. Remote Sensing, vol. 33, no 3, pp.877-894, 1995.

[6] M.C. Dobson, F.T. Ulaby, L.E. Pierce, T.L. Sharik, K.M., Bergen, J. Kellndorfer, J.R. Kendra, E. Li, Y.C. Lin, A. Nashashibi, K. Sarabandi, P. Siqueira, ‘’Estimation of biophysical characteristics in northern Michigan with SIR-C/X-SAR,’’ IEEE Trans. Geosci. Remote Sensing, vol. 33, pp.877-894, 1995.

[7] N.H.F. French, E.S. Kasischke, L.L. Bourgeau-Chavez, P. Harrell, and N.L. Christensen,’’ Monitoring variations in soil moisture on fire disturbed sites in Alaska using ERS-1 SAR imagery,’’ Int. J. Remote Sensing, vol. 17, n°7, pp. 3037-3053, 1996.

21

[8] K.C. McDonald, M.C. Dobson, and F.T. Ulaby,’’ Using MIMICS to model L-band multiangle and multitemporal backscatter from a walnut orchard,’’ IEEE Trans. Geosci. Remote Sensing, vol. 28, no. 4, pp.477-491, 1990.

[9] ESA, ‘’ENVISAT-1. Mission and system summary,’’ ESA Publications, March 97. [10] E. Rignot, J.B. Way, K. McDonald, L. Viereck, C. Williams, P. Adams, C. Payne, W. Wood, and J. Shi,’’ Monitoring of environmental conditions in taiga forests using ERS-1,’’ Remote Sens. Environ., vol. 49, pp. 145-154, 1994.

[11] J.T. Pulliainen, P.J. Mikkelä, M.T. Hallikainen, and J.-P. Ikonen, ‘’Seasonal dynamics of C-band backscatter of boreal forests with applications to biomass and soil moisture estimation,’’ IEEE Trans. Geosci. Remote Sensing, vol. 34, pp. 758-769, 1996. [12] J.B. Way, J. Paris, E. Kasischke et al.,’’ The effects of changing environmental conditions on microwave signatures of forest ecosystems : preliminary results of the March 1988 Alaskan aircraft SAR experiment,’’ Int. J. Remote Sensing, vol. 11, n°7, pp. 1119-1144, 1990. [13] M.C. Dobson, K. McDonald, F.T. Ulaby, and T. Sharik, ‘’Relating the temporal change observed by AIRSAR to surface and canopy properties of mixed conifer and hardwood forests of northern Michigan,’’ in Proc. 3rd Airborne Synthetic Aperture Radar (AIRSAR) workshop, May 20-24, JPL Publications 91-30, pp. 34-43, 1991. [14] M. Moghaddam, S. Durden, and H. Zebker, ‘’Effects of environmental change on radar backscatter in the Oregon transect,’’ in Proc. IGARSS’93 Symp., Tokyo, Japan, 18-21 August, pp. 580-582, 1993. [15] K.J. Ranson and G. Sun, ‘’Effect of environmental temperatures on SAR forest biomass estimates, ‘’ in Proc. IGARSS’97 Symp., Singapore, August, pp. 1722-1724, 1997.

22

[16] G. Kattenborn, E. Nezry, G. De Grandi, A.J. Sieber,’’ High resolution detection and monitoring of changes using ERS-1 time series,’’ in Proc. Second ERS-1 Symposium-Space at the service of our Environment, Hamburg, Germany, October 11-14, pp. 635-642, 1993.

[17] F.M. Seifert, H. Kietzmann, and M. Zink, ‘’Forest monitoring with SIR-C/X-SAR,’’ in Proc. International Symposium on Retrieval of bio- and geophysical parameters from SAR data for land applications, Toulouse, 17-20 Oct, pp. 161-166, 1995.

[18] E. Dufrêne, V. Le Dantec, V. Demarez, J.P. Gastellu-Etchegorry, G. Marty, E. Mougin, C. Proisy, B. Lacaze and S. Rambal,’’ Remote sensing of the Fontainebleau forest during EMAC-94 : Objectives and data collection program,’’ Proc. EMAC 94/95 final results, ESTEC, Noordwijk, The Netherlands, 14-16 April, pp. 91-95, 1997. [19] V. Le Dantec, ‘’Utilisation de la télédétection multi-spectrale en vue de modéliser le bilan carboné d’un massif forestier,’’ DEA Ecologie Générale et Production Végétale, Université de Paris-Sud XI, 35p, 1995. [20] H.H. Bartelink, ‘’Allometric relationships for biomass and leaf area of beech (Fagus sylvatica L.),’’ Ann. Sci. For., no. 54, pp. 39-50, 1997. [21] P. Vanninen, H. Ylitalo, R. Sievänen, and A. Mäkelä, ‘’ Effects of age and site quality on the distribution of biomass in Scots pine (Pinus sylvestris L.),’’ Trees, no. 10, pp. 231-238, 1996. [22] E. Dufrêne and N. Breda, ‘’Estimation of deciduous forest leaf area index using direct and indirect methods,’’ Oecologia, 935, pp. 1-7, 1995. [23] K.S. Fassnacht, S.T. Gower, J.M. Norman, and R.E. McMurtrie, ’’A comparison of optical and direct methods for estimating foliage surface area index in forests,’’ Agricultural and Forest Meteorology, 71, 183-207, 1995.

23

[24] H. Laur, P. Bally, P. Meadows, J. Sanchez, B. Schaettler, and Lopinto E., ‘’Derivation of the backscattering coefficient σo in ESA ERS SAR PRI products,’’ ESA Document No : ESTN-RS-PM-HL09, May 1997.

[25] M.A. Karam, F. Amar, A.K. Fung, E. Mougin, A. Lopes, D.M. Le Vine, and A. Beaudoin,’’ A microwave polarimetric scattering model for forest canopies based on vector radiative transfer theory,’’ Remote Sens. Environ., 53, pp. 16-30, 1995. [26] F. T. Ulaby. and M. A. El-Rayes, “ Microwave Dielectric Spectrum of Vegetation - Part II : Dual Dispersion Model ”, IEEE Trans. Geosci. Remote Sensing, vol. 25, no. 5, pp. 550557, 1987 [27] M. T. Hallikainen, F. T. Ulaby, M. C. Dobson, M. A. El-Rayes and L-K. Wu, ”Microwave Dielectric Behavior of Wet Soil - Part I : Empirical Models and Experimental Observations”, IEEE Trans. Geosci. Remote Sensing, vol. 23, no. 1, pp. 25-34, 1985. [28] J.I.H. Askne, P.B.G. Dammert, L.M.H. Ulander, and G. Smith, “C-band Repeat-Pass Interferometric SAR Observations of the Forest”, IEEE Trans. Geosci. Remote Sensing, vol. 35, no. 1, pp. 25-35, 1997.

24

Table Ia: Measured range of variation of structural forest parameters for the deciduous stands according to their respective development stage.

Parameter Tree Density (N ha-1) Trunk C130 (cm) 2 -1 Basal Area (m ha ) Tree Height (m) Crown Height (m) -1 Total Biomass (t DM ha ) -1 Trunk Biomass (t DM ha ) Branch Biomass (t DM ha-1) Understorey Biomass (t DM ha-1)

Seedlings / Thickets 5980 – 17525 8 3 – 12 5–8 2–3 14 – 47 13 – 42 1–4 0

Mature Sapling Pole stands stands Stands 1990 - 5590 700 - 4950 300 - 1480 12 - 23 19 - 46 37 - 87 9 - 25 15 - 24 20 - 40 8 - 16 13 - 22 20 - 39 1–2 2 - 13 12 - 25 40 - 135 105 - 145 140 - 470 35 - 110 90 -115 110 - 315 5 - 20 15 - 30 30 - 180 0 0 - 120 0 - 55

Seed tree stands 20 – 415 57 - 204 7 –22 28 - 41 16 - 21 100 – 380 60 – 225 40 – 155 0 - 45

Table Ib: Measured range of variation of structural forest parameters for the coniferous stands according to their respective development stage.

Parameter -1

Tree density (N ha ) Trunk C130 (cm) 2 -1 Basal Area (m ha ) Tree Height (m) Crown Height (m) Total Biomass (t DM ha-1) -1 Trunk Biomass (t DM ha ) Branch Biomass (t DM ha-1) -1 Understorey Biomass (t DM ha )

Sapling Pole stands Mature Seedlings / Thickets Stands stands 1480 – 1700 1950 - 4010 1700 450 - 1320 11 – 16 14 - 40 37 45 - 91 2–3 10 - 30 26 26 - 41 4 8 - 13 16 16 - 26 1 4–8 9 8 - 12 7 - 10 30 - 100 95 100 - 195 2–3 13-79 58 78-154 5-7 13-18 13 13-22 0 0-1 23 0 - 50

Seed tree stands 85 - 230 82 - 134 13 - 17 22 - 26 8 - 10 80 - 85 62-63 11-13 0-10

25

Table IIa: Linear coefficients of determination (r2) between the forest structural parameters for the deciduous trees. The minus sign indicates a negative relationship. C-130 Tree Density

-0,50

C-130

Tree Height -0,70

Basal Area -0,33

Crown Height -0,69

Total Biomass -0,50

Trunk Biomass -0,47

Branch Biomass -0,53

0,61

-0,13

0,61

0,21

0,10

0,36

0,54

0,91

0,81

0,78

0,83

0,41

0,86

0,91

0,75

0,68

0,64

0,71

0,99

0,97

Tree Height Basal Area Crown Height Total Biomass Trunk Biomass

0,92

Table IIb: Linear coefficients of determination (r2) between the forest structural parameters for the coniferous trees. The minus sign indicates a negative relationship. C-130 Tree Density C-130 Tree Height Basal Area Crown Height Total Biomass Trunk Biomass

-0,88

Tree Height -0,80

Basal Area -0,29

Crown Height -0,61

Total Biomass -0,49

0,85

-0,34

0,65

0,48

0,55

0,29

0,92

0,83

0,84

0,62

0,82

0,95

0,93

0,85

0,86

0,86

0,75

0,72

Trunk Branch Biomass Biomass -0,56 -0,17

0,98

0,83 0,80

26

Table III: Measured range of variation of temporally varying forest parameters for 2 reference stands (1996 growing season).

Parameter Soil Soil moisture content (cm3 cm-3) Vegetation Plant Area Index (m2 m-2) Leaf length (cm) Leaf width (cm) Leaf thickness (mm) -2 Leaf Mass Area (g m ) -1 Leaf Water Content (g g ) -1 Branch water content (g g ) -1 Bark water content (g g )

Oak (C08)

Beech (H13)

Scots Pine (P08)

0.1 - 0.45

0.1 - 0.6

0.2 - 1.2

1-6 2.6 - 9.85 1.6 - 6.6 0.16 - 0.27 52 - 87 0.45 - 0.7 0.42 - 0.55 0.1 - 0.41

1-7 4 - 8.4 2.4 - 5.1 0.1 - 0.125 46 - 71 0.45 - 0.7 0.4 - 0.5 0.4 - 0.67

1.5 – 3 2.2 – 6.2 1.7 - 1.8 140 – 220 0.48 - 0.78 0.49 - 0.56 0.1 - 0.36

27

Table IV: Input forest parameters used in the simulations (H13 stand).

Top Layer (height = 3m) Leaves Twigs Middle Layer (height = 5.5m) Twigs Secondary Branches Primary Branches Trunk Bottom Layer (height = 6m) Trunk Ground Surface

Sand 65%

β 0, β1, β2, βm, γ, n

Length (cm) 0 - 7.74 0.75

Width (cm) 0 - 4.76 0.4

Density (#/m3) 0 - 1230 27.2

90°, 0, 90°, 0°, 0°, 2 -20°, 0, 90°, 35°, 2

0.75 2.5 2.5 5.5

0.4 1.5 2.5 3.8

27.2 0.2 0.3 0.05

-20°, 0, 90°, 35°, 2 -180°, 0°, 90°, 30°, 2 -180°, 0°, 90°, 30°, 1 -20°, 0°, 20°, 0°, 2

6

8

0.05

-20°, 0°, 20°, 0°, 2

Clay 18%

rms Roughness 1.22 cm

Surface Correlation Length 5 cm

28

Table V: Coefficients of determination r2 and standard errors s (expressed in arithmetic units) associated to the relationship between σo and the logarithm of forest parameters for deciduous trees (wintertime) and conifers (summertime). The clearcuts are not considered in the statistical analysis.

Parameter -1

Tree Density (N ha ) Trunk C130 (cm) Tree Height (m) Basal Area (m2 ha-1) Crown Height (m) Total Biomass (t DM ha-1) Trunk Biomass (t DM ha-1) -1 Branch Biomass (t DM ha )

Deciduous r2 s -0.13 500 0.20 21 0.53 10 0.42 8 0.41 8 0.57 150 0.55 90 0.55 87

Coniferous r2 s -0.23 350 0.53 25 0.69 8 0.55 17 0.64 5 0.53 75 0.61 130 0.40 4

29

LIST OF FIGURES

Fig. 1: Location of the Fontainebleau forest. Fig. 2: Schematic representation of the different successional stages of the deciduous stands. Fig. 3: Mean ERS-1/2 backscattering coefficients and associated dynamic range for forest stands compared with other vegetation types during the study period (April 94 - February 97). Fig. 4: Temporal plots of mean ERS-1/2 backscattering coefficients for forest stands and grassland (April 94 - February 97). Fig. 5: Temporal plots of ERS-1/2 mean backscattering coefficients for the different types of oak stands (April 94 - February 97). Fig. 6: Temporal plots of ERS-1/2 backscattering coefficients for the three reference stands. Rainfalls cumulated over a 24 hour period prior to ERS acquisitions are also indicated (April 94 - December 96). Fig. 7: Seasonal trend of ERS-1/2 backscattering coefficient for deciduous and coniferous stands compared with the seasonal variation of precipitations. Mean observed backscattering coefficients for the considered period are indicated with dot lines (Sp: spring, Su: summer, Au: Autumn, Wi: winter). Fig. 8: Temporal plots of ERS-1/2 σo compared with the seasonal variation of foliage water content (t ha-1), branch water content (t ha-1) and volumetric soil water content (%), for the beech reference stand H13 (January 96 - December 96) Fig. 9: Comparison between simulated σo and experimental data for the H13 stand. Fig. 10: Relationship between ERS-1/2 backscattering coefficient and total standing biomass. Fig. 11a: Relationship between ERS-1/2 backscattering coefficient and total standing biomass for deciduous trees.

30

Fig. 11b: Relationship between ERS-1/2 backscattering coefficient and total standing biomass for pine trees. Fig. 12a: Seasonal variation of the relationships between ERS-1/2 backscattering coefficient and total standing biomass for deciduous trees. Fig. 12b: Seasonal variation of the relationships between ERS-1/2 backscattering coefficient and total standing biomass for pine trees.

31

32

clearcut / seedlings

thickets

40 m

pole stands

mature forest

40 m

seed trees

seed trees / seedlings

40 m

33

0 Crops 

−2

Backscattering Coefficient (dB)

−4

Seedlings & Thickets

−6 −8

Clearcuts

ο

Grassland

•∆ •

−10











Seed Sapling & Mature Trees Pole Stands Stands







•



•∆

•

∆

∆ 







 

−12





ο

−14 −16 −18 −20





ο Crops and Grassland

•

Deciduous trees

∆ Pine Trees

34

−6

Oaks (−) Beeches (−.) Pine Trees (...)

Backscattering Coefficient (dB)

−7

−8

−9

−10

−11

−12

−13

Grassland −14 1994 1995 1996 1997 J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D J F M

35

Oak trees −6

Backscattering Coefficient (dB)

−7

−8

−9

−10

Seedlings −.

Sapling & Pole Stands ...

Mature Stands −

Seed trees −

1994

1995

1996

1997

J FMAM J J A SOND J FMAM J J A SOND J FMAM J J A SOND J FM

36

−6

• Oaks C08 ♦ Beeches H13

−7

•♦ • ♦ • ∆ ∆ • ♦ ∆ • ∆ ∆ ♦ ∆ ∆•

♦ • ♦

∆ ∆

• ♦ ♦ • ♦ ♦ • • ♦ ♦ ♦ ♦ • • • ♦ ♦ ♦• ♦ ♦ •♦ • ♦ • ♦ • • • • ∆ ♦ • ♦ • ∆ ♦ •• ♦ ♦ ♦ ♦ • ♦♦ • ♦ ∆ ∆ ∆ ∆ ∆ ∆ ∆ ∆ ∆ ∆ ∆ ♦ ∆ ∆ ∆

∆ ∆

∆

♦ ♦

∆∆

∆

∆ −10

−11

•

•

•

−9

• •

∆

∆

∆

∆

∆

∆

20

∆

∆

∆

10

0

1994 1995 1996 J FMA M J J A S ON D J F MA M J J A S ON D J F MA M J J A SON D

One−day Cumulated Rainfall (mm)

Backscattering Coefficient (dB)

∆ Pine Trees P08

−8

•

••

37

−6

• Deciduous trees ∆ Pine trees

• • •

−8

•

•

•

•

•

•

• •

∆

−9

∆

∆

∆

400

∆

∆

∆

∆

∆

∆

200

−10

∆

−11

0 Sp94

Su94

Au94

Wi95

Sp95

Su95

Seasons

Au95

Wi96

Sp96

Su96

Au96

Cumulated Rainfall (mm)

Backscattering Coefficient (dB)

−7

38

Backscattering Coefficient (dB)

Beech trees − H13 −6

↓

•

−7

− − − − − Leafy Period − − − − −

•

• −8

−9

•

•

σ°

•

• •

↓

•

••

••

−10

Water Content (tH20/ha)

15

0.6

10

0.4

Foliage Water Content

••••• • •••••••••••••••••••••••••• •••••••••• • ••••••• • • • • •• • Volumetric Soil Water Content •• 1996

5

0

J

F

M

A

M

J

J

A

S

O

0.2

0

N

D

Soil Water Content (cm3/cm3)

Branch Water Content

39

−5

•

ERS−1/2 data

•

• ••

Backscattering Coefficient (dB)

•

•

•

•

•

••

••

•

−10

∇ ♦

♦♦ ♦

♦ ∇

♦ ∇

• ♦

−15

•

•

♦ ♦

♦

∇ • ∇ ∇ ∇∇ ∇

∇

•

•

♦

♦

∇

∇

∇ ♦

♦♦ ∇ ♦∇ • ∇

•

−20

•



Total

• Leaves

♦ Branches

∇ Soil

1996 J

F

M

A

M

J

J

A

S

O

N

D

40

41

42

43

44