Elsevier

Remote Sensing of Environment

Volume 94, Issue 3, 15 February 2005, Pages 364-372
Remote Sensing of Environment

Comparison of time series tasseled cap wetness and the normalized difference moisture index in detecting forest disturbances

https://doi.org/10.1016/j.rse.2004.10.012Get rights and content

Abstract

Vegetation indices and transformations have been used extensively in forest change detection studies. In this study, we processed multitemporal normalized difference moisture index (NDMI) and tasseled cap wetness (TCW) data sets and compared their statistical relationships and relative efficiencies in detecting forest disturbances associated with forest type and harvest intensity at five, two and one year Landsat acquisition intervals. The NDMI and TCW were highly correlated (>0.95 r2) for all five image dates. There was no significant difference between TCW and NDMI for detecting forest disturbance. Using either a NDMI or TCW image differencing method, when Landsat image acquisitions were 5 years apart, clear cuts could be detected with nearly equal accuracy compared to images collected 2 years apart. Partial cuts had much higher commission and omission errors compared to clear cut. Both methods had 7–8% higher commission and 12–22% higher omission error to detect hardwood disturbance when it occurred in the first year of the 2-year interval (as compared to 1-year interval). Softwood and hardwood change detection errors were slightly higher at 2-year Landsat acquisition intervals compared to 1-year interval. For images acquired 1 and 2 years apart, NDMI forest disturbance commission and omission errors were slightly lower than TCW. The NDMI can be calculated using any sensor that has near-infrared and shortwave bands and is at least as accurate as TCW for detecting forest type and intensity disturbance in biomes similar to the Maine forest, particularly when Landsat images are acquired less than 2 years apart. Where partial cutting is the most dominant harvesting system as is currently the case in northern Maine, we recommend images collected every year to minimize (particularly omission) errors. However, where clear cuts or nearly complete canopy removal occurs, Landsat intervals of up to 5 years may be nearly as accurate in detecting forest change as 1 or 2 year intervals.

Introduction

The tasseled cap transformation has been widely used for vegetation mapping and monitoring land cover change (Bauer et al., 1994, Cohen & Spies, 1992, Cohen et al., 1995, Collins & Woodcock, 1996, Dymond et al., 2002, Fiorella & Ripple, 1993a, Fiorella & Ripple, 1993b, Franklin et al., 2002, Skakun et al., 2003). This transform not only provides a mechanism for reducing data volume with minimal information loss but its spectral features can also be directly associated with the important physical parameters of the land surface (Crist & Cicone, 1984, Crist & Kauth, 1986, Crist et al., 1986). The tasseled cap transformation of Landsat thematic mapper (TM) consists of six multispectral features, all of which could be potentially differentiated in terms of stability and change in a multitemporal data set. The first three features usually account for the most variation in a single-date image (Collins & Woodcock, 1996, Crist & Kauth, 1986, Crist, 1985). These first three features have been labeled brightness, greenness and wetness, respectively. The third feature, wetness, has been shown to be sensitive to soil and plant moisture (Crist & Cicone, 1984) and vegetation structure (Cohen & Spies, 1992, Cohen et al., 1995, Fiorella & Ripple, 1993b).

Tasseled cap wetness (TCW) contrasts the sum of the visible and near-infrared bands with the sum of the shortwave infrared bands (SWIR; Crist & Cicone, 1984). The TCW emphasizes both Landsat TM5 (1.55–1.75 μm) and TM7 (2.08–2.35 μm; Collins & Woodcock, 1996). Cohen and Spies (1992) attempted to distinguish old growth and mature forests in the Pacific Northwest using Landsat TM imagery. The tasseled cap brightness and greenness features did not separate old growth and mature forests due to their sensitivity to topography. They reported that the TCW was less sensitive to topographic effects and better correlated with old growth than all other single TM bands and most band ratios. Fiorella and Ripple (1993b) found that TCW and the TM 4/5 ratio were the best band transformations for distinguishing between old growth and mature forest. Cohen et al. (1995) reported that TCW was an important indicator of maturity and structure in closed canopy forest stands and appeared to be more responsive to the interaction between the water content and the structure of the canopy.

Collins and Woodcock (1996) analyzed multitemporal Landsat data and found change in TCW to be a good indicator of conifer mortality and the most consistent single indicator of forest change due to its capture of SWIR changes. The TCW has been applied to detect change caused by bark beetles and harvesting practices in northern forest studies in Canada (Franklin et al., 2000, Franklin et al., 2002, Franklin et al., 2003, Skakun et al., 2003). Franklin et al. (2000) reported that the accuracy of partial harvest change detection, using a variation of the TCW (the enhance wetness difference index), approached 71% over a full range of change conditions in southeastern New Brunswick. Partial harvests had higher forest change detection errors compared to clear cuts.

The normalized difference moisture index (NDMI) is derived from the near-infrared (NIR), e.g., Landsat TM or enhanced thematic mapper plus (ETM+) band 4 and the SWIR band 5 (Eq. (1)).NDMI=NIR(4)SWIR(5)NIR(4)+SWIR(5)

Horler and Ahern (1986) reported that the SWIR bands, compared to others, explained more information about forest structure in conifer and hardwood forests in Western Ontario, Canada. Nemani et al. (1993) applied a SWIR correction factor for the NDVI to improve the relation between NDVI and leaf area index (LAI). Although few vegetation studies have applied the NDMI, several have used the simple ratio TM 4/5 or 5/4 which is similar (Cohen & Fiorella, 1998, Cohen & Spies, 1992, Fiorella & Ripple, 1993b, Sader, 1989, Vogelmann & Rock, 1988). Vogelmann and Rock (1988) evaluated Landsat TM data for its ability to detect and measure damage to spruce-fir stands and concluded that TM Band 5/4 ratio correlated well with ground observations of forest damage defined as percent foliar loss. Hunt and Rock (1989) found that TM 5/4 was linearly correlated to leaf relative water content with each species having a different regression equation. Fiorella and Ripple (1993b) reported that TM 4/5 was highly correlated to wetness (r2=0.97) and had higher correlation with stand age than the tasseled cap indices.

Hardisky et al. (1983) reported that the NDMI was highly correlated with canopy water content and more closely tracked changes in plant biomass and water stress than did the NDVI. Wilson and Sader (2002) compared NDMI and NDVI to detect forest changes at different Landsat acquisition intervals in Maine, USA. In all classification trials, the NDMI change maps had a higher overall accuracy than the NDVI change maps. They concluded that the higher accuracy of the NDMI change maps was due to an increased ability to detect lighter disturbances including partial cuts.

Previous studies suggest that partial forest disturbances are more difficult to detect or are detected with higher error than complete canopy removal (Franklin et al., 2000, Muchoney & Haack, 1994, Sader et al., 2003, Skakun et al., 2003, Wilson & Sader, 2002). With longer gaps between satellite data acquisitions, higher forest change detection errors can be expected (Kimes et al., 1998, Lunetta et al., 2004, Wilson & Sader, 2002). Wilson and Sader (2002) recommended Landsat TM imagery collected less than 3 years apart to detect forest harvesting and other changes that do not remove the entire overstory canopy. In Canada's Acadian forest region, Franklin et al. (2000) suggested annual change detection as an appropriate strategy to eliminate some of the change classification errors and uncertainty caused by high variability within the change classes. To our knowledge, no studies have examined the effect of satellite data gaps on errors in detecting complete and partial forest changes when they occur in both coniferous (softwood) and deciduous (hardwood) forest types.

Over 96% of the Maine forest harvest volume in recent years has been accomplished through partial harvesting (Maine Forest Service, Silvicultural activities reports: 1999–2002); therefore, techniques that improve the detection and monitoring of partial forest disturbances are particularly relevant and challenging. The NDMI is a lesser known and utilized vegetation index in disturbance detection studies than NDVI or TCW. Because NDMI appears to be strongly correlated to TCW, the NDMI should also be a good index for detecting forest disturbances (Wilson & Sader, 2002). The primary purpose of this study is to examine the quantitative relationship between TCW and NDMI and compare associated commission and omission errors in detecting forest disturbances in a northern forest environment. We compared TCW and NDMI from three aspects: (1) forest disturbance by type (hardwood versus softwood); (2) forest disturbance intensity (clear cut versus partial cut); and (3) Landsat acquisition interval effect on detecting forest disturbances by type and intensity class.

The study site is approximately 499,176 ha and includes part or all of 84 townships in northern Maine, USA (Fig. 1). This Acadian forest occupies the northern boundary of temperate forest and southern edge of boreal forest and has a significant marine climatic influence (Loo & Ives, 2003). The upland vegetation is composed of both softwood and hardwood dominant and mixed stands under a variety of age classes from recent clear cut to pole-size and mature forest types. There are very few old growth stands more than 150 years old (Maine Forest Service, 1999a). Soils are derived mainly from glacial till. The terrain is relatively flat to rolling with occasional low mountains and includes abundant lakes, ponds, streams with associated wetland vegetation. The forest ownership is mostly large industrial corporations, family corporations, land management and investment companies and conservation organizations.

The mainly private forests have been actively logged for over 100 years, and there are essentially no urban features or development in the study area (Seymour, 1992). Clear cuts (complete or nearly complete canopy removal) were the dominant forest harvesting practice prior to the 1990s; however, partial harvesting (ranging from approximately 30–80% of the forest volume removed in one harvest) has become dominant over the past decade (Maine Forest Service, 1999b). Partial harvest systems include primarily selection and shelterwood silvicultural systems. The study area (Fig. 1) is the same as reported by Wilson and Sader (2002). High-quality time series Landsat satellite images were available to support this study.

Section snippets

Data acquisition and preprocessing

The Landsat scenes were selected from the overlap region of Path 12, Row 28 and Path 11, Row 28 of the Worldwide Reference System. Three dates of Landsat 5 thematic mapper (1988, 1991 and 1993) and three dates of Landsat 7 ETM+(2000, 2001 and 2002) satellite imagery were acquired in the spring and summer months, representing hardwood “leaf on” conditions. A multidate change detection map (1988–1991–1993) developed in a previous project (Wilson & Sader, 2002) and a 1993 Maine vegetation map (

Linear regression

The results of linear regression between TCW and NDMI images at five dates are listed in Table 1. The TCW explains more than 98% variation of NDMI for four of the five image dates (96% for 2001). This agrees closely with Fiorella and Ripple (1993b) who reported 0.97 r2 between TCW and the simple ratio (TM4/5) on one date of Landsat TM image in Pacific Northwest forest study area. The reason that the 2001 regression equation is different from the others can be explained by the May 25 (early

Conclusions

The TCW and NDMI were highly correlated (>0.95) for all five image dates. Both NDMI and TCW are strongly influenced by NIR (TM 4) and SWIR (TM 5); however, TCW also includes some information from visible bands. By comparing forest harvest type, harvest intensity and intervals in Landsat acquisitions, the differences in change detection commission and omission errors were observed. There was no significant difference between TCW and NDMI for detecting forest disturbances. NDMI produced similar

Acknowledgements

This research was supported by the Maine Agricultural and Forest Experiment Station Grant ME-09608, MAFES Misc. Publ. #2768.

References (39)

  • M.E. Bauer et al.

    Satellite inventory of Minnesota forest resources

    Photogrammetric Engineering and Remote Sensing

    (1994)
  • W.B. Cohen et al.

    Comparison of methods for detecting conifer forest change with Thematic Mapper imagery

  • W.B. Cohen et al.

    An efficient and accurate method for mapping forest clearcuts in the Pacific Northwest using Landsat imagery

    Photogrammetric Engineering and Remote Sensing

    (1998)
  • W.B. Cohen et al.

    Estimating the age and structure of forests in a multi-ownership landscape of western Oregon, U.S.A.

    International Journal of Remote Sensing

    (1995)
  • E.P. Crist et al.

    A physically-based transformation of thematic mapper data—the TM tasseled cap

    IEEE Transactions on Geoscience and Remote Sensing

    (1984)
  • E.P. Crist et al.

    The tasseled cap de-mystified

    Photogrammetric Engineering and Remote Sensing

    (1986)
  • E.P. Crist et al.

    Vegetation and soils information contained in transformed thematic mapper data

  • Erdas
  • M. Fiorella et al.

    Analysis of conifer forest regeneration using Landsat thematic mapper data

    Photogrammetric Engineering and Remote Sensing

    (1993)
  • Cited by (353)

    • Not-so-random forests: Comparing voting and decision tree ensembles for characterizing partial harvest events

      2023, International Journal of Applied Earth Observation and Geoinformation
    View all citing articles on Scopus
    View full text