Project

General

Profile

Wiki ยป

Tuanmu updates

2014-04-15

What I did in past week(s) and I'm working on now

  • The consensus land cover paper has been accepted and will be published online soon.
  • I have finished the second draft of heterogeneity metrics manuscript and hopefully I will distribute it to the group to get some feedback soon.
  • I have generated the EVI-based texture measures at a coarser resolution (2.5 arc minutes, ~5 km). They, together with the 30-arc-second products, are using to examine the heterogeneity-diversity relationship at different scales.
  • I have been generating the EVI-based texture measures for the time period from 2009 to 2013. They will be used to examine the utility of the metrics for monitoring biodiversity changes over time.
  • I tried to generate the texture measures using Google Earth Engine, but it was not successful. Earth Engine has limited function to aggregate pixel values to a coarser resolution and how it resamples images is not clear. So I decided to stick with my bash and python scripts for generating the metrics.

What's next

  • I hope to be able to submit the heterogeneity metrics manuscript soon.
  • I will examine the heterogeneity-diversity relationship across spatial scales and examine the utility of the heterogeneity metrics for monitoring biodiversity changes over time.

2014-02-04

What I did in past week(s) and I'm working on now

  • I have received comments on the consensus land cover manuscript from another round of review. I have been working on the revisions.
  • I have re-processed the MODIS EVI data and calculated texture measures after removing the effects of snow cover. I have conducted the evaluation for the new datasets. Since snow cover only affects the metrics at high latitudes and high elevations, re-processing the data did not change the conclusions.
  • I have been generating the EVI-based texture measures with different grain sizes and examining the relationship between species richness and habitat heterogeneity across scales.
  • I have generated a new heterogeneity metric based on the variability of remotely sensed vegetation phenology for the conterminous US.

What's next

  • I will re-submit the revised manuscript on consensus land cover and finish the texture measures manuscript.
  • I will generate the EVI-based texture measures for different time periods and examine their utility for modeling biodiversity changes.
  • I will modify the approach for generating phenology-based metrics to make it able to be scaled up to the global scale.
  • I will explore the way of generating the heterogeneity metrics on Google Earth Engine.

2013-12-03

What I did in past week(s) and I'm working on now

  • Revised manuscript on consensus land cover has been submitted.
  • I have generated global layers of 12 texture-based and 12 landcover-based heterogeneity metrics.
  • I have been working on the manuscript about MODIS-based texture measures.
  • I have been developing new heterogeneity metrics based on land surface phenology portrayed by a time series of MODIS data.
  • I am testing phenology-based heterogeneity metrics for their utility to detect temporal dynamics of biodiversity.

What's next

  • I will finish the texture measures manuscript, which is the priority.
  • I will finish the test on phenology-based metrics and modify the approach to make it able to be scaled up to the global scale.

2013-09-24

What I did in past week(s)

  • I have submitted the revised MS on the consensus land cover data.
  • I have generated several global data layers of habitat heterogeneity, including first- and second-order texture measures derived from 250m MODIS EVI, metrics derived from the 1km consensus land cover data, 300m GlobCover and 500m MODIS land cover data, and metrics derived from 90m DEM.
  • I have been comparing the usefulness of the above-mentioned heterogeneity metrics for modeling bird species richness and functional diversity using spatial autoregressive models.

What I'm working on now

  • I am evaluating the usefulness of different heterogeneity metrics for modeling biodiversity at the route-level scale.

What obstacles are blocking progress

  • No major obstacle

What's next

  • I will examine the scale effects (different window sizes for metric calculation, different sizes of buffer areas around survey stops, and/or stop-level vs. route-level) on the relationship between biodiversity and habitat heterogeneity.
  • I will draft a manuscript on the comparison of heterogeneity metrics.
  • I will explore MODIS-derived metrics which containing the information on temporal heterogeneity of habitat.

2013-07-16

What I did in past week(s)

  • I have scaled up the approaches for calculating EVI-based heterogeneity metrics to the global scale and written scripts (bash+python) for the process. I have generated 11 global (85N - 70S) land-cover heterogeneity data layers, each of which is composed of 144 20x20 degree (or 15x20 for the region from 85N - 70N) tiles at 1km resolution.
  • I have been using BBS datasets for evaluating the usefulness of the EVI-based heterogeneity metrics for modeling bird species richness at the continental scale (conterminous US). In addition to GLM, I have also built spatial autoregressive models to account for spatial autocorrelation. Results showed that some first- and second-order texture metrics are particularly useful for modeling bird species richness.
  • I have been working on a revision of the manuscript on the consensus land cover data. I have re-done the generation of the consensus dataset with and without the DISCover product and added some additional evaluations.

What I'm working on now

  • I am generating global data layers of other heterogeneity metrics.
  • I am revising the text of the consensus land cover manuscript.

What obstacles are blocking progress

  • No major obstacle

What's next

  • I will finish the revision of the consensus land cover manuscript.
  • I will use the canopy height and biomass datasets for evaluating the metrics.
  • Besides species richness, I will also look at species compositions and functional diversity and evaluate their relationships with habitat heterogeneity.
  • I will examine the scale effects (different window sizes for metric calculation, different sizes of buffer areas around survey stops, and/or stop-level vs. route-level) on the relationship between biodiversity and habitat heterogeneity.

2013-06-18

What I did in past week(s)

  • I have been modeling bird species richness in the conterminous US using heterogeneity metrics derived from a time series of MODIS EVI, and compared the usefulness of those metrics with that of metrics derived from land cover data (the consensus land cover data, GlobCover and MODIS land cover products). RMSE and AIC showed that the models built with the EVI-based metrics had better performance. The models also indicated that some first- and second-order texture metrics are particularly useful for modeling bird species richness.
  • I have been scaling up the approaches for calculating EVI-based heterogeneity metrics to the global scale. Some issues (e.g., projections, tiling and automation) emerged during the process and I have been working on solutions for them.
  • I have obtained and compiled a new version of 30m datasets of canopy height and aboveground biomass for the conterminous US. These datasets will be used for quantifying spatial heterogeneity of vegetation biophysical characteristics and used as reference data for evaluating the EVI-based heterogeneity metrics.

What I'm working on now

  • I am exploring ways to account for spatial autocorrelation in the bird species richness models.
  • I am generating MODIS EVI-based heterogeneity metrics at the global scale.

What obstacles are blocking progress

  • Some additional steps are needed to scale up the metric generation approach, and I need to automate those processes. The processes related to GEE may be more challenging.

What's next

  • I will use the canopy height and biomass datasets for evaluating the metrics.
  • Besides species richness, I will also look at species compositions and functional diversity and evaluate their relationships with habitat heterogeneity.
  • I will examine the scale effects (different window sizes for metric calculation, different sizes of buffer areas around survey stops, and/or stop-level vs. route-level) on the relationship between biodiversity and habitat heterogeneity.
  • I will explore the potential of MODIS EVI-based heterogeneity metrics for modeling biodiversity changes over time.

2013-06-03

What I did in past week(s)

  • I have calculated several heterogeneity metrics from the consensus land cover data for the globe and calculated texture metrics based on EVI values derived from a 10-year time series of MODIS data for the conterminous US.
  • I have examined the correlations between those heterogeneity metrics and bird species richness obtained from the BBS stop-level dataset and started to evaluate and compare the usefulness of the metrics for modeling bird species richness.
  • I have been exploring the ability of some MODIS-derived metrics to quantify both spatial and temporal heterogeneity of land cover (vegetation).

What I'm working on now

  • I am modeling bird species richness with different types of heterogeneity metrics (i.e., metrics derived from categorical land cover data, metrics derived from the proportional consensus land cover data, and texture metrics derived from MODIS EVI) and comparing their utility for the modeling.

What obstacles are blocking progress

  • There are some issues about resampling and reprojection in Google Earth Engine.

What's next

  • Besides species richness, I will also look at species compositions and functional diversity and evaluate their relationships with habitat heterogeneity.
  • I will examine the scale effects (different size of buffer areas around survey stops and/or stop-level vs. route-level) on the relationship between biodiversity and habitat heterogeneity.
  • I will explore the potential of MODIS-derived heterogeneity metrics for modeling biodiversity changes over time.

2013-04-22

What I did in past week(s)

  • I mainly worked on the consensus land cover manuscript in the past couple months and submitted it to GEB last week.
  • I gave an oral presentation about the work on consensus land cover in a seminar at Yale in March and had a talk at the landscape ecology conference in Austin last week ([[https://projects.nceas.ucsb.edu/nceas/documents/50]]).
  • I generated texture measures for Oregon based on the MODIS NDVI product and started to explore new metrics for capturing temporal and/or spatio-temporal heterogeneity of land cover.

What I'm working on now

  • I am familarizing myself with Google Earth Engine and exploring different metrics derived from MODIS time series data for quantifying spatio-temporal heterogeneity of land cover.
  • I am reviewing the literature about remote sensing-based metrics which capture spatial and temporal heterogeneity of land cover or vegetation and working on a review paper on this topic.

What obstacles are blocking progress

  • Good global composites of Landsat images are still needed.

What's next

  • I will compare texture metrics derived from Landsat images with those derived from MODIS data and evaluate their effectiveness at capturing vegetation heterogeneity and modeling bird species richness.
  • I will evaluate the suitability of using Google Earth Engine for generating texture metrics from Landsat and MODIS image composites.
  • I will continue to review the related literature and work on the review paper.

2013-02-11

What I did in past week(s)

  • I have obtained Landsat images (88 images in total) from the GLS 2005 collections for Venezuela and calculated NDVI from those images. I have found many pixels with missing values in this region due to frequent cloud cover and scan line failure of Landsat 7 (see obstacles below).
  • I continued to work on the revision of the consensus land cover manuscript. I have been adding more accuracy metrics for data evaluations and including the MODIS continuous vegetation fields product in the evaluations.
  • I have obtained the MODIS vegetation index product for Oregon and started to explore new metrics for capturing temporal and/or spatio-temporal heterogeneity of land cover.

What I'm working on now

  • I am calculating texture metrics for Venezuela to examine to what extent the missing values will affect the usefulness of those metrics for capturing land cover heterogeneity.
  • I am looking for alternative Landsat image collections (e.g., Google Earth Engine) or remotely sensed data (e.g., MODIS) for quantifying land cover heterogeneity.

What obstacles are blocking progress

  • It is expected that prevalent missing values of Landsat images in the region of Venezuela will significantly reduce the number of valid or good-quality pixels in the images of texture metrics. Alternative Landsat image sources or other remotely sensed data may be needed to resolve this issue.

What's next

  • I will calculate texture metrics from MODIS NDVI data and examine their effectiveness on capturing spatial heterogeneity of land cover by comparing them with the texture metrics derived from Landsat images for Oregon.
  • I will explore and develop metrics capturing spatio-temporal heterogeneity of land cover based on MODIS NDVI data.
  • I will familiarize myself with Google Earth Engine and available data in it, and access the possibility of generating spatial heterogeneity metrics from those data.

2013-01-14

What I did in past week(s)

For consensus land cover data:

  • I have evaluated the ability of the consensus land cover dataset to capture sub-pixel land cover information obtained from 30-m NLCD data. I conducted the evaluation based on both NLCD1992 and NLCD2001 data to examine whether land cover changes in the 1990s affect the evaluation. Results indicate that the consensus land cover dataset can better capture the fine-grain land cover information than any of the four original land cover products. Using different NLCD data as reference data didn't change the conclusion.
  • I have obtained presence/absence data for selected bird species from BBS datasets at the stop level and used the data for evaluating the utility of the consensus land cover dataset for modeling species distributions. I have built both deductive and inductive species distribution models for the evaluation. Results show that in general the consensus dataset suepasses all original land cover products in terms of the utility for modeling species distribution.
  • I have generated an additional consensus land cover dataset with only GlobCover, MODIS and GLC2000 (i.e., without DISCover, which was derived from 1992-1993 AVHRR data). I have compared this additional dataset with the original consensus dataset to examine whether including the older DISCover product introduced additional uncertainty and thus reduced performance of the consensus dataset on capturing sub-pixel land cover information and modeling species distributions. Results show that incorporating the DISCover product doesn't have significant negative effects on the consensus dataset. On the contrary, the DISCover dataset contains some useful land cover information which may improve the utility of the consensus dataset for species distribution modeling.
  • I have put together all results of these analyses and have finished the first draft of a manuscript describing the study.

For habitat heterogeneity

  • I have updated the procedure to generate texture measures from Landsat NDVI images to include more detailed information on NDVI values. I also have updated the procedure to generate more landscape metrics from the NLCD data with more land cover classes.
  • I have re-run the analyses of comparing the usefulness of texture measures and landscape metrics for predicting observed spatial heterogeneity of vegetation structure and for predicting bird species richness. I removed some first-order texture measures, which reflect overall productivity in a pixel. The removal results in a more fair comparison between the texture measures and landscape metrics since the latter cannot capture overall productivity.
  • I have also generated texture measures and landscape metrics in two time periods and used them to build models for predicting bird species richness in the two time periods.
  • Results show that models built with texture measures have better ability to predict observed spatial heterogeneity of canopy height and biomass, compared to the models built with landscape metrics. Results also indicate that while models built with two types of metrics have similar performance in predicting bird species richness in the same time period when the models were built, using texutre measures can increase model's transferability over time.
  • I have put together all these analyses in a poster and present it at the IBS meeting.

What I'm working on now

For consensus land cover data:

  • The manuscript is under internal reviews.

For habitat heterogeneity:

  • I am refining the analyses to control the effects of overall productivity on the spatial heterogeneity of vegetation structure and bird species richness. I have started to write a manuscript reporting the results of the habitat heterogeneity study.

What obstacles are blocking progress

  • Previously mentioned difficulty has been resolved.
  • I need to familiarizing myself with parallelization in Python.

What's next

  • I will finish the data analyses for the habitat heterogenity study and finish the manuscript.
  • I will start to revise the Python code for generating texture measures to parallel the processes, and start to process the images for Venezuela.

2012-11-19

What I did in past week(s)

  • In the past weeks, I have been focused on the updating and evaluating the consensus land cover data.
  • I have fixed a bug in the code for generating the consensus land cover dataset and re-processed the data.
  • I have obtained fine-grain land cover information from NLCD2006 and NLCD1992 datasets.
  • I have used the fine-grain data to evaluate the ability of the consensus dataset to capture sub-pixel land cover information.
  • I have evaluated the utility of the consensus dataset for species distribution modeling using species presence data from GBIF.

What I'm working on now

  • I am exploring ways to obtain species presence/absence information from the Breeding Bird Survey dataset and use the data for better evaluate the utility of the consensus dataset for modeling species distribution.
  • I am also writing up the methods and results of generating and evaluating the consensus data.
What obstacles are blocking progress
  • Presence data obtained from GBIF only allows me to compare species distribution models built with different land cover datasets based on omission errors. Presence/absence data from BBS may provide a better dataset for the evaluation.
  • The current consensus data contain land cover information from the DISCover dataset, which is 10 years older than the other input products. While it may be worth incorporating land cover information derived from more diverse remotely sensed images, land cover changes during the 10-year period may introduce uncertainty. It may be needed to assess whether incorporating the older product is a plus for the consensus dataset.

What's next

  • I will build species distribution models using the BBS data.
  • I will also generate species distribution maps using deductive approaches and validate the maps using the BBS data.
  • I will explore ways to assess whether it is worthy including the older DISCover dataset.

2012-10-8

What I did this past week

For the consensus dataset:
  • I have finished the work on updating and generating a 1-km consensus land cover dataset from four existing global land cover products (DISCover, GLC2000, MODIS and GlobCover).
  • I have evaluated the ability of the consensus land cover dataset to capture sub-pixel land cover infomration using 30-m NLCD data as a reference.
  • I have started to write up the methods and results of the dataset generation and evaluation.
For the habitat heterogeneity:
  • I have built regression models to evaluate the ability of the texuture measures derived from Landsat imagery to model bird species richness obtained from BBS data.
  • I have re-calculated landscape metrics from NLCD data with more detailed information on land cover classes for Oregon and have evaluated their ability to capture vegetation heterogeneity and model bird species richness.
What I'm working on now
  • I am looking at available species distribution data (including range maps and locality data) to select several bird species that are suitable for evaluating the utility of the consensus dataset for species distribution modeling.
What obstacles are blocking progress
  • None

What's next

For the consensus dataset:
  • I will compare the species distribution models built with the consensus datasets vs. those built with other land cover products to examine whether the consensus dataset provides better land cover information for species distribution modeling.
  • I will finish the write-up of the methods and results for the generation and evaluation of the consensus dataset.
For the habitat heterogeneity:
  • I will incorporate spatial components into the regression models of BBS bird species richness to account for spatial autocorrelation.
  • I will start to calculate texture measures for the second test site (i.e., Venezuela).

2012-09-10

What I did this past week
  • I have obtained four global land cover products, pre-processed the data, and finished the coding for generating consensus land cover datasets from them.
  • I have tested the code by generating consensus land cover datasets (12 generalized land cover classes) for Oregon at 500m and 1km resolutions.
  • I have started to explore the possibility of using Forest Inventory and Analysis (FIA) data to quantify habitat heterogeneity as a standard for evaluating the ability of texture measures and landscape metrics to capture the heterogeneity.
  • I have obtained Landsat images (88 images acquired between 2004 and 2007) for Venezuela.
What I'm working on now
  • I am processing the global land cover products and generating consensus land cover datasets for the globe.
What obstacles are blocking progress
  • None
What's next
  • I will evaluate the accuracy of the consensus land cover datasets for capturing sub-pixel land cover heterogeneity.
  • I will continue to look for alternative data on vegetation characteristics that can be used for the analysis on the relationships between texture measures and vegetation heterogeneity.
  • I will start to calculate texture measures for the second test site (i.e., Venezuela).

2012-08-27

What I did this past week
  • I generated spatial autoregressive models to evaluate the ability of texture measures to capture landscape patterns and to capture vegetation heterogeneity.
  • I explored approaches for up-scaling the values of texture measures to coarser spatial scales and evaluated the ability of the up-scaled texture measures to capture vegetation heterogeneity.
  • I reviewed Benoit's first draft of the interpolation review and provided some suggestions and comments.
  • I obtained required global land cover products and started to work on Python code for generating consensus land cover datasets with a modified processing approach.
What I'm working on now
  • I am working on Python code for generating consensus land cover datasets, probably via GRASS GIS, from existing global land cover products.
What obstacles are blocking progress
  • The current datasets that I used for quantifying vegetation heterogeneity are only available for forested areas in the contiguous US. Other datasets on vegetation structure or other characteristics for non-forest vegetation types and/or in other geographic areas are needed.
  • I need to learn about GRASS GIS and the approach to call GRASS functions from Python.
  • Some heterogeneity metrics are sensitive to seasonal and/or inter-annual variability of vegetation among Landsat scenes. This can cause inconsistency in metric values among Landsat scenes.
What's next
  • I will finish the coding for generating consensus land cover datasets.
  • I will look for alternative data on vegetation characteristics that can be used for the analysis on the relationships between texture measures and vegetation heterogeneity.
  • I will start to calculate texture measures for the second test site (i.e., Venezuela).
  • I will continue to work on the issue about vegetation phenology effects.

2012-08-07

What I did this past week
  • I evaluated the ability of texture measures to capture land cover heterogeneity measured by landscape metrics and to capture spatial heterogeneity of vegetation structure by using multiple linear regression.
  • I explored different approaches for upscaling the information captured by texture measures.
  • I finished a draft mentioning an updated approach for generating consensus land cover datasets from existing global land cover products.
What I'm working on now
  • I am evaluating the sensitivity of different texture measures to seasonal and/or inter-annual variability of vegetation (vegetation phenology), and exploring approaches for dealing with the problems of value discontinuities along image edges and inconsistency among images.
What obstacles are blocking progress
  • Some heterogeneity metrics are sensitive to seasonal and/or inter-annual variability of vegetation among Landsat scenes, and the seasonality effects are more influential for some land cover types (e.g., cropland and built-up areas). This can cause inconsistency in metric values among Landsat scenes.
What's next
  • I will continue to work on the issue about vegetation phenology effects.
  • I will further explore the approaches for upscaling texture measures and evaluate the ability of upscaled metrics to capture spatial heterogeneity of land cover/vegetation structure.

2012-07-24

What I did this past week
  • I calculated heterogeneity metrics based on NDVI derived from GLS 2005 data, and resolved the issue on mosaicking metric images based on pixel-wide quality (i.e., the number of valid data within each 500m pixel).
  • I evaluated the seasonality effects on consistency in heterogeneity metric values, compared the sensitivity of different metrics to the effects, and analyzed the effects of land cover types on the sensitivity.
  • I reviewed literature about global land cover products, comparisons among different products and synergies of those products.
What I'm working on now
  • I am working on a draft of modified methods for generating a consensus land cover dataset.
What obstacles are blocking progress
  • Some heterogeneity metrics are sensitive to seasonal and/or inter-annual variability of vegetation among Landsat scenes, and the seasonality effects are more influential for some land cover types (e.g., cropland and built-up areas). This can cause inconsistency in metric values among Landsat scenes.
  • Some global land cover products lack information on class-specific accuracy and/or information on pixel-wise data quality, which I would like to incorporate into the procedure for generating a consensus land cover dataset.
What's next
  • I will finish the draft of modified methods for generating consensus land cover data.
  • I will work on the issue about seasonality effects.

2012-07-10

What I did this past week
  • I compared the usefulness of different heterogeneity metrics for estimating NBCD canopy height and biomass data using linear regression models.
  • I reviewed the methods used for generating the consensus land cover dataset from multiple land cover products.
  • I started to process GLS 2005 data and calculate heterogeneity metrics based on the data provided from the NASA group.
What I'm working on now
  • I am working on a draft of modified methods for generating a consensus land cover dataset.
What obstacles are blocking progress
  • NBCD canopy height and biomass datasets have values of 0 for mixed forest land cover (according to the NLCD), and thus cause false spatial heterogeneity. These datasets may not be suitable for evaluating the usefulness of the heterogeneity metrics.
  • Because the GLS data from the NASA group are also scene-based, I need to develop an approach to mosaic Landsat scenes or derived metric layers.
What's next
  • I will finish the draft of modified methods for generating consensus land cover data.
  • I will develop an approach for mosaicking scene-based Landsat data and/or derived metrics.
  • I will look for other datasets of vegetation structure.

2012-06-26

What I did this past week
  • I calculated first- and second-order texture measures based on NDMI for entire Oregon.
  • I re-calculated texture measures based on NDVI with a higher radiometric resolution.
  • I obtained the datasets of canopy height and aboveground biomass from NBCD and calculated first-order texture measures of the datasets for entire Oregon.
  • I evaluated the correlations among different heterogeneity metrics and evaluated their correlations with spatial heterogeneity of canopy height and biomass.
What I'm working on now
  • I am evaluating the correlations among the metrics and their ability to capture spatial heterogeneity of canopy height and biomass in different land cover types and ecoregions.
What obstacles are blocking progress
  • None
What's next
  • I will continue to evaluate the usefulness of those heterogeneity metrics for capturing spatial heterogeneity of land cover.

2012-06-19

What I did this past week
  • I compared the values of three types of metrics (first-order texture measures, second-order texture measures and landscape metrics) within EPA Level III Ecoregions.
  • I revised the Python code to directly read data from EOS-HDF files using the "pyhdf" package.
  • I calculated the Normalized Difference Moisture Index (NDMI) based on near-infrared and middle-infrared bands of Landsat images for entire Oregon.
What I'm working on now
  • I am calculating texture measures based on NDMI with a higher radiometric resolution (6 bits now vs. 5 bits before).
What obstacles are blocking progress
  • There is a problem to read image data (at a 30 m resolution) for the entire Oregon at once using gdal (a memory issue). Tiling the image may be necessary.
  • Current Python code for reading image data cannot maintain projection information from EOS-HDF files.
What's next
  • I will complete generating texture measures based on NDMI.
  • I plan to update the texture measures based on NDVI using a higher radiometric resolution.
  • I plan to compare texture measures calculated from different remotely sensed data (i.e., NDVI vs. NDMI).

2012-06-12

What I did this past week
  • I generated data layers of four first-order texture measures, eight second-order texture measures, 12 landscape-level landscape metrics and one class-level landscape metric (one layer for each of 10 different land cover types) with the fixed-grid approach for entire Oregon.
  • I compared the values of these metrics calculated in the area of one Landsat scene. Most of second-order texture measures are highly correlated with one another. Similarly, many landscape metrics are also highly correlated. However, there are only low to medium correlations between texture measures and landscape metrics. This suggests that tecture measures derived from NDVI and landscape metrics derived from categorical land cover data may provide different information on land cover heterogeneity.
  • I compared metric values calculated from difference Landsat scenes in their overlapping regions to evaluate the effects of vegetation seasonality on the metric values. Preliminary results show that different metrics have different sensitivity to vegetation seasonality, and the sensitivity is also related to land cover types.
  • I summarized the progress of this project for a NASA report.
What I'm working on now
  • I am comparing values of different heterogeneity metrics within and among different land cover mosaics.
  • I am looking for available datasets on vegetation structure and/or land cover patterns to facilitate an evaluation of the ability of different metrics to capture those information.
What obstacles are blocking progress
  • There is no major obstacle.
What's next
  • I will continue to familiarize myself with GRASS.
  • I will further examine the information on land cover heterogeneity captured by different metrics.

2012-06-05

What I did this past week
  • I tested the performance of different approaches for running Fragstats from Python and revised the Python script to improve the performance. Now it only took ~1 hr to calculate 14 landscape-level metrics and 1 class-level metric for 1 Landsat scene.
  • I looked at the SDMTools package in R and r.le and r.li functions in GRASS. While SDMTools can only calculate patch-level metrics, the functions in GRASS may be considered as an alternative to Fragstats.
  • I started to familiarize myself with GRASS.
What I'm working on now
  • I am calculating Fragstats metrics for the entire Oregon based on NLCD land cover data.
  • I am conducting preliminary comparison between landscape metrics and texture measures obtained for an area covered by a Landsat scene.
What obstacles are blocking progress
  • There is no major obstacle.
What's next
  • I will complete the calculation of landscape metrics for the entire Oregon.
  • I will further analyze and compare the landscape metrics with texture measures derived from GLS Landsat imagery.

2012-05-29

What I did this past week
  • I contacted the authors of Fragstats and resolved the problem of running Fragstats via the command line.
  • I wrote Python code to calculate landscape metrics via Fragstats using the fixed grid approach.
  • I revised the code for calculating the first- and second-order texture measures and implemented a defined class in the code to better conduct image processing and data format conversion.
What I'm working on now
  • I am testing different approaches for calculating landscape metrics via Fragstats and polishing the Python code.
What obstacles are blocking progress
  • The current code for calculating landscape metrics runs slowly. It took 6 hrs to calculate 7 landscape metrics using the fixed grid approach for a 500-by-500 pixel NLCD image ( 1,000 MODIS 500m pixels).
What's next
  • I will explore different approaches for calculating landscape metrics via Fragstats and compare their efficiency.
  • I will compare the landscape metrics with texture measures for several small regions.

2012-05-20

What I did this past week
  • I revised the python code for calculating GLCM texture metrics (now it's 2-3 times faster than before).
  • I evaluated the effects of seasonality and different approaches (moving window va. fixed window) on the first-order and GLCM texture metrics.
  • I calculated several landscape metrics based on NLCD land cover data for small test areas using Fragstats.
What I'm working on now
  • I am working on the code to run Fragstats from a command line and exploring the way to calculate landscape metrics for each MODIS 500m grid (fixed window approach).
What obstacles are blocking progress
  • Fragstats can only run in the Windows environment and cannot handle an entire Landsat scene with the moving window approach (memory issue).
What's next
  • I plan to figure out the best way to calculate landscape metrics with the fixed window approach.
  • I plan to compare some landscape metrics with the first-order and GLCM metrics in some small test areas.
  • I plan to start to generate texture metrics for the entire Oregon after obtaining mosaicked Landsat products from NASA.

2012-05-13

What I did this past week
  • Calculating first-order and GLCM texture metrics using the moving window approach for two Landsat scenes
  • Writing Python code to calculate first-order and GLCM texture metrics using fixed windows (MODIS 500m grid) for three Landsat scenes
  • writing Python code to calculate the number of Landsat pixels with valid values for each MODIS 500m pixel (this information may be used as a quality indicator for generated texture metrics)
  • Comparing the values of texture metrics calculated using the moving window approach vs. those calculated using fixed windows
  • Comparing the values of texture metrics within the overlapping regions among Landsat scenes (which were acquired in different seasons of different years) to evaluate the effects of seasonality on the metric values
What I'm working on now
  • Calculating texture metrics for more Landsat scenes to evaluate the effects of seasonality on metric values in the regions of different land cover types
  • Reading the user manual of Fragstats and testing Fragstats for generating landscape heterogeneity metrics from a categorical land cover map (NLCD)
What obstacles are blocking progress
  • Fragstats can only run under the Windows operating systems. It may need to write my own code to calculate landscape metrics in the Linux environment.
What's next
  • Generate landscape heterogeneity metrics using Fragstats and compare them with texture metrics derived from NDVI
  • Continue to evaluate the effects of seasonality and different approaches (moving window vs. fixed window)
  • Generate texture metrics for the entire Oregon after obtaining mosaicked Landsat products from NASA

2012-05-07

What I did this past week
  • Reviewing the literature on remote sensing derived metrics of habitat heterogeneity.
  • Landscape metrics
  • First-order texture measures
  • GLCM (gray level co-occurrence matrix) texture measures
  • Spatial autocorrelation related metrics
  • Calculating NDVI from the GLS (Global Land Survey) surface reflectance products for the entire Oregon
  • Identifying and masking out the pixels with problematic surface reflectance values
  • Using the QA (quality assessment) information to mask out water bodies and cloud-contaminated pixels
  • Testing my Python codes for calculating GLCM texture measures in small areas
What I'm working on now
  • Generating first-order and GLCM texture measures at two spatial scales (3x3 and 11x11 pixels) for a Landsat scene
  • Continuing the literature review on other metrics (fractal analysis)
What obstacles are blocking progress
  • Processing time
  • It took ca. 4 days to calculate the GLCM homogeneity from NDVI with a 3x3 moving window for a single Landsat scene
  • Seasonality problem
  • Landsat scenes acquired in different seasons of different years may cause inconsistency among scenes
What's next
  • Write a Python code for calculating GLCM texture measures using fixed windows (MODIS 500m grid)
  • Compare the results from the moving window and the fixed window approaches
  • Evaluate the effects of the seasonality problem on texture measures