the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Annual forest maps in the contiguous United States during 2015–2017 from analyses of PALSAR-2 and Landsat images
Jie Wang
Xiangming Xiao
Yuanwei Qin
Jinwei Dong
Geli Zhang
Xuebin Yang
Xiaocui Wu
Chandrashekhar Biradar
Yang Hu
Abstract. Annual forest maps at a high spatial resolution are necessary for forest management and conservation. Large uncertainties remain among the existing forest maps, because of different forest definitions, satellite datasets, in-situ training datasets, and mapping algorithms. In this study, we generated annual forest maps and evergreen forest maps at a 30-m resolution in the Contiguous United States (CONUS) during 2015–2017 by integrating microwave data (Phased Array type L-band Synthetic Aperture Radar (PALSAR-2)) and optical data (Landsat) using Knowledge-based algorithms. The resultant PALSAR-2/Landsat-based forest maps (PL-Forest) were compared with five major forest datasets in the CONUS: (1) the Landsat tree canopy cover from Global Forest Watch datasets (GFW-Forest), (2) the Landsat Vegetation Continuous Field datasets (Landsat VCF-Forest), (3) the National Land Cover Database 2016 (NLCD-Forest), (4) the Japan Aerospace Exploration Agency (JAXA) forest maps (JAXA-Forest), and (5) the Forest Inventory and Analysis (FIA) data from the USDA Forest Service (FIA-Forest). The forest structure data (tree canopy height and canopy coverage) derived from the lidar observations of the Geoscience Laser Altimetry System (GLAS) onboard of NASA's Ice, Cloud, and land Elevation Satellite (ICESat-1) were used to assess the five forest datasets derived from satellite images. Using the forest definition by the Food and Agricultural Organization (FAO) of the United Nations, more forest pixels from the PL-Forest maps meet the FAO’s forest definitions than the GFW-, Landsat VCF-, and JAXA-Forest datasets. Forest area estimates from the PL-Forest were close to those from the FIA-Forest statistics but higher than the GFW-Forest, NLCD-Forest and lower than the Landsat VCF-Forest, which highlights the potential of using both PL-Forest and FIA-Forest datasets to support the FAO's Global Forest Resources Assessment. Furthermore, the PL-based annual evergreen forest maps (PL-Evergreen Forest) showed reasonable consistency with the NLCD product. Together with our previous work in South America and monsoon Asia, this study further demonstrates the potential of integrating PALSAR and Landsat images for developing annual forest maps and forest-type maps at high spatial resolution across the scales from region to the globe, which could be used to support FAO Global Forest Resources Assessments. The PL-Forest and PL-Evergreen Forest datasets are publicly available at https://doi.org/10.6084/m9.figshare.21270261 (Wang et al., 2022).
- Preprint
(5494 KB) - Metadata XML
- BibTeX
- EndNote
Jie Wang et al.
Status: open (until 27 Sep 2023)
-
CC1: 'Comment on essd-2022-339', cao zhiyue, 06 Jun 2023
reply
Annual PL-Forest and PL-Evergreen forest datasets are critical for forest resources assessment and carbon sink estimation. Particularly, the high spatial resolution forest datasets are relatively limited over national or larger spatial scales. This study proposed an interesting approach by combining PARSAR-2 and Landsat to generate the annual forest and evergreen forest maps. The accuracy has been improved more or less. However, this study was conducted for the CONUS. I am curious about the application of the approach. If it can be applied into other regions, such as the north region of China?
Citation: https://doi.org/10.5194/essd-2022-339-CC1 -
CC2: 'Reply on CC1', jie wang, 12 Jun 2023
reply
Thanks for your comments and questions. In this study, we developed an approach by integrating microwave data (PALSAR-2) and optical data (Landsat) to map annual forests and evergreen forests. It is a general approach that can be applied into other regions. In our pervious studies, we have combined PALSAR and Landsat/MODIS images for developing annual forest maps in South America and monsoon Asia. These studies suggested the robustness of the approach. As this study was conducted at a national scale, the accuracy of the resultant maps could be improved in a regional study by adjusting the thresholds of HV, Difference (HH-HV), Ratio (HH/HV), and NDVImax using the field samples from the specific region.
Citation: https://doi.org/10.5194/essd-2022-339-CC2
-
CC2: 'Reply on CC1', jie wang, 12 Jun 2023
reply
-
CC3: 'Comment on essd-2022-339', Rehman Khan, 21 Jun 2023
reply
It is a study on the annual forest mapping at 30-m spatial resolution at a national scale. This work involves a large amount of data processing. May I know how did you process the data? In Figure 9, the comparisons between your results and others were mainly distributed in the small regions. Could you provide more comparisons located in large regions?
Citation: https://doi.org/10.5194/essd-2022-339-CC3 -
CC4: 'Reply on CC3', jie wang, 26 Jul 2023
reply
Thanks for your positive comments. We appreciate your suggestions on improving the manuscript. We processed all the Landsat and PALSAR-2 images based on the Google Earth Engine (GEE) platform. We also used R to do data analysis and the software of Arcmap to generate all the maps. The algorithms for mapping forests and evergreen forests have been described in the manuscript in detail. Figure 9 shows the landscapes of Google Earth for six random locations to present the differences between our results and JAXA forests in 2016 at the pixel scale, as these two products have been produced considering radar signatures. The six sample areas really do not show the comparisons at large regions such as the eastern area of the CONUS. We will improve this point if we have a chance to revise it.
Citation: https://doi.org/10.5194/essd-2022-339-CC4
-
CC4: 'Reply on CC3', jie wang, 26 Jul 2023
reply
-
RC1: 'Comment on essd-2022-339', Anonymous Referee #1, 15 Aug 2023
reply
The preprint “Annual forest maps in the contiguous United States during 2015-2017 from analyses of PALSAR-2 and Landsat images” introduced 30 m spatial resolution forest maps for CONUS during 2015-2017 by integrating microwave data (PALSAR-2) and optical data (Landsat) using knowledge-based algorithms. The results were compared to four existing/operational forest products (GFV, VCF, NLCD, and JAXA) and FIA statistic data from USDA. The potential of PL-Forest was demonstrated by its forest area being close to the FIA-Forest statistics, but higher than GFV, NLCD, and lower than VCF. Also, PL-Forest evergreen forest showed reasonable consistency with the NLCD product.
The authors presented two main reasons for PL-Forest generation, including (1) “It is noticed that the high-spatial-resolution forest maps are relatively few for the years after 2010” and (2) “The combination of the optical and microwave data could take advantage of the optical remote sensing sensors that capture the light and forest canopy interaction and microwave sensors that capture the microwave and forest structure (tree trunk and branch) interaction without cloud contamination.”, “To date, no study has combined PALSAR and Landsat images during 2015-2017 to map annual forest distributions in the CONUS”. However, I think these two aspects are not significant enough to a demand for producing new annual forest maps for CONUS. First, as the authors mentioned, “In the United States, FIA and NLCD are the primary databases used by managers, researchers, and policymakers”. It is because these two datasets are high quality and have been carefully examined before releasing to the public. Second, while the US can map forest annually, the forest remains stable and the NLCD production interval of 3 years (2021, 2019, 2016, 2013, 2011, 2008, 2006, 2004, 2001) is set to capture meaningful changes that occur gradually over time. If there are disturbances (e.g., wildfires), they should be captured in regional or local regions by other advanced monitoring systems in the US (Fire and Emmision monitoring) and also captured within a 3-year period of 30 m NLCD. Thus, the authors may consider to:
(1) explain why we need annual PL-Forest maps.
(2) describe the uncertainties in the NLCD or other Landsat-based forest products that require the integration of microwave and optical data. I believe microwave sensors will contribute more to other regions such as tropical Amazon or monsoon Asia, but still unclear in the US. This could have been paid more attention to show the advantages of involving microwave data (e.g., high-cloudy areas) compared to the use of only optical data. The authors could show some examples of uncertainties in the NLCD and better performance of PL-Forest in the results. Also, keep in mind that the combination of Landsat 8/9 or the NASA harmonized Landsat and Sentinel-2 (HLS) currently gives more opportunity to obtain good observations.
(3) exhibit the forest decrease in the Midwest region in Fig 7.
Citation: https://doi.org/10.5194/essd-2022-339-RC1 -
AC1: 'Reply on RC1', Xiangming Xiao, 10 Sep 2023
reply
We appreciate your time and efforts to provide the valuable comments and suggestions to improve this manuscript. We have carefully addressed your comments and suggestions. To help read our responses, we labeled all the comments and suggestions at the corresponding locations and used the Track Change mode to show all the revisions in the revised manuscript. Please see our point-to-point responses to each comment and suggestion.
1. explain why we need annual PL-Forest maps.
Response: Thanks for your comments and suggestions. We clarified this point from three aspects: First, the combination of optical and microwave data has the potential to improve the accuracy of forest mapping. This study also suggested that the accuracy of PL-Forest maps reached more than 90 %. Second, we produced annual forest maps by satellite-based approaches with comparable accuracy with FIA data, which has the potential to improve the efficiency of forest resources survey. Finally, although most forests remain stable annually, we cannot ignore the annual abrupt changes, which affects the annual forest carbon fluxes and cannot be detected by a multi-year identification. PL-Forest maps provided accurate forest information annually, which can capture the inter-annual dynamics of forests.
We revised the manuscript in Lines of 75-101 by adding detail information of FIA and NLCD data. For example, “FIA is a field survey of forest plots and reports information on the status and trends of forests in the United States. A subset of plots is measured every year with revisit intervals of 5 to 10 years depending on the state (Burrill et al. 2021; Hoover et al. 2020). The NLCD provides updated datasets continuously every about three years, which was generated by change detection algorithms for only a time period and has a certain amount of commission errors (Jin et al. 2013a)”. “It remains untested about their application potential for the annual management of forest resources.”
2. describe the uncertainties in the NLCD or other Landsat-based forest products that require the integration of microwave and optical data. I believe microwave sensors will contribute more to other regions such as tropical Amazon or monsoon Asia, but still unclear in the US. This could have been paid more attention to show the advantages of involving microwave data (e.g., high-cloudy areas) compared to the use of only optical data. The authors could show some examples of uncertainties in the NLCD and better performance of PL-Forest in the results. Also, keep in mind that the combination of Landsat 8/9 or the NASA harmonized Landsat and Sentinel-2 (HLS) currently gives more opportunity to obtain good observations.
Response: Thanks for your comments and suggestions. We revised the manuscript see Lines of 90-101 following your suggestions.
First, as an optical sensor, the good observations of Landsat will be affected by cloud cover. Microwave data has strong penetration, which can avoid the influence of cloud coverage. However, the PALSAR-based forest maps often have commission errors caused by buildings, rocks, and high biomass crops. As a result, the combination of the optical and microwave data could take advantage of the optical remote sensing sensors that capture the light and forest canopy interaction and microwave sensors that capture the microwave and forest structure (tree trunk and branch) interaction without cloud contamination.
Additionally, an assessment study suggested that the complementarity of optical and SAR datasets improved the discriminative properties for forest mapping compared to the individual dataset (Lehmann et al. 2015). For example, uncertainties of Landsat-based forest maps could be caused by the re-planted areas with small- or medium-size trees or regions with some vegetation types like highland scrub. These regions could be identified correctly by PALSAR data (Lehmann et al. 2015).
Improved forest mappings have been reported in a number of studies by using integrated PALSAR and Landsat data in tropical regions (Lehmann et al. 2015; Reiche et al. 2015; Thapa et al. 2014), and PALSAR and MODIS data in monsoon Asia and several sample regions of the word (Qin et al. 2016b; Zhang et al. 2019). However, it remains unclear about the potential to improve the annual forest monitoring in the CONUS.
3. exhibit the forest decrease in the Midwest region in Fig 7.
Response: We appreciate your suggestions. We revised the figure with showing the forest decrease and increased from 2015 to 2017. See Fig7e, f.
-
AC2: 'Reply on RC1', Xiangming Xiao, 10 Sep 2023
reply
We appreciate your time and efforts to provide the valuable comments and suggestions to improve this manuscript. We have carefully addressed your comments and suggestions. To help read our responses, we labeled all the comments and suggestions at the corresponding locations and used the Track Change mode to show all the revisions in the revised manuscript. Please see our point-to-point responses to each comment and suggestion.
1. explain why we need annual PL-Forest maps.
Response: Thanks for your comments and suggestions. We clarified this point from three aspects: First, the combination of optical and microwave data has the potential to improve the accuracy of forest mapping. This study also suggested that the accuracy of PL-Forest maps reached more than 90 %. Second, we produced annual forest maps by satellite-based approaches with comparable accuracy with FIA data, which has the potential to improve the efficiency of forest resources survey. Finally, although most forests remain stable annually, we cannot ignore the annual abrupt changes, which affects the annual forest carbon fluxes and cannot be detected by a multi-year identification. PL-Forest maps provided accurate forest information annually, which can capture the inter-annual dynamics of forests.
We revised the manuscript in Lines of 75-101 by adding detail information of FIA and NLCD data. For example, “FIA is a field survey of forest plots and reports information on the status and trends of forests in the United States. A subset of plots is measured every year with revisit intervals of 5 to 10 years depending on the state (Burrill et al. 2021; Hoover et al. 2020). The NLCD provides updated datasets continuously every about three years, which was generated by change detection algorithms for only a time period and has a certain amount of commission errors (Jin et al. 2013a)”. “It remains untested about their application potential for the annual management of forest resources.”
2. describe the uncertainties in the NLCD or other Landsat-based forest products that require the integration of microwave and optical data. I believe microwave sensors will contribute more to other regions such as tropical Amazon or monsoon Asia, but still unclear in the US. This could have been paid more attention to show the advantages of involving microwave data (e.g., high-cloudy areas) compared to the use of only optical data. The authors could show some examples of uncertainties in the NLCD and better performance of PL-Forest in the results. Also, keep in mind that the combination of Landsat 8/9 or the NASA harmonized Landsat and Sentinel-2 (HLS) currently gives more opportunity to obtain good observations.
Response: Thanks for your comments and suggestions. We revised the manuscript see Lines of 90-101 following your suggestions.
First, as an optical sensor, the good observations of Landsat will be affected by cloud cover. Microwave data has strong penetration, which can avoid the influence of cloud coverage. However, the PALSAR-based forest maps often have commission errors caused by buildings, rocks, and high biomass crops. As a result, the combination of the optical and microwave data could take advantage of the optical remote sensing sensors that capture the light and forest canopy interaction and microwave sensors that capture the microwave and forest structure (tree trunk and branch) interaction without cloud contamination.
Additionally, an assessment study suggested that the complementarity of optical and SAR datasets improved the discriminative properties for forest mapping compared to the individual dataset (Lehmann et al. 2015). For example, uncertainties of Landsat-based forest maps could be caused by the re-planted areas with small- or medium-size trees or regions with some vegetation types like highland scrub. These regions could be identified correctly by PALSAR data (Lehmann et al. 2015).
Improved forest mappings have been reported in a number of studies by using integrated PALSAR and Landsat data in tropical regions (Lehmann et al. 2015; Reiche et al. 2015; Thapa et al. 2014), and PALSAR and MODIS data in monsoon Asia and several sample regions of the word (Qin et al. 2016b; Zhang et al. 2019). However, it remains unclear about the potential to improve the annual forest monitoring in the CONUS.
3. exhibit the forest decrease in the Midwest region in Fig 7.
Response: We appreciate your suggestions. We revised the figure with showing the forest decrease and increased from 2015 to 2017. See Fig7e, f.
-
AC1: 'Reply on RC1', Xiangming Xiao, 10 Sep 2023
reply
-
RC2: 'Comment on essd-2022-339', Anonymous Referee #2, 10 Sep 2023
reply
The authors have undertaken the task of developing 30-meter resolution forest maps across the United States for the years 2015-2017. They achieved this by integrating conventional optical data (Landsat) with microwave data (Phased Array type L-band Synthetic Aperture Radar, PALSAR-2), which improved the overall accuracy of the maps. Additionally, the authors distinguished between evergreen and deciduous forests using satellite-based time series water-related indices and vegetation greenness-related indices. The accuracy of their PALSAR-2/Landsat annual forest maps was validated using fully independent datasets and compared with five existing forest cover datasets, with their results demonstrating superior performance. The authors present new estimates of forest cover and suggest that their work can contribute to more accurate forest mapping and the investigation of climate change and anthropogenic impacts on forests. The paper is well-written, the methods are clearly articulated, and the results support the conclusions drawn. However, there are several aspects of the paper's narrative that could benefit from further clarification to enhance the overall quality of the study.
Firstly, the paper lacks novelty, as it appears to apply a similar method to another region, in comparison to the authors' previous studies. While the study is undoubtedly important for mapping forests in the CONUS region, it would be beneficial to highlight any novel aspects beyond the geographical and temporal scope.
We encourage the authors to provide additional information and engage in a more extensive discussion about the uncertainties associated with their methodology and the resulting forest maps. Furthermore, the limitation of the study's focus on the 2015-2017 period restricts its ability to monitor and analyze long-term forest changes, and this should be addressed.
The paper cites the existence of other studies that have used integrated PALSAR and satellite data to generate forest maps. It would be interesting to see a comparative analysis of the results from this study with those from previous efforts.
Several specific points require attention in the manuscript:
On line 160, the abbreviation "TOA" needs to be explained to readers who may not be familiar with the term.
The legend in Figure 4 (Line 184) should be made clearer to enhance the understanding of the figure.
On line 336, the reference to a "stronger relationship" could benefit from additional discussion and context.
Overall, while this study contributes valuable insights into forest mapping, addressing these concerns and incorporating additional details and comparisons could further elevate the paper's quality.
Citation: https://doi.org/10.5194/essd-2022-339-RC2 -
RC3: 'Comment on essd-2022-339', Anonymous Referee #3, 19 Sep 2023
reply
Summary
Mapping the temporal dynamics of forests at a high resolution is essential. Here Wang et al., combined the advantages of microwave and optical images, and generated high-resolution forests over continues US during 2015-2017 using an empirical threshold-based method. The manuscript is generally well organized, and easy to follow. My major concern is about its innovation. I also have few minor concerns listed as below.
Specific comments
- The major concern for me is the innovation for this study. What’s the advantage of the PL-Forests data comparing to other datasets? At least not very clear for the current version. Maybe necessary accuracy comparison (e.g., Table 2) is needed to show the accuracy advantage. Or, the major advantages of the data need to be clearly demonstrated compared to at least JAXA which also covers 2015-2017 at a high resolution.
- Line 52-53, which year?
- Line 79-80: it’s hard to say ‘few’ since all the data except one in Table 1 have years after 2010.
- Line 90-91, since a major contribution for this study is combining the advantages of microwave and optical images for mapping forest, necessary experiments are needed to show the accuracy improvement when combining these two datasets.
- Section 2.4, what’s the spatial and classification accuracy for the sample data in Fig.,4?
- Line 178-179, how did you identify land cover changes? If a sample changed from other land cover types to Forest during 2015-2017, shall we keep this sample or delete it? How many forest samples before and after the removing.
- Line 201-202: since the canopy height and cover will change over time, why there is no effect of time differences? I feel confused.
- Line 203-204, ‘the time differences could have small effect on the assessment’, what do you mean ‘time differences’? if it’s difference for canopy height or cover, it may not be necessarily correct.
- Figure4, clarify what do ‘NL’, ‘BL’, and ‘ML’ respectively mean in its caption.
- Line 210: The PL-Forest map is during 2015-2017, while the GFW product is in 2010. Unless the interannual changes are ignorable, the two products cannot be compared across different years to show their definition differences.
- Section 2.7, how to justify these thresholds are optimal?
- Line 265-266: why change ‘NFN’ to ‘NNN’? why not ‘FFM’ or ‘NFF’? similar to ‘FNF’
- Line 265-266: such rules seem very empirical. How many or what a percentage of pixels show a ‘NFN’ or ‘FNF’?
Citation: https://doi.org/10.5194/essd-2022-339-RC3
Jie Wang et al.
Data sets
30m Forest and Evergreen Forest in 2015 to 2017 Jie Wang https://doi.org/10.6084/m9.figshare.21270261
Jie Wang et al.
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
501 | 120 | 29 | 650 | 7 | 8 |
- HTML: 501
- PDF: 120
- XML: 29
- Total: 650
- BibTeX: 7
- EndNote: 8
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1