Flooded wetlands map ping from Sen ti nel-2 im ag ery with spec tral wa ter in dex:
a case study of Kampinos Na tional Park in cen tral Po land
Tatiana SOLOVEY1, *
1 Pol ish Geo log i cal In sti tute – Na tional Re search In sti tute, Rakowiecka 4, 00-975 Warszawa, Po land
Solovey, T., 2020. Flooded wetlands map ping from Sen ti nel-2 im ag ery with spec tral wa ter in dex: a case study of Kampinos Na tional Park in cen tral Po land. Geo log i cal Quar terly, 64 (2): 492–505, doi: 10.7306/gq.1509
As so ci ate Ed i tor: Stanis³aw Wo³kowicz
Flood mon i tor ing of wetlands and floodplains is a new is sue in re mote sens ing, as com pared to the map ping of open wa ter bod ies. The method based on spec tral wa ter in di ces, cal cu lated on the ba sis of green, red and shortwave in fra red bands, is one of the most pop u lar meth ods for the rec og ni tion of a wa ter body in multispectral im ages. The re cently in tro duced Sen ti - nel-2 sat el lite can pro vide multispectral im ages with high spa tial res o lu tion. This new data set is po ten tially of great im por - tance for flood map ping, due to its free ac cess and the fre quent re visit ca pa bil i ties. In this study, three pop u lar wa ter in di ces (Mod i fied Nor mal ized Dif fer ence Wa ter In dex, Nor mal ized Dif fer ence Pond In dex and Nor mal ized Dif fer ence Tur bid ity In - dex) were used. The ef fi ciency of the pro posed method was tested ex per i men tally us ing the Sen ti nel-2 im age for the Kampinos Na tional Park in Po land. The ex per i ment com pared four ex trac tion al go rithms in clud ing three based on in di vid ual wa ter in di ca tors and one on a com bi na tion of them. The re sults showed that the 10-metre false col our com pos ite pro duced sig nif i cantly im proved the rec og ni tion of flood ing in wet land ar eas by com par i son with sin gle spec tral wa ter in di ces. In this way, flooded wetlands were mapped based on the Sen ti nel-2 data set for the years 2017–2018.
Key words: re mote sens ing, Sen ti nel-2, flooded wetlands map ping, Mod i fied Nor mal ised Dif fer ence Wa ter In dex (MNDWI), Nor mal ised Dif fer ence Pond In dex (NDPI), Nor mal ised Dif fer ence Tur bid ity In dex (NDTI).
INTRODUCTION
Wetlands by def i ni tion are in ter me di ate eco sys tems be - tween typ i cally aquatic and typ i cally ter res trial ones, formed un - der the in flu ence of con stant or pe ri odic sat u ra tion of the ground and char ac ter ized by the pres ence of hy dro philic veg e - ta tion and or ganic de pos its. Wetlands play a key role in im prov - ing wa ter qual ity, mit i gat ing floods (Acreman and Holden, 2013;
Loveline, 2015) and drought (Ilnicki, 2002), ab sorb ing car bon di ox ide and ox y gen emis sions (Kayranli et al., 2009), pro vid ing nat u ral hab i tats and sup port ing biodiversity (Tobolski, 2003;
£achacz, 2004; Mitsch, 2009; Mitsch and Gosselink, 2015).
Map ping of wetlands has al ways been nec es sary, since a reg is - ter of these ar eas is oblig a tory for en vi ron men tal pro tec tion au - thor i ties and wa ter man age ment.
Flood in un da tion is a key hy dro log i cal char ac ter is tic of floodplain wetlands. High wa ter gen er ates flood ing in wetlands, and the range, fre quency, du ra tion and depth of flood ing de ter - mine the dis tri bu tion, type and vi tal ity of the wet land veg e ta tion.
Floods play a key role in sup ple ment ing wet land ground wa ter
stocks. There fore, it is nec es sary to mon i tor the spa tial ex tent of flood ing.
Re mote sens ing has proven to be a use ful and fre quently used tool for wetlands mon i tor ing, as the data ob tained can pro - vide mac ro scopic, real-time, dy namic in for ma tion that is not avail able from in situ tech nol ogy (Bourgeau-Chavez et al., 2001; Kasischke et al., 2003; Chen et al., 2004; Jones et al., 2009; Brisco et al., 2011; Budzyñska et al., 2011; Du et al., 2011; Feng et al., 2012; Solovey, 2013, 2017; Nandi et al., 2017; Whyte et al. 2018; Wu, 2018). Var i ous meth ods have been used in wetlands mon i tor ing in clud ing sin gle band den sity slic ing (Butera, 1983; Mar ti nez and Le Toan, 2007; Melack and Hess, 2010; Morandeira et al., 2016; Moser et al., 2016), un su - per vised and su per vised clas si fi ca tion (Ramsey and Laine, 1997; Zomer et al., 2009; Mwita et al., 2013; Huang et al., 2014a; Napiórkowska, 2014; White et al., 2015) and spec tral wa ter in di ces (Li and Chen, 2005; Huang et al., 2014b; Li et al., 2016; Nandi et al., 2017). The wetlands map ping method based on spec tral wa ter in di ces is cur rently very pop u lar be cause it is easy to ap ply, ef fi cient and has low com put ing costs. The most use ful spec tral wa ter in di ces for wet land clas si fi ca tion are the Nor mal ized Dif fer ence Veg e ta tion In dex (NDVI; Kayastha et al., 2012), the Land Sur face Wa ter In dex (LSWI; Dong et al., 2014), the Nor mal ized Dif fer ence Wa ter In dex (NDWI; McFeeters, 1996; Seiler et al., 2009; Dvorett et al., 2016), Mod i fied Nor mal - ized Dif fer ence Wa ter In dex (MNDWI; Xu, 2006; Davranche et al., 2010) and the Soil and At mo sphere Re sis tant Veg e ta tion In dex (SARVI; Huete et al., 1997).
* E-mail: tatiana.solovey@pgi.gov.pl
Received: July 12, 2019; accepted: October 4, 2019; first published online: February 6, 2020
MODIS, Land sat and cur rently, Sen ti nel sat el lite im ages are the most used in mon i tor ing of sea son ally flooded wetlands.
MODIS data has a low spa tial res o lu tion and a high time res o lu - tion; on this ba sis, flooded wetlands can not be ac cu rately de ter - mined as well as pro vid ing iden ti fi ca tion of smaller ob jects, whereas im ages with a better spa tial res o lu tion, such as Land - sat data, al ways have a low time fre quency and a se ries of ir reg - u lar im ages due to clouds. The Eu ro pean Space Agency (ESA) has made great prog ress in in creas ing the ef fi ciency of flooded wetlands map ping by in tro duc ing a new op ti cal, pre cise spa tial res o lu tion us ing the Sen ti nel-2 sat el lite. Sen ti nel-2 im ages have great po ten tial for wet land map ping on a re gional scale due to prop er ties such as spa tial res o lu tion of 10 m for four bands and a 10-day re turn fre quency, and free ac cess. Sen ti nel-2 of fers multispectral im ages and has a to tal of 13 bands, in which four bands [blue, green, red and Near, In fra, Red (NIR)] have a spa - tial res o lu tion of 10 m, four veg e ta tion bands of Red Edge and two bands of Short Wave In fra Red (SWIR) have 20 m res o lu - tion, while the Coastal Aero sol, Wa ter Va por and Cir rus Bands have 60 m res o lu tion. How ever, con sid er ing the four fine spec - tral res o lu tion bands, a pan chro matic band can be pro duced and used in Sen ti nel-2 im age fu sion for pro duc ing ten fine spa - tial res o lu tion bands (Selva et al., 2015).
This ar ti cle de scribes flooded wetlands mon i tor ing us ing a com bi na tion of three spec tral wa ter in di ces to iden tify floodplains within wetlands in clud ing: the Mod i fied Nor mal ized Dif fer ence Wa ter In dex (MNDWI), the Nor mal ized Dif fer ence Pond In dex (NDPI) and the Nor mal ized Dif fer ence Tur bid ity In - dex (NDTI). The com bi na tion of these in di ces helps the iden ti fi - ca tion of wa ter res er voirs, wa ter and swamp veg e ta tion and wa - ter tur bid ity, which makes it easy to iden tify wetlands.
Xu (2006) pro posed a MNDWI which uses a short wave in - fra red (SWIR) beam to re place the NIR band used in NDWI (McFeeters, 1996) as the NDWI in di ca tor is sen si tive to built-up ar eas and of ten causes over es ti ma tion of wa ter res er voirs.
Many pre vi ous re search stud ies have shown that MNDWI is more suit able for en rich ing wa ter in for ma tion and can iden tify wa ter res er voirs with greater ac cu racy than NDWI (Xu, 2006; Li et al., 2013; Du et al., 2014; Singh et al., 2015).
NDPI (Lacaux et al., 2007) can ef fec tively cap ture wa ter and swamp veg e ta tion oc cur ring in wetlands with sur face wa ter, in con trast to the clas sic NDVI (Tucker and Sell ers, 1986), which did not work well for veg e ta tion in ar eas flooded with a shal low layer of wa ter. Lacaux et al. (2007) pro posed NDTI, which ef fec - tively cap tures tur bid wa ter res er voirs prev a lent in wetlands that were of ten con fused with open soil on re mote sens ing im ages as the in crease in wa ter tur bid ity and as so ci ated ra dio met ric re - sponse causes the tur bid wa ter res er voirs to be have like open soils (Guyot, 1989). The NDTI in di ca tor uses green and red bands of im ages from the re mote sens ing based on the level of wa ter tur bid ity in creas ing due to the in crease of par ti cles sus - pended in wa ter, which causes a greater re flec tion of the red band than the green (Is lam and Sado, 2006).
Based on the ob served ad van tages of the MNDWI, NDPI and NDTI in di ca tors as well as the po ten tial of the Sen ti nel-2 da ta base re sources, we de cided to test the use ful ness of com - bin ing these in di ca tors as an RGB com pos ite for map ping flooded wetlands in the Vistula val ley. An ad van tage of this ap - proach is the ease of the method and the avail abil ity of Sen ti - nel-2 data, which is par tic u larly im por tant for ex plor atory map - ping with lim ited re sources.
The pur pose of this study was to: (1) cre ate a se ries of tem - po ral com pos ite im ages of the spec tral wa ter in di ces MNDWI, NDPI and NDTI with a spa tial res o lu tion of 10 m based on Sen - ti nel-2 im ages for the years 2017–2018; (2) flooded wetlands
map ping with the use of su per vised clas si fi ca tion tech niques;
and (3) as sess ment of the use ful ness of the 10 m im ages cre - ated in flooded wetlands map ping.
STUDY AREA
The re search area is lo cated in the cen tral part of the Vistula val ley in Po land. Part of the Vistula val ley dis cussed in this study is lo cated in the Kampinos Na tional Park (KPN), which is also a Natura 2000 area and spreads over an area of 385 km2 (Fig. 1). The KNP is lo cated in the Cen tral Mazovian Low land in the Vistula val ley, north-west of War saw.
The av er age height of the KPN area, which is gen er ally even, is ~80 m above sea level (Krogulec, 2004). The land - scape here is char ac ter ized by lat i tu di nal pat terns. From the north spreads the Vistula flood ter race. Far ther to the south runs the ter race of the Kampinos For est con sist ing of two al ter - nat ing dune and marshy strips, and the ter race of the
£owicko-B³oñska Plain (Krogulec, 2011). The cli mate is mod er - ate with an av er age an nual tem per a ture over many years (1994–2016) of 8.7°C and an av er age an nual pre cip i ta tion of 568.1 mm (Olszewski et al., 2018).
The ba sic eco sys tem for which the KNP was es tab lished is the valu able nat u ral wetlands, which cover 176.9 km2 (Kopeæ et al., 2013). The north ern swamp belt, with an area of
~10,000 ha, is much more de vel oped than the south ern one.
Most swamps here have been turned into mead ows and pas - tures. The south ern swamp belt, with an area of ~7,000 ha, is frag mented into sev eral val leys that no lon ger form a dis tinct belt. Wet mead ows and sedges pre vail in this belt (Michalska- Hejduk, 2001).
Nat u ral valu able wet land com mu ni ties of the non-for est ar - eas of the KNP are: vari able hu mid ity molinia mead ows of Molinion, hu mid mead ows of Molinietalia, ex ten sively used fresh mead ows of Arrhenatherion, tran si tion bogs and quak ing bogs (mostly with Scheuzerio-Caricetea veg e ta tion), mari gold mead ows with Calthion, Phragmition rushes and, most wide - spread, large reed rushes of Magnocaricion (Michalska- Hejduk, 2004).
The swamps (for ested wetlands) of the KNP are mainly rep re - sented by al der and ri par ian for ests, which oc cupy al most half of the area (8810 ha) of the boggy belt. The red-necked al der Ribeso nigri-Alnetum and the Fraxino-Alnetum ash-al der for est are typ i cal (Michalska-Hejduk et al., 2011). Low er ing of ground wa ter level is clearly con sid ered to be the great est threat to the na ture of the KPN. The main rea sons for the deg ra da tion of wetlands were the stop ping of flood ing as a re sult of the con struc tion of the Vistula and Bzura river em bank ments as well as drain age chan nels and ditches (Okruszko et al., 2011). There fore, it is ex tremely im por - tant to mon i tor flooded wetlands of the KPN.
DATASET
The dataset used in this study is the Sen ti nel-2 Level 1C prod uct (Ta ble 1) in the Uni ver sal Transvers Mercator (UTM)/WGS84 pro jec tion, which was down loaded from the ESA Sen ti nel-2 Pre-Op er a tions Hub (https://scihub.co per ni - cus.eu). A sub set cov er ing 40 ´ 17 km and can tered at 52024’35”N, 20017’3”E was used for the case study (Fig. 2).
The false col our com pos ite of the Sen ti nel-2 im age at 10 m res - o lu tion of the study area is shown in Fig ure 2A. The im ages of the green band at 10 m, the NIR band at 10 m, the SWIR band at 20 m, re spec tively, are shown in Fig ure 2B–D, and these four
bands were in volved in the cal cu la tion of wa ter in di ces of the MNDWI, NDPI and NDTI. For the anal y sis, a Sen ti nel-2 im age was used with a zero cloud in 2017–2018. Date of ac qui si tion is shown in Ta ble 2.
METHODOLOGY
SPECTRAL WATER INDICES
Mod i fied Nor mal ised Dif fer ence Wa ter In dex (MNDWI).
The Mod i fied Nor mal ised Dif fer ence Wa ter In dex (MNDWI; Xu, 2006) is de fined as:
MNDWI Green SWIR
Green SWIR
= -
+
r r
r r
[1]
where: rGreen – the top of the at mo sphere (TOA) reflectance value of the green band; rSWIR – the TOA reflectance value of the SWIR band 12.
Com pared to the raw dig i tal num bers (DN), the TOA reflectance is more suit able for cal cu lat ing the MNDWI (Chander et al., 2009; Li et al., 2013; Ko et al., 2015). The freely avail able Sen ti nel-2 Level-1C dataset is al ready a stan dard prod uct of TOA reflectance (Drusch et al., 2012).
Fig. 1. Lo ca tion of the study area within a Dig i tal El e va tion Model (DTM) of Po land
Sen sor name Sen sor type Band in for ma tion Band num bers Res o lu tion [m]
Sen ti nel-2 op ti cal
Blue (490 nm) Band 2 10
Green (560 nm) Band 3 10
Red (665 nm) Band 4 10
NIR (842 nm) Band 8 10
SWIR (1610 nm) Band 11 20
SWIR (2190 nm) Band 12 20
T a b l e 1 Summary of the re motely sensed datasets used for this study
Fig. 2. Im age ac quired on 2 Oc to ber 2017
A – 10 metre false col our map (R: Band 4; G: Band 3; B: Band 8) study area; B – 10 m green Band 3; C – 10 m red Band 8; D – 20 m SWIR Band 11; the two square frames shown in di cate the
lo ca tions of sub ar eas A and B
Com pared to NDWI (Gao, 1996), wa ter bod ies have greater pos i tive val ues in the MNDWI be cause wa ter bod ies ab sorb more SWIR light than NIR light; soil, veg e ta tion and built-up ar - eas have smaller neg a tive val ues be cause they re flect more SWIR light than green light (Sun et al., 2012).
The Nor mal ised Dif fer ence Wa ter In dex (MNDWI; Gao, 1996) is de fined as:
NDWI Green NIR
Green NIR
= -
+
r r
r r
[2]
where: rGreen – the TOA reflectance value of the green band; rNIR – the TOA reflectance value of the NIR band.
For Sen ti nel-2, the green band has a spa tial res o lu tion of 10 m, while the SWIR band (Band 12) has a spa tial res o lu tion of 20 m. Thus, the MNDWI needs to be cal cu lated at a spa tial res - o lu tion of ei ther 10 or 20 m. In this study, the spa tial res o lu tion of the SWIR band was in creased from 20 to 10 m by us ing pan-sharp en ing al go rithms. For this pur pose, the High Pass Fil - ter (HPF) was se lected (Chavez et al., 1991) be cause the HPF pro duces a sharp ened 10 m SWIR band with a higher qual ity with no ref er ence (QNR) value.
If the spa tial res o lu tion of Band 12 is in creased from 20 to 10 m, the MNDWI with a spa tial res o lu tion of 10 m, MNDWI10m, can then be cal cu lated as:
MNDWI m
m 10 m
3 12
10
3 12
= - 10
+
r r
r r
[3]
where: r3 – the TOA reflectance value of the green band and re fers to the TOA reflectance value of Band 12 at 10 m, which is pro duced by downscaling the orig i nal 20 m Band 12. This is achieved by us ing pan-sharp en ing al go rithms.
Nor mal ised Dif fer ence Pond In dex (NDPI). The clas sic NDVI did not work well in the case of veg e ta tion in wetlands cov ered with a shal low layer of wa ter, which is why the NDPI was de vel oped.
The Nor mal ised Dif fer ence Pond In dex (NDPI; Lacaux et al., 2007) is de fined as:
NDPI Green SWIR
Green SWIR
= -
+
r r
r r
[4]
where: rSWIR – the TOA reflectance value of the SWIR Band 11;
rGreen – the TOA reflectance value of the green band.
As in the case of the MNDWI, for the NDPI cal cu la tion, the spa tial res o lu tion of the SWIR band Sen ti nel-2 band should be in creased from 20 to 10 m by us ing pan-sharp en ing al go rithms based on the High Pass Fil ter (HPF).
The NDPI with a spa tial res o lu tion of 10 m, NDPI10m, can then be cal cu lated as:
NDPI m
m 10 m
3 11
10
3 11
= - 10
+
r r
r r
[5]
where: r1110mre fers to the TOA reflectance value of Band 11 at 10 m, which is pro duced by downscaling the orig i nal 20 m Band 11 and r3
is the TOA reflectance value of the green band.
NDTI. The Nor mal ised Dif fer ence Tur bid ity In dex (NDTI) (Lacaux et al., 2007) is de fined as:
NDTI d Green
d Green
= -
+
r r
r r
Re Re
[6]
where: rRed – the TOA reflectance value of the red band; rGreen – the TOA reflectance value of the green band.
Ac qui si tion
date Wa ter in di ces
Min. Max. Mean SD
wa ter body
flood.
wet land
wa ter body
flood.
wet land
wa ter body
floodD.
wet land
wa ter body
flood.
wet land 29/03/2017
veg e ta tion pe riod
MNDWI 0.44 –0.38 0.77 0.42 0.65 0.03 0.06 0.17
NDPI 0.39 –0.54 0.74 0.30 0.61 –0.19 0.07 0.18
NDTI –0.04 –0.16 0.04 0.14 0.00 0.01 0.02 0.06
2/10/2017 veg e ta tion pe riod
MNDWI –0.37 –0.68 0.81 –0.47 0.25 –0.57 0.24 0.04
NDPI –0.48 –0.80 0.77 –0.66 0.10 –0.73 0.25 0.03
NDTI –0.36 0.04 0.14 0.30 –0.18 0.18 0.13 0.06
26/12/2017
MNDWI 0.12 –0.37 0.40 0.03 0.24 –0.14 0.05 0.08
NDPI 0.13 –0.35 0.39 0.03 0.25 –0.17 0.06 0.08
NDTI –0.11 –0.14 0.13 0.23 0.01 –0.02 0.05 0.07
8/01/2018
MNDWI 0.40 0.21 0.92 0.59 0.64 0.39 0.14 0.08
NDPI 0.27 0.08 0.98 0.47 0.58 0.31 0.14 0.08
NDTI –0.49 –0.04 0.03 0.18 –0.18 0.08 0.11 0.05
19/03/2018 veg e ta tion pe riod
MNDWI 0.42 –0.26 0.64 0.48 0.53 0.17 0.05 0.17
NDPI 0.36 –0.39 0.59 0.37 0.48 0.03 0.05 0.19
NDTI –0.04 –0.01 0.03 0.11 0.00 0.03 0.01 0.03
T a b l e 2 Max i mum (Max), min i mum (Min), mean and stan dard de vi a tion (SD) val ues of wa ter bod ies and flooded wetlands
within the MNDWI, NDTI and NDPI
FLOODED WETLANDS MAPPING AND ASSESSING THE ACCURACY OF RESULTS
Af ter the cre ation of the MNDWI, NDPI and NDTI, floodplain wetlands can be mapped based on a false col our com pos ite of Red (NDTI), Blue (MNDWI) and Green (NDPI). The tests car - ried out to se lect the col our com po si tion that al lows the best rec - og ni tion of the ob jects dis cussed show that in the RNDTIGNDPIBMNDWI com po si tion the floodplain wetlands are pre - sented in a more ex pres sive way.
The next stage is re lated to the per for mance of a pixel clas - si fi ca tion su per vised by the max i mum like li hood method (Lewiñski, 2007) in the Erdas Imag ine en vi ron ment. When de - cid ing on the choice of classes, it was noted that the classes were rep re sented by ho mo ge neous ob jects, re gard ing the iden - ti fi ca tion of which there is no doubt. Based on the anal y sis of the spec tral char ac ter is tics of the ini tially se lected classes, it was de cided to sep a rate 2 classes of ob jects, which are:
1. Wa ter res er voirs (wa ter bod ies);
2. Flooded wetlands with a trans lu cent plant cover (in the re search area mainly from Phragmition, Sparganio- Glycerion and Magnocaricion com pounds).
The def i ni tion of flooded wetlands in for est ar eas was aban - doned due to the in abil ity to clearly de fine this class based on the ap plied RNDTIGNDPIBMNDWI com po si tion. The se lec tion of the right scat ter plot is made on the ba sis of field ob ser va tions and use of a nu mer i cal ter rain model, as sum ing that the out flow de - pres sions have the high est prob a bil ity of flood ing. In or der to ver ify the cor rect ness of the train ing fields se lected for in di vid - ual classes, the spec tral dis tance method ac cord ing to Jeffreys-Matusit (Jensen, 1996) was used. The mea sure ment of the spec tral dis tance JMab was cal cu lated us ing the for mula (Jensen, 1996):
( )
JMab = 2 1-eBhatab [7]
where: JMab – J-M dis tance be tween classes a and b, Bhatab – Bhattacharyya dis tance be tween classes a and b.
Fig . 3. Sub area A of the 10 m false col our com pos ite RNDTIGNDPIBMNDWI, MNDWI, NDPI i NDTI im age for ac qui si tion dates: (A) 29.03.2017; (B) 2.10.2017; (C) 26.12.2017; (D) 8.01.2018; and (E) 19.03.2018
A thresh old for the max i mum spec tral dis tance JMab equal to 2 was used, in di cat ing com plete class sep a ra tion.
In this study, the over all ac cu racy (OA) and Kappa co ef fi - cient were used to as sess the clas si fi ca tion ac cu racy on the ba - sis of the er ror ma trix (Foody, 2002). Monserud and Leemans (1992) sug gested that Kappa co ef fi cient val ues <0.4 rep re sent poor or very poor agree ment, val ues from 0.4 to 0.55 rep re sent fair agree ment, val ues from 0.55 to 0.7 rep re sent good agree - ment, val ues from 0.7 to 0.85 rep re sent very good agree ment, and val ues >0.85 rep re sent an ex cel lent agree ment be tween im ages.
The anal y sis was di vided into three main stages:
1. Au to matic de tec tion of flood ex tent us ing im age thresh - old seg men ta tion based on thresh old val ues for the MNDWI, NDPI and NDTI;
2. Semi-au to matic de tec tion of the ex tent of flood ing us ing the clas si fi ca tion su per vised by the colour com pos ite of Red (NDTI) – Blue (MNDWI) – Green (NDPI);
3. Val i da tion of flooded and wa ter body maps based on the Kappa co ef fi cient.
RESULTS AND DISCUSSION
The anal y sis was car ried out in the pre- and post-veg e ta tive pe riod in 2017–2018, when Sen ti nel-2 im ages with no clouds were avail able (Ta ble 2). For each ac qui si tion date for sub ar eas A and B four wa ter in dex im ages were de vel oped, RNDTIGNDPIBMNDWI, MNDWI, NDPI and NDTI, as shown in Fig - ures 3 and 4. The re sults of the de tec tion of flooded wetlands and wa ter body im ages in Fig ure 2 were com pared.
For wa ter bod ies, most MNDWI val ues are >0.5, while NDPI val ues for them are >0.3. The MNDWI in di ca tor pro vides the best distinguishability of wa ter body com pared to NDPI and NDTI. Poor suit abil ity for iden ti fi ca tion of a wa ter body is in di - cated by the NDTI in dex, be cause the wa ter map ping val ues (from –0.03 to 0.03) are mostly sim i lar to ar a ble land with a veg - e ta tion cover.
For flooded wetlands with a trans lu cent plant cover, the MNDWI val ues are very dif fer ent, as shown in Fig ures 3 and 4.
In com par i son with a wa ter body, flooded wetlands with a trans - lu cent plant cover have much lower MNDWI val ues. Out side Fig . 4. Sub area B of the 10 m false col our com pos ite RNDTIGNDPIBMNDWI, MNDWI, NDPI i NDTI im age for ac qui si tion dates: (A)
29.03.2017; (B) 2.10.2017; (C) 26.12.2017; (D) 8.01.2018; and (E) 19.03.2018
the veg e ta tion pe riod, the dif fer ences be tween the MNDWI val - ues of wa ter bod ies and flooded wetlands are re duced. The weak ness of the MNDWI in di ca tor is the sim i lar ity be tween flooded wetlands and ar a ble land with out veg e ta tion. By us ing the NDPI in di ca tor, flooded wetlands can be slightly dif fer en ti - ated com pared to the MNDWI in di ca tor. The NDTI in di cates the worst re sults, but its ad van tage is the detectability of rush veg e - ta tion, which in di rectly iden ti fies sea son ally flooded wetlands. In com par i son to the wa ter body, flooded wetlands are less well cap tured on im ages with sin gle wa ter in di ces. The com bi na tion of these in di ces strength ens the range of ar eas flooded with wa - ter and cov ered with wa ter and swamp veg e ta tion, which al lows the most pre cise iden ti fi ca tion of floodplain wetlands.
Ta ble 2 con tains a list of sta tis ti cal pa ram e ters of train ing fields for wa ter bod ies and flooded wetlands with ref er ence to wa ter in di ces – MNDWI, NDTI, NDPI.
For flooded wetlands, the min i mum, max i mum and av er age val ues of MNDWI are smaller than the MNDWI val ues for a wa - ter body. These dif fer ences vary de pend ing on the stage of de - vel op ment of swamp veg e ta tion. In au tumn, these dif fer ences reach their larg est sizes, against slightly smaller dif fer ences in the early spring pe riod, and they are min i mal in win ter. The stan dard de vi a tion in win ter for wa ter bod ies and flooded wetlands shows sim i lar low val ues. In con trast, in early spring the value of stan dard de vi a tion in creases sig nif i cantly for flooded wetlands. To sum up, based on MNDWI, worse clas si fi - ca tion re sults due to poor distinguishability of wa ter bod ies and flooded wetlands can be ex pected in win ter and the best re sults avail able in early spring. In au tumn, due to the green col our of wa ter bod ies, the classes ana lysed are of ten con fused.
A sim i lar trend to that of the MNDWI is shown by the NDPI.
The av er age NDPI value com pared to the av er age MNDWI value is mostly ~0.05 for wa ter bod ies and ~0.15 for flooded wetlands. Us ing NDPI, you can best sep a rate the wa ter res er - voir from flooded wetlands based on a thresh old of zero.
Flooded wetlands with a trans lu cent veg e ta tion cover com - pared to wa ter bod ies al ways have an NDPI value be low or around zero. The only ex cep tion is when the wetlands were flooded with wa ter above the veg e ta tion cover such as hap - pened on Jan u ary 8, 2018.
Us ing the NDTI, worse re sults of ex trac tion of wa ter body and flooded wetlands are ob tained in com par i son with the MNDWI and NDPI. How ever, based on the NDTI, better dis - crim i na tion can be achieved be tween wa ter bod ies and flooded wetlands. Most of the NDTI val ues of wa ter body are <0, against a back ground of very dif fer ent val ues for flooded wetlands.
Fig ure 5 shows the di a grams of scat ter ing of bright ness val - ues of pix els of a train ing field for wa ter in dex pairs: MNDWI and NDTI, MNDWI and NDPI as well as NDTI and NDPI. Train ing fields have been de fined for 5 classes, which are: wa ter body, flooded wetlands, ar a ble land with veg e ta tion cover, ar a ble land with out plant cover, and for ests. The least sep a rated class on all graphs is ar a ble land with out a plant cover, which spo rad i - cally shows sim i lar val ues to wa ter in di ces from flooded wetlands.
Based on the pair of MNDWI and NDTI, the worst iso la tion re sults were ob tained be tween the classes analysed. In this case, the spec tral sim i lar ity of ar a ble land with out veg e ta tion cover and flooded wetlands is clearly vis i ble.
On the ba sis of the NDTI and NDPI pair, a sup pos edly low level of sep a ra tion be tween classes of ar a ble land with out veg - e ta tion cover and flooded wetlands is ob tained, but the qual ity of sep a ra tion of ar a ble land with out veg e ta tion cover and wa ter bod ies is im proved, be cause shal low and tur bid wa ter body in the area of re search have sim i lar fea tures to open soil.
On the ba sis of a pair of MNDWI and NDPI, the best sep a - ra tion re sults are ob tained be tween the ma jor ity of classes, ex - cept for ests and ar a ble land with a veg e ta tion cover.
When de cid ing on the se lec tion of an op ti mal set of spec tral wa ter in di ces for flooded wetlands clas si fi ca tion, at ten tion was drawn to the dis tance be tween classes in the graphs de picted in Fig ure 5. It was found that in the early spring pe riod the best re sults of the flooded wetlands clas si fi ca tion are en sured by the com bi na tion of the MNDWI and NDPI in di ces (Fig. 5A, E). In win ter (no snow cover) and late au tumn, the best flooded wetlands re sults can be achieved by com bin ing the MNDWI, NDPI and NDTI in di ces (Fig. 5B–D). Ad di tion of the NDTI im - proves the sep a ra tion of flooded wetlands from ar a ble land with out plant cover.
The fi nal flooded wetlands maps were made on the ba sis of the su per vised clas si fi ca tion us ing the com bi na tion of wa ter in - di ces as a false col our com pos ite RNDTIGNDPIBMNDWI and train ing fields for two classes – wa ter bod ies and flooded wetlands. Five maps were ob tained, parts of which for sub area A and B are shown in Fig ures 6 and 7. On the maps flooded wetlands are shown in brown, wa ter bod ies are blue.
In the clas si fi ca tion per formed, 0.25 ha was as sumed as the ref er ence unit. The size of the adopted ref er ence unit re sults from the in ter pre ta tion as sump tions, ac cord ing to which, in the course of vi sual in ter pre ta tion, ob jects with di men sions of 5 ´ 5 mm can be de fined and rec og nized on the scale of the map. This means that on a scale of 1:10 000 (on the scale of the sat el lite im age in ter pre ta tion), one can in prac tice rec og nize an area of 0.25 ha.
Fig ures 6 and 7 show a high level of rec og ni tion of non-for - est flooded wetlands out of the grow ing sea son (Figs. 6B–D and 7B–D). In the veg e ta tion pe riod, the rec og ni tion of non-for - est flooded wetlands is sig nif i cantly worse (Figs. 6A, E and 7A, E). Then only low veg e tated wetlands can be mapped, for ex - am ple those with sparsely veg e tated ar eas with short grasses and small wet land plant spe cies. It is not pos si ble to dis cern wa - ter on the sur face in highly veg e tated ar eas, con sist ing of larger spe cies of veg e ta tion (e.g., Phragmition and Magnocaricion).
The ap plied false col our com pos ite RNDTIGNDPIBMNDWI en sures sat is fac tory dif fer en ti a tion of wa ter bod ies from flooded ar eas based on the ob ject-pixel het er o ge ne ity fea ture. Wa ter bod ies are char ac ter ized by a ho mo ge neous ob ject-pixel tex ture in con trast to clearly het er o ge neous im ages of flooded ar eas.
The key prob lem in the clas si fi ca tion was the sim i lar pixel viv id ness for flooded wetlands and ar a ble land with out a plant cover. With the help of the NDVI in dex, ar a ble land with out plant cover was ef fec tively elim i nated, leav ing only veg e ta tion in wetlands and in wa ter. In gen eral, flooded ag ri cul tural ar eas, es pe cially grass lands, have been found to strongly over lap with flooded wetlands. Due to the au thor’s con sid er able ex pe ri ence in rec og niz ing wetlands in the re search area, the flooded ag ri - cul tural ar eas were ex pertly elim i nated. This sit u a tion al lows us to con clude that the ap plied false col our com pos ite RNDTIGNDPIBMNDWI al lows ef fec tive sep a ra tion of all flooded ar - eas. How ever, the sep a ra tion of flooded wetlands re quires the in tro duc tion of ad di tional in ter pre ta tional in di ca tors. In this case, the NDVI in di ca tor was used. The method based on the com bi na tion of Sen ti nel-1 and Sen ti nel-2 ra dar data also show good re sults of flooded wetlands sep a ra tion from flooded ag ri - cul tural ar eas (Whyte, 2018).
Over all ac cu racy (OA) and Kappa in di ca tors were used to quan tify the ac cu racy of flooded wetlands maps of re gions A and B. Ta ble 3 shows that the maps orig i nat ing from the data as of 29/03/2017, 26/12/2017 and 19/03/2018 have a rel a tively sim i lar high ac cu racy be cause a poorly de vel oped veg e ta tion
cover or its ab sence pro motes the vis i bil ity of the wa ter sur face.
The maps as of 08/01/2018 show slightly worse ac cu racy due to the pres ence of slight snow fall. The larg est er rors are ob - tained in the pe riod of ac tive bio mass de vel op ment, ex em pli fied on Oc to ber 2, 2017.
Thanks to the ap proach based on mul ti lat eral tri als and er - rors, it has been found that a het er o ge neous wet land en vi ron - ment can be sat is fac to rily di vided into seg ments to gen er ate flooded ar eas. It has been dem on strated that in di vid ual spec tral wa ter in di ca tors are not suf fi cient to iden tify flood ing within the var ied wetlands land scapes, that are dif fi cult to seg ment. The MNDWI in di ca tor, which was ef fec tive in iden ti fy ing wa ter bod - ies, did not pro vide rec og ni tion of flooded veg e ta tion. There - fore, a syn er gis tic ap proach may be more ef fec tive, as was found in this study us ing a com bi na tion of spec tral wa ter in di ca -
tors vi su al ized by the false col our com pos ite RNDTIGNDPIBMNDWI. The re sults of flood map ping tested in two re gions within the Kampinos Na tional Park show fairly good re sults, where the over all map ac cu racy is >90%, and the Kappa co ef fi cient is
>0.80. How ever, the ap proach ap plied does not work in for est ar eas and shows worse ef fi cacy at times of lux u ri ant veg e ta tion de vel op ment. The Sen ti nel-2 data ex per i ment also showed that the ap proach taken is use ful for map ping flood ing in gen eral, not only in wet land ar eas. Very good re sults of iden ti fy ing flood - ing have been ob tained for ag ri cul tural land.
The anal y sis of sea sonal changes in flood ing has shown a large vari abil ity in the ranges of flooded ar eas. In moist pe ri - ods, the area of flooded ar eas within sub-re gions A and B is greater by 7.2 times com pared to the dry pe riod. Quite a large in crease in the ex tent of flood ing in wet pe ri ods in the re search Fig . 5. Graphs of scat ter ing of bright ness val ues of pix els of train ing fields for wa ter in dex pairs MNDWI and NDTI, MNDWI and
NDPI as well as for NDTI and NDPI for ac qui si tion dates: (A) 29.03.2017; (B) 2.10.2017; (C) 26.12.2017; (D) 8.01.2018; and (E) 19.03.2018
Fig. 6. Sub area A of the 10 m false col our com pos ite RNDTIGNDPIBMNDWI and the re sult ing maps of flooded wetlands for ac qui si tion dates: (A) map of flooded wetlands (29.03.2017); (B) map of flooded wetlands (2.10.2017); (C) map of flooded wetlands
(26.12.2017); (D) map of flooded wetlands (8.01.2018); (E) map of flooded wetlands (19.03.2018)
Fig . 7. Sub area B of the 10 m false col our com pos ite RNDTIGNDPIBMNDWI and the re sult ing maps of flooded wetlands for ac qui si tion dates: (A) map of flooded wetlands (29.03.2017); (B) map of flooded wetlands (2.10.2017); (C) map of flooded wetlands
(26.12.2017); (D) map of flooded wetlands (8.01.2018); (E) map of flooded wetlands (19.03.2018)
area is re lated to the im peded out flow of rain wa ter due to the flat to pog ra phy and poor sur face per me abil ity. Peat bogs show the most sta ble trend of sur face vari abil ity. On the other hand, the high est vari abil ity was ob served in flood wa ters along ditches and drain age chan nels, in which wa ter ac cu mu lates only for short pe ri ods af ter rains, but dur ing the dry sea son they be come dry. More over, the ob ser va tions have shown that a poorly lo cated drain age net work, which of ten in ter sects the nat u ral drain age routes of rain wa ter and thereby blocks out - flow, in ten si fies flood ing in the re search area. Our re sults show the ef fec tive ness of the method of re-nat u ral iza tion of the Kampinos Na tional Park Wetlands by stop ping the land use pro posed by Kopeæ et al. (2013).
CONCLUSIONS
The newly in tro duced Sen ti nel-2 sys tem pro vides high res o - lu tion multispectral im ages and a sat is fac tory time res o lu tion, thereby pro duc ing an im por tant data set for map ping of floods.
This pa per pro poses a new method of map ping of flood ing from Sen ti nel-2 im ages by cre at ing a 10 m false col our com pos ite RNDTIGNDPIBMNDWI in the con text of wet land ar eas. The wa ter fea tures needed for in ves ti ga tions of flooded ar eas are ex - tracted us ing the SWIR, green and red bands. The ex per i ment on the Sen ti nel-2 sub set from the Kampinos Na tional Park in Po land shows that the com bi na tion of the wa ter in di ca tors MNDWI, NDPI and NDTI is more ef fec tive in im prov ing the rec - og ni tion of flood ing, es pe cially in wetlands, than in di vid ual spec - tral wa ter in di ces. The idea of the pro posed con nec tion lies in the mu tual com ple ment ing of spec tral wa ter in di ces in the rec - og ni tion of flood ing. While the MNDWI in di ca tor best iden ti fies open wa ter, the NDPI cap tures veg e ta tion in wetlands and wa - ter, and the NDTI re duces the im pact of open soils that may be
con fused with tur bid wa ter res er voirs. As a re sult, fea tures are cre ated that are eas ier to dis tin guish be tween flooded ar eas and wa ter bod ies. How ever, the ap proach used does not work in for est ar eas and shows in fe rior ef fec tive ness dur ing pe ri ods of lux u ri ant veg e ta tion de vel op ment. The maps of the ex tent of flood ing in non-for est ar eas cre ated on the ba sis of the pro - posed method show sat is fac tory ac cu racy. In ad di tion, the time vari abil ity of ranges and the lo ca tion of flooded ar eas was tracked in this study. It has been seen that flood ing ar eas are lo - cated along the rows and drain age ca nals that cross the priv i - leged rain wa ter run off paths. Flood ing with a sta ble trend of vari abil ity is usu ally lo cated in the un drained de pres sions. Fu - ture stud ies should use ef fec tive im age sharp en ing al go rithms that will better ac com mo date flooded wetlands and the spec tral and spa tial char ac ter is tics of the Sen ti nel-2 im ag ery.
The ad van tage of this ap proach is the sep a rate rec og ni tion of wa ter res er voirs and ar eas flooded with wa ter with a trans lu - cent veg e tal cover, which en sures the dis tinc tion be tween flooded ar eas and wa ter res er voirs. In ad di tion, the use of the false col our com pos ite RNDTIGNDPIBMNDWI pro vides the abil ity to dis tin guish be tween flooded ar eas flooded and per ma nent wetlands. The re sults ob tained in this work show that the com - bined RNDTIGNDPIBMNDWI in dex cal cu lated on the ba sis of Sen ti - nel-2 data pro vides sat is fac tory ac cu racy of flood map ping and can be used on non-for est ar eas. An ad di tional ad van tage is the sim ple cal cu la tion method and quick map ping with lim ited re sources, which is of par tic u lar im por tance when mon i tor ing floods and clas si fi ca tion of wetlands on a re gional scale.
Ac knowl edg ments. This pa per is fi nan cially sup ported by the Pol ish Geo log i cal In sti tute – Na tional Re search In sti tute (Grant 61-8509-1701-00-0 and 62.9012.1954.00.0). The au - thor wishes to thank the re view ers, Prof. I. Nyambe and M. Ste - fouli, for their very con struc tive sug ges tions and com ments.
REFERENCES
Acreman, M., Holden, J., 2013. How wetlands af fect floods.
Wetlands, 33: 773–786.
Bourgeau-Chavez, L., Kasischke, E., Brunzell, S., Mudd, J., Smith, K., Frick, A., 2001. Anal y sis of space-borne SAR data for wet land map ping in Vir ginia Ri par ian Eco sys tems. In ter na - tional Jour nal of Re mote Sens ing, 22: 3665–3687.
Brisco, B., Kaper, M., Hirose, T., Tedford, B., Liu, J., 2011. Eval u - a tion of C-band po lar iza tion di ver sity and polarimetry for wet - land map ping. Ca na dian Jour nal of Re mote Sens ing, 37: 82–92.
Budzyñska, M., D¹browska-Zieliñska, K., Turlej, K., Malek, L., Bartold, M., 2011. Mon i tor ing przyrodniczy Bagien Biebrzañ -
skich z zastosowaniem teledetekcji (in Pol ish). Woda-Œro - dowisko- Obszary Wiejskie, 3: 39–64.
Butera, M.K., 1983. Re mote sens ing of wetlands. IEEE Trans ac - tions on Geoscience and Re mote Sens ing, 3: 383–392.
Chander, G., Markham, B.L., Helder, D.L., 2009. Sum mary of cur - rent ra dio met ric cal i bra tion co ef fi cients for Land sat MSS, TM, ETM+, and EO-1 ALI sen sors. Re mote Sens ing of En vi ron ment, 113: 893–903.
Chavez, P.S., Sides, S.C., An der son, J.A., 1991. Com par i son of 3 dif fer ent meth ods to merge multiresolution and multispectral
Ac cu racy in di ca tors Ac qui si tion date
29/03/2017 2/10/2017 26/12/2017 8/01/2018 19/03/2018
Kappa A
B
0.8700 0.8954
0.8334 0.8289
0.8905 0.8828
0.8579 0.8756
0.8971 0.9070
OA A
B
97.61%
94.91%
97.13%
94.28%
94.67%
96.02%
93.12%
97.78%
94.99%
96.79%
T a b l e 3 Kappa and over all ac cu racy (OA) of re sul tant flooded wetlands maps in sub ar eas A and B
data-Land sat tm and spot pan chro matic. Photogrammetric En - gi neer ing and Re mote Sens ing, 57: 295–303.
Chen, Q.L., Zhang, Y.Z., Ekroos, A., Hallikainen, M., 2004. The role of re mote sens ing tech nol ogy in the EU wa ter frame work di - rec tive (WFD). En vi ron men tal Sci ence & Pol icy, 7: 267–276.
Davranche, A., Lefebvre, G., Poulin, B., 2010. Wet land mon i tor ing us ing clas si fi ca tion trees and SPOT-5 sea sonal time se ries. Re - mote Sens ing of En vi ron ment, 114: 552–562.
Dong, Z.Y., Wang, Z.M., Liu, D.W., Song, K.S., 2014. Map ping wet - land ar eas us ing land sat-de rived NDVI and LSWI: a case study of west songnen plain, North east China. Jour nal of the In dian So ci ety of Re mote Sens ing, 42: 569–576.
Drusch, M., Del Bello, U., Carlier, S., Colin, O., Fernandez, V., Gascon, F., Hoersch, B., Isola, C., Laberinti, P., Martimort, P., Meygret, A., Spoto, F., Sy, O., Mar chese, F., Bargellini, P., 2012. Sen ti nel-2: ESA’s op ti cal high-res o lu tion mis sion for GMES op er a tional ser vices. Re mote Sens ing of En vi ron ment, 120: 25–36.
Du, Y., Xue, H.P., Wu, S.J., Ling, F., Xiao, F., Wei, X.H., 2011. Lake area changes in the mid dle Yang tze re gion of China over the 20th cen tury. Jour nal of En vi ron men tal Man age ment, 92:
1248–1255.
Du, Z.Q., Li, W.B., Zhou, D.B., Tian, L.Q., Ling, F., Wang, H.L., Gui, Y.M., Sun, B.Y., 2014. Anal y sis of Land sat-8 OLI im ag ery for land sur face wa ter map ping. Re mote Sens ing, 5: 672–681.
Dvorett, D., Da vis, C., Papes, M., 2016. Map ping and hydrologic at tri bu tion of tem po rary wetlands us ing re cur rent Land sat im ag - ery. Wetlands, 36: 431–443.
Feng, L., Hu, C.M., Chen, X.L., Cai, X.B., Tian, L.Q., Gan, W.X., 2012. As sess ment of in un da tion changes of Poyang Lake us ing MODIS ob ser va tions be tween 2000 and 2010. Re mote Sens ing of En vi ron ment, 121: 80–92.
Foody, G.M., 2002. Sta tus of land cover clas si fi ca tion ac cu racy as - sess ment. Re mote Sens ing of En vi ron ment, 80: 185–201.
Gao, B.C., 1996. NDWI - A nor mal ized dif fer ence wa ter in dex for re - mote sens ing of veg e ta tion liq uid wa ter from space. Re mote Sens ing of En vi ron ment, 58: 257–266.
Guyot, G., 1989. Sig na tures spectrales des sur faces naturelles.
Télédétection satellitaire, 5, Col. SAT, Ed. Paradigme.
Huang, C., Chen, Y., Wu, J.P., 2014a. DEM-based mod i fi ca tion of pixel-swap ping al go rithm for en hanc ing floodplain in un da tion map ping. In ter na tional Jour nal of Re mote Sens ing, 35:
365–381.
Huang, C.Q., Peng, Y., Lang, M.G., Yeo, I.Y., McCarty, G., 2014b.
Wet land in un da tion map ping and change mon i tor ing us ing Land sat and air borne LiDAR data. Re mote Sens ing of En vi ron - ment, 141: 231–242.
Huete, A., Liu, H., Batchily, K.V., Van Leeuwen, W., 1997. A com - par i son of veg e ta tion in di ces over a global set of TM im ages for EOS-MODIS. Re mote Sens ing of En vi ron ment, 59: 440–451.
Ilnicki, P., 2002. Torfowiska i torf (in Pol ish). AR Poznañ.
Is lam, M., Sado, K., 2006. Anal y ses of ASTER and spectroradiometer data with in situ mea sure ments for tur bid ity and trans par ency study of lake Abashri. In ter na tional Jour nal of Geo-In for ma tion, 2: 31–45.
Jensen, J.R., 1996. In tro duc tory Dig i tal Im age Pro cess ing, a Re - mote Sens ing Per spec tive. Sec ond edi tion. Prentice Hall, Up - per Sad dle River, New Jer sey.
Jones, K., Lanthier, Y., van der Voet, P., van Valkengoed, E., Tay - lor, D., Fernández-Prieto, D., 2009. Mon i tor ing and as sess - ment of wetlands us ing Earth Ob ser va tion: the GlobWetland pro ject. Jour nal of En vi ron men tal Man age ment, 90:
2154–2169.
Kasischke, E., Smith, K., Bourgeau-Chavez, L., Romanowicz, E., Brunzell, S., Rich ard son, C., 2003. Ef fects of sea sonal hydrologic pat terns in south florida wetlands on ra dar back scat - ter mea sured from ERS-2 SAR im ag ery. Re mote Sens ing of En - vi ron ment, 88: 423–441.
Kayastha, N., Thomas, V., Galbraith, J., Banskota, A., 2012. Mon - i tor ing wet land change us ing inter-an nual Land sat time-se ries data. Wetlands, 32: 1149–1162.
Kayranli, B., Scholz, M., Mustafa, A., Hedmark, A., 2009. Car bon stor age and fluxes within fresh wa ter wetlands: a crit i cal re view.
Wetlands, 30: 111–124.
Ko, B.C., Kim, H.H., Nam, J.Y., 2015. Clas si fi ca tion of po ten tial wa - ter bod ies us ing land sat 8 OLI and a com bi na tion of two boosted ran dom for est clas si fi ers. Sen sors, 15: 13763–13777.
Kopeæ, D., Michalska-Hejduk, D., Krogulec, E., 2013. The re la - tion ship be tween veg e ta tion and ground wa ter lev els as an in di - ca tor of spon ta ne ous wet land res to ra tion. Eco log i cal En gi neer - ing, 57: 242–251.
Krogulec, E., 2004. Ocena podatnoœci wód podziemnych na zanieczyszczenia w dolinie rzecznej na podstawie przes³anek hydrodynamicznych (in Pol ish). Wyd. Uniwersytetu Warszawskiego, Warszawa.
Krogulec, E., 2011. Charakterystyka uwarunkowañ hydroekologicznych (in Pol ish). In: Ochrona i renaturyzacja mokrade³ Kampinoskiego Parku Narodowego (eds. T.
Okruszko, W. Mioduszewski and L. Kucharski): 73–92.
Wydawnictwo SGGW, Warszawa.
Lacaux, J.P., Tourre, Y.M., Vignolles, C., Ndione, J.A., Lafaye, M., 2007. Clas si fi ca tion of ponds from highspatial res o lu tion re mote sens ing: Ap pli ca tion to Rift Val ley Fe ver ep i dem ics in Sen e gal.
Re mote Sens ing of En vi ron ment, 106: 66–74.
Lewiñski, S., 2007. Obiektowa klasyfikacja zdjêæ satelitarnych jako metoda pozyskiwania informacji o pokryciu i u¿ytkowaniu ziemi (in Pol ish). Instytut Geodezji i Kartografii, Seria monograficzna, 12. Warszawa.
Li, J.H., Chen, W.J., 2005. A rule-based method for map ping Can - ada’s wetlands us ing op ti cal, ra dar and DEM data. In ter na tional Jour nal of Re mote Sens ing, 26: 5051–5069.
Li, W.B., Du, Z.Q., Ling, F., Zhou, D.B., Wang, H.L., Gui, Y.M., Sun, B.Y., Zhang, X.M., 2013. A com par i son of land sur face wa - ter map ping us ing the nor mal ized dif fer ence wa ter in dex from TM, ETM plus and ALI. Re mote Sens ing, 5: 5530–5549.
Li, W., Qin, Y., Sun, Y., Huang, H., Ling, F., Tian, L., Ding, Y., 2016.
Es ti mat ing the re la tion ship be tween dam wa ter level and sur - face wa ter area for the Danjiangkou Res er voir us ing Land sat re - mote sens ing im ages. Re mote Sens ing Let ters, 7: 121–130.
Loveline, E., 2015. Im pacts of wet land deg ra da tion in Niger delta Ni ge ria and its sig nif i cance in flood con trol. In ter na tional Jour nal of Re mote Sens ing, 4: 177-184.
£achacz, A., 2004. Mokrad³a w krajobrazie – wybrane pojêcia (in Pol ish). Woda-Œrodowisko-Obszary-Wiejskie 4, 2a (11):
295–301.
Mar ti nez, J., Le Toan, T., 2007. Map ping of flood dy nam ics and spa - tial dis tri bu tion of veg e ta tion in the Am a zon Floodplain us ing multitemporal SAR data. Re mote Sens ing of En vi ron ment, 108:
209–223.
McFeeters, S.K., 1996. The use of the nor mal ized dif fer ence wa ter in dex (NDWI) in the de lin ea tion of open wa ter fea tures. In ter na - tional Jour nal of Re mote Sens ing, 17: 1425–1432.
Melack, J.M., Hess, L.L., 2010. Re mote sens ing of the dis tri bu tion and ex tent of wetlands in the Am a zon basin. Eco log i cal Stud ies, 210: 43–59.
Michalska-Hejduk, D., 2001. Stan obecny i kierunki zmian roœlinnoœci nieleœnej Kampinoskiego Parku Narodowego.
Monographia botanica, 89: 1–134.
Michalska-Hejduk, D., 2004. Najcenniejsze przyrodniczo obszary
³¹k i turzycowisk zachodniej czêœci Kampinoskiego Parku Narodowego oraz propozycje ich ochrony(in Pol ish). Parki Narodowe Rezerwaty Przyrody, 23: 203–218.
Michalska-Hejduk, D., Kopeæ, D., Kucharski, L., Kêb³owska, A., Otrêba, A., Kloss, M., Dembek, A., 2011. Roœlinnoœæ terenów mokrad³owych – stan zachowania i tendencje dynamiczne (in Pol ish). In: Ochrona i renaturyzacja mokrade³ Kampinoskiego Parku Narodowego (eds. T. Okruszko, W. Mioduszewski and L.
Kucharski): 119–141. Wydawnictwo SGGW, Warszawa.
Mitsch, W.J., 2009. Wet land eco sys tems: John Wiley & Sons.
Mitsch, W.J., Gosselink, J.G., 2015. Wetlands. 5th ed: 155–204.
Wiley, Hoboken, NJ, USA.
Monserud, R.A., Leemans, R., 1992. Com par ing global veg e ta tion maps with the Kappa sta tis tic. Eco log i cal Mod el ling, 62:
275–293.
Morandeira, N.S., Grings, F., Facchinetti, C., Kandus, P., 2016.
Map ping plant func tional types in floodplain wetlands: an anal y - sis of C-Band polarimetric SAR data from RADARSAT-2. Re - mote Sens ing, 8: 174.
Moser, L., Schmitt, A., Wendleder, A., Roth, A., 2016. Mon i tor ing of the lac Bam wet land ex tent us ing dual-po lar ized X-band SAR data. Re mote Sens ing, 8: 302.
Mwita, E., Menz, G., Misana, S., Becker, M., Kisanga, D., Boehme, B., 2013. Map ping small wetlands of Kenya and Tan - za nia us ing re mote sens ing tech niques. In ter na tional Jour nal of Ap plied Earth Ob ser va tion and Geoinformation, 21: 173–183.
Nandi, I., Srivastava, P.K., Shah, K., 2017. Floodplain map ping through sup port vec tor ma chine and op ti cal/in fra red im ages from Land sat 8 OLI/TIRS sen sors: case study from Varanasi.
Wa ter Re sources Man age ment, 31: 1157–1171.
Napiórkowska, M., 2014. Mon i tor ing wetlands eco sys tems us ing ALOS PALSAR (L-Band, HV) sup ple mented by op ti cal data: a case study of Biebrza Wetlands in North east Po land. Re mote Sens ing, 6: 1605–1633.
Okruszko, T., Mioduszewski, W., Kucharski, L., 2011. Ochrona i renaturyzacja mokrade³ Kampinoskiego Parku Narodowego (in Pol ish). Wydawnictwo SGGW, Warszawa.
Olszewski, A., Wierzbicki, A., Degórska, A., Ferchmin, M., Gudowicz, J., Lenartowicz, M., Otrêba, N., 2018. Raport stacji bazowej zintegrowanego monitoringu œrodowiska przyrodniczego „Po¿ary” za rok 2017 (in Pol ish). Maszynopis, KPN, Izabelin.
Ramsey, E.W., Laine, S.C., 1997. Com par i son of Land sat the matic map per and high res o lu tion pho tog ra phy to iden tify change in com plex coastal wetlands. Jour nal of Coastal Re search, 13:
281–292.
Seiler, R., Schmidt, J., Diallo, O., Csaplovics, E., 2009. Flood mon i tor ing in a semi-arid en vi ron ment us ing spa tially high res o - lu tion ra dar and op ti cal data. Jour nal of En vi ron men tal Man age - ment, 90: 2121–2129.
Selva, M., Aiazzi, B., Butera, F., Chiarantini, L., Baronti, S., 2015.
Hy per-sharp en ing: a first ap proach on SIM-GA data. In ter na - tional Jour nal of Ap plied Earth Ob ser va tion and Geoinformation and Re mote Sens ing, 8: 3008–3024.
Singh, K.V., Setia, R., Sahoo, S., Prasad, A., Pateriya, B., 2015.
Eval u a tion of NDWI and MNDWI for as sess ment of waterlog - ging by in te grat ing dig i tal el e va tion model and ground wa ter level. Geocarto In ter na tional, 30: 650–661.
Solovey, T., 2013. Zastosowanie metod teledetekcyjnych do identyfikacji obszarów podmok³ych na Nizinach Œrodkowopolskich (in Pol ish). Biulletyn Pañstwowego Instytutu Geologicznego, 454: 133–139.
Solovey, T., 2017. Iden ti fi ca tion of the Rozwarowo Marshes us ing ra dar re mote sens ing. Geographia Polonica, 90: 431–440.
Sun, F.D., Sun, W.X., Chen, J., Gong, P., 2012. Com par i son and im prove ment of meth ods for iden ti fy ing waterbodies in re motely sensed im ag ery. In ter na tional Jour nal of Re mote Sens ing, 33:
6854–6875.
Tobolski, K., 2003. Torfowiska na przyk³adzie Ziemi Œwieckiej (in Pol ish). Wyd. Towarzystwo Przyjació³ Dolnej Wis³y, Œwiecie.
Tucker, C.J., Sell ers, P.J., 1986. Sat el lite re mote sens ing of pri - mary pro duc tiv ity. In ter na tional Jour nal of Re mote Sens ing, 7:
1395–1416.
White, L., Brisco, B., Dabboor, M., Schmitt, A., Pratt, A., 2015.
A collection of SAR meth od ol o gies for mon i tor ing wetlands. Re - mote Sens ing, 7: 7615–7645.
Whyte, A., Ferentinos, K.P., Petropoulos, G.P., 2018. A new syn - er gis tic ap proach for mon i tor ing wetlands us ing Sen ti nels-1 and 2 data with ob ject-based ma chine learn ing al go rithms. En vi ron - men tal Mod el ling & Soft ware, 104: 40–54.
Wu, Q., 2018. GIS and re mote sens ing ap pli ca tions in wet land map - ping and mon i tor ing. Com pre hen sive Geo graphic In for ma tion Sys tems, 2: 140–157.
Xu, H.Q., 2006. Mod i fi ca tion of nor mal ised dif fer ence wa ter in dex (NDWI) to en hance open wa ter fea tures in re motely sensed im - ag ery. In ter na tional Jour nal of Re mote Sens ing, 27:
3025–3033.
Zomer, R.J., Trabucco, A., Ustin, S., 2009. Build ing spec tral li brar - ies for wetlands land cover clas si fi ca tion and hyperspectral re - mote sens ing. Jour nal of En vi ron men tal Man age ment, 90:
2170–2177.
