给自己泼盆冷水

Don't Become a Scientist!

American universities train roughly twice as many Ph.D.s as there are jobs for them.

扩招的压力也许并不仅存在于本科和硕士生中间。虽然现在博士在中国还不是太过剩，但谁也无法断言4年后社会拥有多少 PhD 又需要多少 PhD。

虽然作者讨论的是美国，讨论的是基础研究，但中国的应用科学和工科博士现在也不少了，反倒是基础研究的博士比较少，或者说真正有实力读基础学科的 PhD 的人太少。不知道3、4年后会不会出现博士为了一个普通技术员的岗位争得头破血流的场面。

不过真正把环境科学读到博士的人整个大学本科阶段我也只见过两三位，应该还不会太过剩。

然而读了博士以后，即使不会因为生计所迫当一个又一个的博士后，大概也要为研究经费操心。可能很多时间都不能用来研究而是写一个又一个 proposal 。用现实的眼光看，这些都是投身研究工作必须面对的问题，现在也应该有心理准备。

最后是关于博士和科学家生活的光明面：

Science is fun and exciting. The thrill of discovery is unique.
 

Considerable Progress!

On the base of yesterday's trial, today I reduced the model area to Bleaklow only.  Surrounding areas are defined as inactive cells and holes in DTM are filled with elevation data from OS map.  The simulation now still cannot converge but the water budget gives good consistency (inconsistency about 0.01%).  The map of water head draw down and dry cells also make sense. 

The question now I have is why the water table draw down so fast in just one day.  I would also like to know the storage coefficient of water in the soil -- absence of this parameter may lead to what I see in the simulation.  Maybe I should ask the authors of MODFLOW whether such rapid draw down is normal and if I ignored something in the simulation.

Weird “unable to converge” problem

Today I tried to solve the converge problem which prevent my model from simulating groundwater flow in more than 1 time step or stress period. I deactivated all modules one by one, then tried to adjust time parameters. None of these work.

Then I adjusted layer type. When I set layer 2--11 as type 0: confined, and layer 1 as type 0: confined or type 2: confined/unconfined with constant transmissivity, the simulation can work until stress period 51, time step 9. Otherwise the model cannot converge and simulation stops.

It was found later that both type 0 and type 2 can be assigned to layer 1 and 2, but type 3 and 1 will make converge problems. If type 2 is assigned to other layers, the problem also occurs.

Another model is tested. It has 30x30x3 cells and similar configuration with my working model. Strange enough, this model works well. I doubt the transmissivity settings may lead to the problem of my working model.

Another file gives possible causes of the problem. Some are very possible to present in this simulation. I will need to see to them. If there are questions unsolvable without acceptable changes of the model, I may consider using conceptual models instead of spatial explicit model.

The result of many times of simulation, variable control and trial error: the dry cells generated during simulation is the key reason of convergency difficulties. MODFLOW is not suitable to deal with non-saturate flow. My counter measure thus is to use more carefully designed parameters in the model, do a single-period, single-time-step simulation and identify those cells destined to become dry and exclude them (be assigning them as inactive cells) in consequential multi-period simulations. Besides, there are some ``holes'' in the DEM, if they are in my concerned area, I should define them as inactive cells and pay attention to their adjacent cells to ensure they have correct elevation data.

A somewhat better simulation result

After I redefined the boundary conditions, got rid of RIVER package, adjusted STR1 package and gave river cells constant head, I got such simulation result:

     CUMULATIVE VOLUMES      L**3       RATES FOR THIS TIME STEP      L**3/T
     ------------------                 ------------------------           

     IN:                                      IN:
           ---                                      ---
       CONSTANT HEAD =    63661356.0000         CONSTANT HEAD =     2092976.1300
              DRAINS =           0.0000                DRAINS =           0.0000
       RIVER LEAKAGE =           0.0000         RIVER LEAKAGE =           0.0000
                  ET =           0.0000                    ET =           0.0000
            RECHARGE =      871753.8130
              RECHARGE =       28660.4004
            TOTAL IN =    64533108.0000              TOTAL IN =     2121636.5000
          OUT:                                     OUT:
          ----                                     ----
       CONSTANT HEAD =    55723300.0000         CONSTANT HEAD =     1831999.0000
              DRAINS =     8146114.5000                DRAINS =      267817.4690
       RIVER LEAKAGE =           0.0000         RIVER LEAKAGE =           0.0000
                  ET =      747962.5630                    ET =       24590.5508
            RECHARGE =           0.0000              RECHARGE =           0.0000           
           TOTAL OUT =    64617380.0000             TOTAL OUT =     2124407.0000
            IN - OUT =      -84272.0000              IN - OUT =       -2770.5000  
 PERCENT DISCREPANCY =          -0.13     PERCENT DISCREPANCY =          -0.13

The discrepancy of the result is much smaller than previous one (82%).  Now I have a model that makes sense.

The next problem would be adjusting parameters using inverse models and make the result converge so multi period simulation is possible.

MODFLOW: Progress so far and problems

Today I finished inputing parameters to PMWIN and had a first run of the model.

It turns out that some of those characters may not be defined correctly. The water budget for the area do not balance thus the simulation cannot continue after the first time step. My first guess is that the boundary condition of the model is not correct: the rivers and streams should be constant head and the boundary of simulated area may be too low and should not be defined as active cells.

Individual modules may also have problems in their parameters. Tomorrow I will look at the output file and determine the need to change.

Protected: 牢骚

机房里总是太吵。

GIS 专业的人为什么总是有那么多的 GROUP WORK 要带到机房来做呢？今天早上做了什么完全没有印象，头嗡嗡直响。耳边全是他们的说话声和笑声。

现在完全不想见人，只想一个人呆一会儿。

明天看看本科生的机房里是不是安静一些，不然这活没法干了。 

Fieldwork Phase 1

I just come back from it.

I have encountered more sphagnum patches than I expected, though some are rather stressed, especially those on the south-facing slopes. Perhaps it is because of more sunlight and less moisture. Besides, there are very few sphagnum patches larger than 5 metres in any dimension, except some in SK0993 in those gullies.

Besides, I just found that many sphagnum patches can be found in these ESA vegetation types: cottongrass moorland, dry bog heather or non-heather dominated. I'll do a chi-square test to see if the hypothesis is true: that land use type is related to sphagnum presence.

A possible spectral signature can be found using about 20 sphagnum patches I found this time. The problem is the difference between aerial photographs made large-scale classification complicated: you have to identify sphagnum patches one square km by another. This could mean extensive fieldwork next phase.

The validation of these signatures is another problem.  I can guess they are pretty good predictions, but this must be verified using more fieldwork data.

I may need to do more study on sphagnum habitats considering elevation and hydrology.  I also need to talk with Jack in the coming weeks.  For the next period, the main focus will be on hydrology models.