MULTITEMPORAL ERS AND ENVISAT IMAGERY FOR THE ESTIMATION OF THE REFORESTATION PROCESS OF BURNED AREAS

Transcription

1 MULTITEMPORAL ERS AND ENVISAT IMAGERY FOR THE ESTIMATION OF THE REFORESTATION PROCESS OF BURNED AREAS F. Catalucci (1), F. Del Frate (1), A. Minchella (1), M.Paganini F (2) (1) Tor Vergata University - Dipartimento di Informatica Sistemi e Produzione Via del Politecnico, 1 I Rome, Italy - delfrate@disp.uniroma2.it (2): ESA/ESRIN Via G. Galilei, I Rome, Italy, marc.paganini@esa.int ABSTRACT In this study the potentialities of ERS/ENVISAT-SAR multitemporal data for the analysis of the reforestation process over burned areas are investigated. The summer 2000 fire devastating a large part of the Castel Fusano pinewood (near Rome) is the selected test case. A first objective is to evaluate the capability of detecting the boundaries of the burned area only using SAR imagery and considering optical data (Landsat) as ground truth. A second one is to retrieve information on the stage of the reforestation process from multitemporal backscattering signature. To this purpose both areas within and out of the fire scar have been considered and their SAR returns have been examined over several passes of the satellites before and after the event. 1. INTRODUCTION According to personnel working in environmental agencies and state institutions the rate of biomass regrowth over burned areas can be a crucial factor for the damage assessment related to a forest fire event. In some cases the reforestation process is very fast and after two years the burned areas will be completely repopulated. In other cases such a process can take even decades so that the environmental and economic impact of the fire event is much stronger. The knowledge on the repopulation capability over burned areas of different types of vegetation is so far sparse and limited to local experiences. Satellite data can alternatively offer the possibility of a more sistematic investigation on the topic. Indeed, previous studies have shown how measurements from spaceborne payloads may be used either for the fire impact assessment or for the characterization of the forest recovery from the fire [1] [5]. As far as SAR measurements are concerned, different factors stemming from a fire event may be potential causes of backscattering intensity variations, even at C band. For example changes in the dihedral scattering between tree trunks and the ground surface, variations of the influence of the soil moisture and soil roughness in the backscattering, variations of the water content and of the biomass characterizing the canopy and the understory. Moreover, significant information can also be provided by the measured level of coherence [6]. Indeed, over burned areas, volume backscattering significantly decreases and is replaced by surface backscattering which allows a much better correlation, resulting in greater coherence, between the two SAR images taken over the considered area. In this work we considered the very dramatic fire event which occurred in the Castel Fusano pinewood, located few kilometers away from Rome main urban area, on July 2000 [7],[8]. More than 250 ha went burned during the fire, modifying impressively the forest local environment. The investigation presented here focuses on the potentialities of multitemporal ERS-SAR images, eventually enhanced by new Envisat-ASAR imagery, either for the detection of fire scar or to derive some index significant for the monitoring of the recovery processing. 2. THE TEST AREA The Castel Fusano pinewood belongs to the state natural reserve of the Roman Coast, which is located to the south-west of the delta of the River Tevere, few kilometres away from the main Rome urban center. The whole park extends over an area of approximately ha and is characterized by the presence of different types of surroundings, typical of the coastal area of the Mediterranean. The arboreal vegetation is dominated by pines and oaks. Moreover, a large part of the Park is covered by the tall Mediterranean bush. The pinewood occupies about 6% (1000 ha) of the entire reserve and its first species were introduced by the coast starting from 18 th century. A view of the pinewood is reported in Fig. 1. Until the 3 rd of July, 2000, the pinewood was only slightly damaged by not Proc. of the 2004 Envisat & ERS Symposium, Salzburg, Austria 6-10 September 2004 (ESA SP-572, April 2005)

2 Fig. 1: A view of Castel Fusano pinewood before the fire event particularly serious frequent fires. Nevertheless on 3rd and 4th of July 2000, the pinewood went severely burned because of some fires that ruined about 350 ha of the wood, about 250 ha of wood were destroyed by the fire, while other 100 ha were seriously damaged (Fig. 2). Particularly devastating was the 4th of July fire, which had more than one fire point in the South- West part of the pinewood, where it borders the Castelporziano Presidential Estate. Soon after the fire, Municipality of Rome has decided a plan for the reclamation of the area and for restoring the local ecosystem. In the manner provided for by the plan some main actions consist in the cutting and removal of arboreus vegetation irremediably damaged (Fig. 3), in the singling out the domestic pines for which spontaneous recovery was still possible, in the replantation of the domestic pines and of the other most ecologically relevant types of vegetation. Moreover, experimental areas particularly dedicated to the monitoring of the only spontaneous regrowth processes have been identified. Fig.2: An area of the pinewood devasted by the fire 3. METHODOLOGY A significant set of ERS-SAR images has been collected before and after the fire event consisting in an overall number of 32 satellite passes. The images have been chosen according to some criteria. We remind that the fire event occurred on 4 July The considered time window goes from May 1999 to October 2003 so it comprises one year before the fire event and three years after. Most of the images belong to the same frame and track to facilitate coregistration operations. Fig. 3: Cutting of arboreus irremediably damaged vegetation As an example, in Fig. 4 the full image taken on 8/7/2001is illustrated, the square box indicates the area of interest, the city of Rome is clearly visible on the north-east direction. For each pass, investigation on meteorological condition have been carried out. These latter are indicated in the rightmost two columns of Table 1 where the complete list of the considered images is reported. In particular the UP column shows

3 the number of days occurring between the last rainy day and the day of the measurement ( - means that such number was greater than 7). One Landsat image, taken on 25/09/2000, has been also analysed for a comparison of the results on the detection of the fire scar. The images have been processed by means of BEST, the sotware tool released by ESA. Main processing consisted in the extraction of the regions of interest, calibration, coregistration and georeferencing. This last step was necessary for drawing in the images the edges, originally on land maps, characterizing the different types of intervention scheduled by the municipality plan. For the purpose of our study we decided to consider 4 different regions of interest: A region belonging to an area of the pinewood which was not damaged by the fire event (about 37 ha) A region within the area struck by the fire but essentially consisting of bare soil (about 6.1 ha) A region destroyed by the fire where the regrowth process started completely spontaneous with no man-made actions (about 5.8 ha) A region which underwent artificial interventions of reclaiming and replantation (about 16.8 ha) Localization of the described regions on a SAR image is shown in Fig. 5 with different colors. Fig. 4: ERS-SAR image taken on 8/7/2001 Fig. 5: The 4 analyzed sub-areas of interest: not burned pinewood (green), bare soil (orange), burned with artificial interventions (red), burned with only spontaneous regrowth (light-blue) Table 1: List of the images considered for the study Date Days from fire event Satell. Orbit Frame Weather condition 01/05/ ERS Cloudy 4 04/07/ ERS Clear 3 16/10/ ERS Misty 1 17/10/ ERS Cloudy 2 26/11/ ERS /11/ ERS /01/ ERS Cloudy 0 05/02/ ERS /04/00-80 ERS Clear 0 14/05/00-51 ERS Cloudy 5 18/06/00-16 ERS Cloudy 2 07/07/00 +3 ERS Clear - 01/10/ ERS Cloudy 0 05/11/ ERS Cloudy 2 10/12/ ERS Cloudy 7 14/01/ ERS Cloudy 5 18/02/ ERS Cloudy - 29/04/ ERS Clear 2 08/07/ ERS Cloudy 6 16/09/ ERS Cloudy 2 25/11/ ERS Cloudy 6 05/01/ ERS Clear 5 14/04/ ERS Rainy 1 23/06/ ERS Clear - 28/07/ ERS Clear - 01/09/ ERS Cloudy 1 10/11/ ERS Clear 2 19/01/ ERS Clear 1 30/03/ ERS Mist - 08/06/ ERS Cloudy - 17/08/ ERS Clear - 26/10/ ERS Clear 3 UP

4 4. RESULTS 4.1 Fire scar detection To analyse the capability of detecting the fire scar with the SAR data we considered the image taken by Landsat (25/9/2000) as ground truth. Indeed, Fig. 6 shows how the scar is easily detectable by using Landsat 7, 4 and 3 bands. By chance, the first ERS pass after the fire event was very early, on July the 7 th. Though a general lower value of backscattering can be visible in the burned area, this is not clearly discernible in the SAR image which is reported in Fig. 7. For sake of comparison, in Fig. 8 a SAR image obtained by averaging all the available passes preceding the event is also shown. The lower value of the backscattering should be given to the fact that the return is dominated by the soil contribution. In summer season this is characterized by low values of moisture and, in turn, of backscattering. The same type of hypothesis may be considered for a comment of figures 9 and 10, given by averaging on two different sets of multitemporal images. Fig. 8: Average image obtained considering only available images before the fire event SAR Fig. 9: Average image obtained considering only wintertime images after the fire event Fig. 6: Landsat image of the test area (bands 7, 4, 3) Fig. 7: SAR image taken on 7/7/2000 Fig. 10: Average image obtained considering only summertime images after the fire event

5 The image of Fig. 9 derives from the averaging of the first summertime dates available after the fire event, conversely, Fig. 10 is the result of the average of the first wintertime images taken after the event. In both figures, we again can observe the significant effect of the soil moisture on the return corresponding to the burned area. In summertime soil moisture values are lower and, as seen before for a single image, this provokes a lower value of the backscattering (about 2 db) with the respect to the surrounding not-burned area, where the volume scattering of the trees dominates. The behaviour is opposite during wintertime, when an higher soil moisture makes the backscattering of the pixels corresponding to the fire scar higher (more than 1 db) than that of those belonging to the surrounding areas. 4.2 Regrowth process monitoring The influence of the soil moisture on the backscattering seems to be significant also for understanding how the regrowth process may be monitored by multitemporal measurements. In fact, as the reforestation goes on in the burned area, the corresponding backscattering tends to be less affected by the surface scattering effects typical of the soil, becoming more similar the volume scattering typical of the parts of the pinewood not affected by the fire. This can be noted in Fig. 11 where four different lines, corresponding to the areas of interest described in section 3, are plotted. Different comments are suggested by the figure. The first one is on the behaviour of the bare soil field. It is clear how the seasonal cycle of the soil moisture drives the behaviour of the backscattering in the five considered years. Particularly, it is interesting to observe the 5 peaks of low values, regularly occurring in the summer season, when the dieletric characteristics of the soil are described by low values of the soil moisture and, hence, of the dielectric constant. On the other hand, the backscattering time sequence of the pinewood region not affected by the fire event is much more stable with smaller fluctuations which may be scarcely correlated to the soil moisture values. The time behaviour of the other two regions is rather similar. In fact, for both of them, we can separate among three different phases. Before the fire event they tend to be closer to the unburnt pinewood. Right after the event closer to the bare soil. In the period corresponding to the last two years, the similarity to bare soil seems to decrease. The previous results suggest a possibility for monitoring the pinewood regrowth in the burned regions. This might consist in tracking how much the -6-7,5 Backscattering coefficient -9-10, , , /11/03 01/08/03 01/05/03 01/02/03 01/11/02 01/08/02 01/05/02 01/02/02 01/11/01 01/08/01 01/05/01 01/02/01 01/11/00 01/08/00 01/05/00 01/02/00 01/11/99 01/08/99 01/05/99 Unburnt Burnt artificial Bare soil Burnt spontaneus Fig. 11: Time behaviour of backscattering corresponding to the 4 types of considered areas

6 backscattering behaviour of the regions is similar to the backscattering of the bare soil. In particular, a measure of this similarity could be represented by the evaluation, in the region to be monitored, of the backscattering excursion from the positive peak (taken during the winter season) to the negative peak (taken during the summer season) and comparing it to the one corresponding to the bare soil area. An exercised like that has been applied to our data set. Four annual peak-to-peak transitions (occurring in a time window from November 1999 to June 2003) of backscattering have been considered. The backscattering difference for each transition has been also calculated for bare soil, for not burned region and for the average of the two burned region. The results are shown in table 2 where the values of the first line correspond to the only transition considered before the fire event, the second line to the transition right after the fire event, the third and the fourth line to the transitions occurring in the following two years. We see that the values corresponding to the bare soil transitions are the greatest. The values of the not burned surface are lower and seem to be rather uncorrelated with those of the bare soil. As far as the burned areas are concerned we see that, before the fire event, the difference with the bare soil is rather high (4 db), soon after the fire event is small (0.4 db), but it starts progressively to increase (0.9 and 1.4 db) in the following two years. Table 2: Peak-to-peak annual transitions for different types of region of interest soil unburnt burnt Trans. 1 (db) Trans. 2 (db) Trans. 3 (db) Trans. 4 (db) CONCLUSIONS In this study a multitemporal analysis of the backscattering coefficient measured over the Castel Fusano pinewood, partially destroyed in a fire event of July 2000, has been carried out. The retrieval of the fire scar does not seem to be feasible with one look ERS SAR imagery. However, due to the increased influence of soil moisture in the backscattering, the fire scar might be detected with multitemporal measurements taken on the same season. Moreover, the measurement, taken throughout at least one year, of the similarity between the backscattering of the burned area and the backscattering of a bare soil around or inside the burnted area, may provide an index of the reforestation process. The presented results are still preliminary. In fact, other significant features still need to be investigated. First of all, the cross-polarized backscattering return which is available with the Envisat-ASAR. The coherence level derived by SAR interferometry may also be very significant and it is worthy to be considered, although tandem mission stopped few months before event. Finally, textural feautures calculated over the regions of interest can also provide additional information. A scattering electromagnetic model, developed and made available by University of Tor Vergata, Rome, will be also used for the understanding and the validation of the results that will be obtained. ACKNOWLEDGEMENTS Prof. P. Ferrazzoli of Tor Vergata University, Rome, and the research group coordinated by Prof. L. Manes of La Sapienza University, Rome, are gratefully ackoweledged for their helpful comments and suggestions. 6. REFERENCES [1] Viedma O., Meliá J., Segarra D., García-Haro J., Modeling rates of ecosystem recovery after fires by using Landsat TM data, Remote Sensing of Environment, 61, pp , 1997 [2] Díaz-Delgado R., Lloret F., Pons X., Influence of fire severity on plant regeneration by means of remote sensing imagery, International Journal of Remote Sensing, 24 (8), pp , 2003 [3] Sun G., Rocchio L., Masek J., Williams D., Ranson K.J., Characterization of Forest Recovery From Fire Using Landsat and SAR Data, Proceedings of Geoscience and Remote Sensing Symposium 2002, vol. 2, pp , 2002 [4] Bourgeau-Chavez L.L., Harrell P.A., Kasischke E.S., French N.H.F., The detection and mapping of Alaskan wildfires using a spaceborne imaging radar system, International Journal of Remote Sensing, 18 (2), pp , 1997 [5] Kasischke E. S., Bourgeau-Chavez L.L., French N.H.F., Observations of variations in ERS-1 SAR image intensity associated with forest fires in Alaska, IEEE Transactions on Geoscience and Remote Sensing, Vol. 32, No. 1, 1994 [6] Antikidis E., Arino O., Laur H., Arnaud A., ERS SAR Coherence & ATSR Hot Spots: a sinergy for mapping deforested areas. The special case of the 1997 fire event in Indonesia, Proceedings of the 2 nd International Workshop on Retrieval of Bio- & Geophysiacal Parameters from SAR data for Land Applications, ESTEC, Noordwijk, The Netherlands, 1998 [7] D. Monaco, The Castelfusano pinewood: a case study, Proceedings International Workshop on "Improving dispatching for forest fire control", Creete, 6-8 December, 2001 [8]