You are here

Summer 2013 Blog - Alan Cai

Alan Cai is spending twelve weeks in the Bay Area working at Energy Biosciences Institute. 

August 9, 2013

My internship at the Energy Biosciences Institute is winding down.  In the weeks since my exciting trip to Illinois for the EBI retreat, I have been continuing to use my ever-improving ArcGIS skills to analyze land use change in Mato Grosso province, Brazil.  Our central goal this summer has been to determine where deforestation has occurred in the past decade, and what the converted land is being used for now.  To do this, we have relied on maps of land cover produced by a collaborator at Brown University, Stephanie Spera.  Blocks of land (referred to as pixels) are classified as forest, pasture/cerrado, soy crops, cotton crops, soy/cotton crop rotation, soy/corn crop rotation, irrigated, or unclassified. The land cover of adjacent years is compared pixel-by-pixel to determine extent of deforestation, conversion to pasture, and crop cultivation.


The extent of water cover in Mato Grosso province, Brazil. The large expanses of water along the eastern and southern borders of the province are wetlands.

The land classification is done using a multi-step decision tree approach that distinguishes among land uses according to the changes in EVI through a growing season, defined here as August-July. The trendline of EVI through a growing season reflects the phenology of the plant cover and can be used to classify the land type. For example, if the standard deviation of EVI measurements for an August-July period is below 0.05 or the minimum is above 0.3, then that pixel is classified as forest.  (Tropical rainforests do not grow and lose their leaves on a seasonal cycle, as is the case with deciduous species, so the standard deviation should be low.  They are also heavily vegetated, with multiple layers of plant matter, so the EVI should be relatively high.)  If it doesn’t meet this criteria, the pixel is then examined to determine if it is cerrado/pasture.

There are at least two shortcomings to relying just on this method.  First, simply looking at adjacent years leads to many cases of a pixel flipping back and forth from deforested to forested on an almost annual basis, which is clearly not realistic.  Perhaps the EVI trendline analysis was not precise enough.  In any case, we decided to alter the definition of land use change to require a longer-term change.  For example, deforestation was redefined to cover pixels which went from forest one year to non-forest the next, and were still non-forest the year after that.  Making this change reduced the estimated deforested area by 30% on average for the time period 2002-2011.


A trendline of EVI (the red and blue lines) between 2000 and 2012 for a random point on Brazil’s land surface. Notice the annual rise and fall in EVI.

Another needed change was to apply a “water-mask” to our maps.  There are some regions of Mato Grosso province known to be covered by wetlands, yet they were being classified as forest, crops, and other things by the algorithm.  By eliminating the pixels that overlapped with a map of water cover, we further reduced the estimated deforested area by 32% on average.  Applying both of these changes led to an average 53% reduction.  There might be over corrections to apply.

There are still several questions left to be answered, such as lining up our estimated deforestation, pasture creation, and agricultural cultivation with the farms, ranches, and other large properties in Mato Grosso to see how the owners are utilizing their land differently.  In my next and last blog post, I will summarize my internship experience and what I took away from this summer.

July 23, 2013

Last Wednesday night I returned to Berkeley from a 3-day retreat in Champaign, Illinois with Cyrus and the rest of my EBI colleagues.  The trip was exciting, educational, and very enjoyable.  I will spend this post describing what happened and what I learned.  This is my one chance to write a blog entry in “traveler-mode,” as most of my fellow Energy Corps members have been doing all summer, so please bear with me!


Alan on the bus ride to the University of Illinois at Urbana-Champaign.

Most of Sunday was spent in transit from Berkeley to UIUC (University of Illinois at Urbana-Champaign).  Stepping outside of Midway Airport in Chicago, I immediately noticed that it was hot (about 85 degrees) and humid.  I was sweating almost immediately.  Having stayed mostly in the Bay Area and San Diego these last few years, I have probably been spoiled by mild weather at this point.

A highlight of Sunday night was a “game” the retreat organizers had us play in order to interact with people outside our research area.  EBI’s research on bioenergy spans five categories: feedstock development, biomass depolymerization, biofuels production, fossil fuel bioprocessing, and lastly environmental, social and economic (ESE) impacts.  Our name tags came equipped with some 24 stickers (mine were ESE, since my research on land use change in Brazil falls into environmental impacts).  Our goal was to exchange stickers and get a group of 6 (other than the 5 above categories, EBI administration also had their own stickers).  These could be traded for an alcoholic beverage coupon.  It was entertaining to see everyone, from the lowliest undergraduates like me, all the way up to the directors of EBI, participate in this meet-and-greet.  I met some interesting people who I ended up hanging out with for the rest of the retreat.

Monday was mostly a whirlwind of lectures and panel discussions.  I don’t pretend to have more than a background knowledge of most of what was discussed, but I tried to follow what was being said.  During the lunch break, Cyrus and I ventured out into the blazing heat (85 F and 70% humidity) to look around at the UIUC campus.  There were some elegant brick buildings that I would associate with pioneer-era America, but for the most part it was depopulated and not too lively.  And the campus is very flat, something I just could not get used to.

In the evening, I had a chance to present the research poster my group prepared.  This was a very exciting opportunity to share the results, so far, of my work (I will detail them in a future blog post).  Although I was perhaps a little nervous, I was reminded myself that even if I did not have a complete grasp of land use change in Brazil after working for a month and a half, my audience would know even less.  That was comforting.  I was happy to see so many people willing to have me explain the poster, having spent so many hours creating the graphics, writing the project methodology, and proofreading the text.


A panorama of the University of Illinois at Urbana-Champaign campus.

Between stints of standing next to my poster, I also looked at other people’s work.  Interestingly enough, a number of people at EBI are focused on biofuel developments in Brazil.  Brazil derives much of its fuel from sugarcane that has been transformed into ethanol, and there seems to be increasing interest in Brazil within EBI.  I saw a poster that explored potential financial reasons for fluctuations in sugarcane production, and one which examined policy and governmental aspects of Brazil’s biofuels use.

That night after dinner, Phil New, who is the Biofuels Chief Executive Officer for BP, gave what I thought was the most compelling speech of the retreat. He explained the breadth of BP’s investment in biofuels, including expanding operations in Brazil, and more generally expressed confidence in the viability of BP’s investment in biofuel operations and EBI.

We ended the day by making a quick trip to a bar in downtown Champaign.  I noticed the bars were very full, for 10 pm on a Monday – perhaps this was the only time of day where it was reasonably comfortable to be outside?


Alan and the rest of the research team standing in front of their poster, entitled “Pasture to Crop Transitions in Cerrado Brazil.” From left to right: Cyrus Blankinship, Jeanny Wang, Project Supervisor Avery Cohn, and Alan.

The highlight of Tuesday was definitely visiting the UIUC Energy Farm, a vast expanse of cropland located about a 15 minute drive from the center of the campus.  Again, the weather was difficult to endure, but the speakers were engaging and there was a well-timed gelato break as well.  Presenters showed off various crops they were growing, such as Miscanthus giganteus, switchgrass, and sugarcane, as well as the metrics used to assess which was best suited for use as a biofuel feedstock (for example, biomass per acre, resistance to drought, fertilizer needs, and growth rate).

On Wednesday, after leaving Champaign, I had an opportunity to visit Argonne National Laboratory outside of Chicago.  Again, the research topics were mostly outside of what I have been doing this summer, but I appreciate the willingness of the employees there to explain their work.  I was especially interested to hear about a project exploring the use of land with marginal suitability for agriculture, for bioenergy crops instead.

Overall, I thought the retreat was well-organized, engaging, and a refreshing change of pace.  I’m happy to have gotten to know so many of my colleagues, though I now realize that with my internship ending in three weeks, there won’t be much more time to connect with them.  Also, having already helped create and present a research poster, I now have a leg up on the Cal Energy Corps symposium happening in October.  I’m very glad I went!

July 2, 2013


The three administratively-defined biomes of Brazil’s Mato Grosso province: Amazonia (green), or rainforest; cerrado (tan), or savanna; and pantanal (blue), or wetland. However, the actual distribution of biomes is not as neat and clear-cut as these boundaries would suggest.

The past few weeks have mostly consisted of me reacquainting myself with the map-making software ArcGIS and using it to do analyze changing land cover in the Brazilian province of Mato Grosso.  I am trying to figure out the deforested area on a year-to-year basis using three datasets.  They are based on data from two sources: a satellite belonging to a NASA program called LANDSAT, as well as MODIS, a spectroradiometer instrument on board another NASA-operated satellite.

I am still learning about how exactly these satellites work, but from what I can understand, they shoot light of different wavelengths towards a place on the surface of the earth.  The light can be absorbed or reflected, depending on what land cover is present  For example, a green forest canopy would absorb non-green colors of the spectrum and reflect green light back towards the satellite.  (This is why we see green objects as green – it is the color that the object sends out and which our eyes take in.)  By contrast, an agricultural field would probably have a different profile of absorbency – and it would change greatly over the course of a year, as crops are planted, grow, and are harvested.  Tropical rainforests, by contrast, don’t change so much in response to the seasons.

These satellites can scan the entire globe once every few days, so it is possible to observe changes in land cover in a particular location over longer periods of time, say over a growing season.  The more frequently a place is scanned, the higher “temporal resolution” any resulting dataset has.


An image of Mato Grosso from space.

The other measure of detail is spatial resolution.  The satellites shoot a finite number of beams towards Earth, so even though each measurement is only truly accurate for one point on the ground, analysts obviously have to apply it to a larger area.  The more data points there are, the finer-resolution the resulting image is – sort of like how many megapixels your smartphone’s camera is capable of.

The 3 datasets have differing spatial resolutions, but the one I’ve spent the most time analyzing has a pixel size of about 238 by 238 meters – or 56,571 m^2 .  For comparison, an American football field is 5,351 m^2 (yes, I am dead set on using metric units – the heresy!). You can see a preview in the images embedded in this post.

Outside of work, there’s not much that’s noteworthy.  I am definitely looking forward to EBI’s retreat in Illinois in about 2 weeks, and there is a chance I might visit my hometown of San Diego next week for ESRI’s annual conference.  ESRI is the company that makes the ArcGIS mapping software I’m using.


The black pixels together represent the extent of deforestation between mid-2002 and mid-2011 according to the PRODES dataset, one of the three Alan is analyzing.

I have to mention that I’ve decided to try to learn some Brazilian Portuguese this summer.  I haven’t found any fluent speakers, so I am relying on some mp3 lessons instead.  So far I can say the following: Oi! Bom dia.  Eu sou Americano de San Diego.  Eu falo um pouco de Portugues. Eu entendo Ingles e Chines.

Look for my next blog post, by which time I will have hopefully learned some of the rules of Portuguese spelling, and also figured out how to type letters with accent marks on top of them.  Ate logo!

May 30, 2013


Alan in front of Energy Biosciences Institute Building.

This summer I am spending 12 weeks working at the Energy Biosciences Institute (EBI) in Berkeley, CA on a project related to sustainable land use in Brazil.  EBI is an organization focused on bioenergy research, and  is co-led by UC Berkeley, LBNL, the University of Illinois at Urbana-Champaign, and the energy company BP.  I am really excited to be working with my fellow Cal Energy Corps participant Cyrus Blankinship on this research, as well as Avery Cohn, a recent Berkeley PhD graduate who is heading up the study.

The basic gist of my project is that I will use a number of software programs, including ArcGIS (Geographic Information Systems), R, and Python, to develop a land cover classification algorithm for Brazil’s cerrado (savannah) biome, and track shifts in land use over the last dozen or so years.  Specifically, I want to see changes in forest, cerrado, pasture, and crop cover.  The maps I produce will be used for socio-economic and behavioral research in the later phases of this study.

The project has implications for Brazil’s actions to mitigate climate change, because as demand for livestock and agricultural products grow, so does pressure to cut down forest and cerrado, which releases greenhouse gases.  There is evidence suggesting that conversion of lower-productivity pastureland to higher-productivity cropland would be helpful in Brazil’s efforts to curb its greenhouse gas emissions. If new cropland can be created out of converted pastureland rather than by clearing pristine habitat, then the threat to native land (and the associated biodiversity, ecosystem services, and climate-related degradation) will be somewhat reduced.

Though Cyrus and I aren’t in an exotic foreign country like most of the Cal Energy Corps researchers, we do get to work in the brand-new, eco-friendly Energy Biosciences Building (EBB) in the northwestern corner of the Berkeley campus.  Also, we plan on attending and presenting our work at EBI’s biannual retreat in Illinois in July, so we will get to do some traveling after all.

These first two weeks of my internship have mainly been about project orientation, reading papers for background knowledge, and setting up our computers and software for the work ahead.  New team members are also still being added.  Soon, however, we will start doing some exciting geospatial analysis!

UC Berkeley logo

© Copyright UC Regents. All Rights Reserved.
Privacy Statement