Diario de ladybird_sunbathing

1. Resights of ringed/flagged birds 2017 spring (个人旗标记录汇总)

● Color Rings (彩环)
https://www.inaturalist.org/observations/ladybird_sunbathing?utf8=%E2%9C%93&q=color+rings&search_on=&quality_grade=any&reviewed=&geoprivacy=&identifications=any&captive=&place_id=&swlat=&swlng=&nelat=&nelng=&taxon_name=Shorebirds+and+Allies&taxon_id=67561&day=&month=&year=2017&order_by=observations.id&order=desc&rank=&hrank=&lrank=&taxon_ids%5B%5D=&d1=&d2=&created_on=&site=&tdate=&list_id=&filters_open=true&view=map

● Engraved flag (标码旗标)
https://www.inaturalist.org/observations/ladybird_sunbathing?utf8=%E2%9C%93&q=engraved+flag&search_on=&quality_grade=any&reviewed=&geoprivacy=&identifications=any&captive=&place_id=&swlat=&swlng=&nelat=&nelng=&taxon_name=Shorebirds+and+Allies&taxon_id=67561&day=&month=&year=2017&order_by=observations.id&order=desc&rank=&hrank=&lrank=&taxon_ids%5B%5D=&d1=&d2=&created_on=&site=&tdate=&list_id=&filters_open=true&view=map

● Plain flag (无码旗标)
https://www.inaturalist.org/observations/ladybird_sunbathing?utf8=%E2%9C%93&q=plain+flag&search_on=&quality_grade=any&reviewed=&geoprivacy=&identifications=any&captive=&place_id=&swlat=&swlng=&nelat=&nelng=&taxon_name=Shorebirds+and+Allies&taxon_id=67561&day=&month=&year=2017&order_by=observations.id&order=desc&rank=&hrank=&lrank=&taxon_ids%5B%5D=&d1=&d2=&created_on=&site=&tdate=&list_id=&filters_open=true&view=map

To find out what color-combined flags represent what location information, browse the protocol:
http://eaaflyway.net/documents/Protocol_birds%20marking.pdf

1. 鸭绿江口湿地国家级自然保护区现状

保护区只有一条底线原则：不得近一步地围垦。在此底线之内，人们可以进行一切不违法（包括不违反《野生动物保护法》）的活动。

自上世纪80年代的围海造田起，北井子镇的海岸线已被推进了五公里。现在，保护区由两大主要部分组成：农业用地和渔业用地，前者为距海堤200米以上的区域；后者包括海堤内侧的海产品养殖塘，以及海堤外侧的滩涂、海域。渔民在潮间带放养四角蛤蜊、托氏昌螺等，在距海堤两公里处的固定渔网上收取虾爬（Stomatopoda）和鱼等，在潮下带渔船捕捞作业。

协调渔民生计与跨境科研保护、规章监督、设施维护、宣传，这些我们期望保护区运营的工作，在鸭绿江口几乎全然没有。相反，保护区的地权不在丹东环保局手中，而归属于各村承包滩涂养殖的私人老板。我们要下滩拍鸟、挖泥，得先跟保护区打声招呼，再分别与各片滩涂的私人老板协商，每人送条烟，以解决这一季度的通行证。毕竟是人家用来生钱的地盘，说不让下滩，你一点办法没有。这不，五月底泥螺（野生）长出来了，又不给下滩了，怕把泥螺踩完了。我们又得找村长好说歹说，最后他准我们下滩，我们给他提供精美照片以做乡镇形象建设宣传。

衡量一个领域的发达程度，我通常看它在大众之间的普及程度。鸭绿江口湿地的生态保护当然远未飞入寻常百姓家，绝大多数人都不知道这里有什么鸟，它们的习性，从哪里来，到哪里去；渔民把垃圾随手丢在滩上；养殖塘里更换海产品时，用毒药杀死原有的海产品残余，再放新的进去养。一天下午，一位住在养殖塘坝埂上的年轻渔民用我的单筒望远镜观察了滩上和人工潮沟里的水鸟，才知道滩涂上有这么大群的鸟，而且不止一种。难怪，广袤的滩涂，鸟的羽色混迹其间，普通人没有望远镜，又不关心，哪能了解这么多呢？

每个月两次大潮，会有大批游客和观鸟爱好者前往近丹东港的2号点观鸟。那时潮水会赶着上万只鸻鹬类（主要是斑尾塍鹬（Limosa lapponica）和大滨鹬（Calidris tenuirostris））到地势相对较高的东西两侧。梅爷爷多次建议保护区，在那个时候做些科普、宣传，未果。当然不会有结果，因为保护区不希望外界知道这里鸟的数量下降了，最好什么也不要知道。某一次大潮数鸟，我们偶遇保护区的副局长老王，跟他聊了很久。他给我们讲述年初他在西澳大利亚与Chris Hassell等人的水鸟环志之旅（新西兰一NGO出资培训保护区职员），讲网捕大杓鹬的不幸溺水，给我们看手机里绚丽的西澳风光。我知道，老王不是坏人，我们都一样，喜欢天高任鸟飞的唯美。想必是评价体系的问题吧，虽然鸟变少并非全部归咎于保护区管理不当（我们也不明原因，揣测丹东港西航道的封堵可能改变了原有沙粒组成），但在那个职位上，人总要为上级负责。我敢说，环保局、林业局、村知书、农民、渔民、利益集团，没有一个是坏人，虽然我还不清楚这一整套错综复杂体制的运作纠葛。

尽其所能让公众知晓情况，理当是生态保护学家的义务，然而这在鸭绿江也受到了挑战。我被告知，当被问到那些鸟吃什么时，切忌说到蛤，因为滩上的白蚬子（Mactra veneriformis）多是渔民放养的，如果渔民知道鸟会吃他们的蚬子，则他们可能会采取极端的手段来维护自己的经济利益，比如下毒。有谁知道滩上的鸟吃蛤？我们科研团队，司机隋师傅，也许帮我们挖泥的刑师傅也知道。我们的终极理想是什么？每个人都知道或可以查阅到生态系统中的每一环。然而现在呢？一道难以跨越的障碍横亘其间……戏剧性的是，僵局在四月中下旬被打破了——4号点滩涂的老板下滩捡起一坨鸟粪，发现是碎砚壳，恍然明白鸟吃了他养的四角蛤蜊（Mactra veneriformis）。老板刚投了800万元的蚬苗（指甲盖大小），全给野鸟当饲料了咋整？于是他开始雇人每天下滩放炮赶走鸟儿（买炮、雇人的开销小于白蚬子被鸟吃掉的损失）。

我钟爱这样的系统：在设计之初它就考虑并解决了所有问题；如果运营的时候出现问题，则再一劳永逸地解决并不遗余力地维护。我问中国多久能建成像北美、澳洲那样成熟的国家公园体系，Jason说50年，而我和梅爷爷都没个数。

1. Current situations of Yalujiang Estuary Wetland National Nature Reserve

Only one bottom line of the reserve: no further reclamation is permitted. Within this bottom line, people can do anything not against the law (including not against "Wildlife Protection Act").

Since 1980s' reclamation (sea burying -> field), the coast of Town Beijingzi has been pushed toward the sea by 5 km. Now the reserve consists of 2 major parts: agricultural & fishery land. The former is the area which locates over 200 meters off the sea wall. The latter comprises the marine product cultivating ponds inside the sea wall, and the mudflat and waters outside. Fishermen cultivate bivalves (Mactra veneriformis), Thomas snails (Umbonium thomasi), etc. on the intertidal zone, harvest mantis shrimps and fish on the fixed net 2 km off the sea wall, fish on the subtidal zone by fishing vessel.

Coordinating fishermen's livelihood and international scientific protection, regulation enforcement and supervision, facility maintenance, outreach, etc., all that we expect a nature reserve to operate is next to none in here Yalujiang estuary. On the contrary, the land ownership doesn't belong to Dandong Environmental Protection Bureau, but to the contractors who are in charge of each village's sea product cultivation respectively. In order to video birds and dig mud, we scientists ought to inform both the reserve and the contractors. The critical process in negotiating with the contractors of 5 villages respectively is to send each one a carton of cigarettes, thus guaranteeing this season's passport. After all, it's their land for money raising. We would have no way if they simply say no. Look, when mud snails (wild) prospered in late May, they no longer let us go onto the mudflat in case we stepped onto the mud snails so caused damage. Then we had to talk with the village head until they rendered us the permission. In return, we gave them delicate photos for the town's image construction.

To judge the development level of an area, I like to see its popularity among the public. The eco-protection of Yalujiang estuary wetland hasn't reached folk. Most people have no idea of the birds, their habit, where they're from and to. Fishermen readily throw garbage on mudflat. Each time altering sea products, farmers remove the remaining creatures in the pond by chemical poison. An afternoon, a young fisherwoman who lived on the embankment didn't realize so large-scaled flock and there were more than one species until she used my telescope to watch the shorebirds on the mudflat and in the artificial channel. Birds' plumage mixed in vast mudflat, not surprising that average people know little without lens and awareness.

During the spring tide twice every month, groups of tourists and bird lovers come to Site 2 nearby Dandong Port, when the tide drives dozens of thousands of shorebirds (mainly bar-tailed godwits (Limosa lapponica) and great knots (Calidris tenuirostris)) to the higher east and west. Grandpa Mel has more than once suggested to the reserve that they hand out leaflets about the birds, but achieved no reaction. Of course they won't take action because the reserve don't expect the outside to know about the decline of birds in here. Knowing nothing is the best. Once a time we bumped into Mr. Wang the vice-president of the reserve when counting birds, and had a long chat with him. He told us about his bird banding trip with a group led by Chris Hassell in Western Australia early this year (a New Zealand NGO contributes to training the reserve's staff), storying the far eastern curlew (Numenius madagascariensis) drowned, showing us the pictures in his cell phone of splendid WA scenery... I know Mr. Wang isn't a bad man. We are same, both yearning for the aesthetic bird flying sky. Supposed to be the problem of the evaluation system, isn't it? The decrease of birds shouldn't be totally owing to the reserve's improper management (we don’t why either, perhaps the block of Dandong Port west channel causes the grain composition to change), but on that position, people have to be responsible for higher-ups, and have to watch their own official career. I bet no one is a bad guy, not the environmental protection bureau, forestry bureau, village secretary, peasants, fishermen, interest groups, although I'm unclear of the complex system's operation.

It should be conservationists' duty to inform the public as much as possible, but this guide line faces a challenge in Yalu River. I was told not to say mactra when asked about the birds' menu. Because the fishermen would possibly take extreme means such as poisoning birds to protect their properties, the mactra, semi-cultivated. Who knows those birds eat mactra? We scientific team, driver Mr. Sui, maybe Mr. Xing too who helps benthos survey. What's our ultimate dream? Everyone knows or can refers to each element in an ecosystem. But what about now? An unconquerable barrier lies between... Dramatically, the deadlock was broken in mid-late April - the contractor of Site 4 went onto his mudflat and picked up a bird poop... "Isn't it my white mactra?" shocked him finally understood that the birds had been eating his sea products. He just had planted CNY 8 million mactra juveniles (about a nail's size), couldn't be devoured by birds! So he employed people to drive away birds by fireworks since then everyday (the fee is lower than the loss without action).

I cherish a system in which it's taken into consideration and solved all potential problems in design. If any bug occurs during operation, then it gonna solve it once and for all and spare no effort to maintain. I ask when China can build such a national park system like North American and Oceania do. Jason says 50 years. Grandpa Mel and I have no idea.

1. 潜在的技术发展对生态研究的推动

我们给觅食的鸻鹬类录像，为了数据化他们在一定时间内啄泥、吃食、排便、呕吐、斗殴的频次，以及食物种类等信息。这些数据需要人工回放视频，敲键盘记录。如果以0.2倍速回放，则分析这上百段五分钟的视频至少需要原始总时长5倍的时间——200小时！如果交由机器完成这些庞大、无聊的工作量，则可大大节省人力。

有一期《自然》杂志的封面文章讲的是通过计算机分析视频中鸟群飞行轨迹，找出领头鸟。看来人工智能在动物行为学上的应用已经有先例了。水鸟的觅食视频分析，我想，也应该由计算机代替人。通过逐帧提取，电脑可以记录鸟喙的每一次啄泥，识别吃到的每一只贝壳或蠕虫。电脑的性能恒稳，不会受心情影响；人眼盯着屏幕，手指敲键盘，难免会出错。另外，我们还对鸟群扫行为谱，得出不同时间鸟群觅食的比例。这也可以用机器分析，什么样姿态的鸟是在觅食或歇息，并数出个数、种类。

人工智能/神经网络是什么呢？通俗地讲，就是喂给机器10双鞋子的照片，告诉它，这些是鞋子；当把第11双鞋拿给它看时，机器会告诉我，这是鞋子。有人会说，稍有常识的人们都能辨识出来的东西，交给机器做甚？那么，如果对象的可辨性达到模棱两可、以假乱真的程度，比如听一段声音让受试者判别其为鸟叫还是人吹的口哨（或者干脆就听一段诗朗诵问你这是人声还是机器模拟声），或者信息繁杂到难以立即算出结果的程度，比如采编的新书应当放在图书馆哪片架位上，那么让机器来做将是明智之举。在刚结束的最后一届ImageNet挑战赛上，算法对物体识别的准确率即以97.3%完克人类的94.9%。相信在积累足够量的数据之后，iNaturalist终有一天可以根据图片自动鉴别物种，省去冗长的人工鉴别。当机器在某方面做得不亚于人类时，机器在该角色上便可取代人类。不过，悲观地讲，这种人工智能技术多久能普遍地应用在生态研究，或者说何时能去分析鸟类觅食视频，现在还很难说，因为缺乏经济的驱动。一款人工智能产品要做得好，研发期得投入大量的资源去训练、调试。仅仅十双鞋子的样本量是远远不够的；谷歌和亚马逊分析的是全球用户的大数据。云端的GPU和数据集决定了样本训练的效率和质量，即便是高等院校的实验室，硬件资源也无法与巨头企业望其项背，因而研发效率也是被后者数量级上的碾压。生态保护毕竟不是自动驾驶、智能家居那样前景光明，有大量市场需求，有大公司扶植的行业，即便是自动驾驶、智能家居也还没有同我们幻想的那样发达、普及，所以当前看来，革新技术以解放劳力的希冀似乎还很渺远。我并不懂神经网络算法，上面所写的多是从身边IT工程师那听来的，难免错误百出，还望不懂行的读者莫要迷信，懂行的随手指证，感谢！

如果抛开阻碍不谈，并且做进一步的设想，那么最好把滩涂所有实地的任务也一并交给机器完成。我们每天顶多四个人下滩拍鸟，扛着三脚架、单筒等设备去找鸟群，接近之后扫描行为谱，再数鸟，再盯个体录像，并记录时间、地点等信息。如果空中的无人机、滩涂上的固定相机和可自动驾驶的间谍相机联合执行任务，那么无人机能最快地考察鸟的分布。有了实时的鸟群信息，机器人便被派遣去拍摄，而鸟群的组成、尺寸、时间、坐标等信息都已被无人机侦测下来了，游猎型摄像机只需专心致志地录像。如果视频实时同步到云端，则鸟群的觅食情况也将立即被分析、量化；如果发现绑旗标的鸟，也当即记录入库。固定着的相机起到守株待兔的作用，也许可以根据底栖动物采样样地有针对性地设定。让机器人拍摄还有一个好处是避免了人为干扰对鸟行为的影响。特别是到后期鸟少食物也少的情况下，接近鸟群非常困难，冷不丁就飞走了。有时鸟群见人来了，便往相反的方向行进觅食。至于底栖动物采样，让机器去挖泥、筛泥、识别蠕虫、活贝壳等，原理与先前的技术一样。如果退一步讲，机器还不足以胜任挖泥的话，那么在辅助方面，人工智能有望替人脑优化合理的行程。结合滩涂地貌、潮汐、风向等环境因素，计算机会规划出采样路线，并综合考虑到各类型的泥有不同难度，是否绕过渔网，安全通过潮沟等等。当实时环境变化时（风向影响水线位置），或者采样者没有按计划或临时改变计划需要新的导航时，人工智能也要随机应变。这需要一款集大成的产品，同步功能非常关键（Google Home?）。我们去挖前一天录像的觅食地，出发前得记下其坐标（通常录像和挖泥不是同一人），如果挖泥的人忘了记，或者当新任务来临却发现自己的设备里还没有从电脑导入49块样地的坐标时，就很尴尬了。

也许有人会问：“如果所有事情都给机器做了，那我们人还做什么？”我们当然永远有事可做！从原始社会工具的发明到今天的信息化系统，技术的革新推动着生产力，而我们人总是永远在工作。单谈生态学，研究与保护的进程永远不嫌快。我们不清楚机器代替人力后自己会做什么，因为认知受当下的生活方式所限。最后用Kevin Kelly在《必然》一书中总结的一种循环出现的模式做结尾（电子工业出版社；周峰、董理、金阳译）：

1. 机器人（电脑）干不了我的工作。
2. 好吧，它会许多事情，但我做的事情它不一定都会。
3. 好吧，我做的事情它都会，但它常常出故障，这时需要我来处理。
4. 好吧，它干常规工作时从不出错，但是我需要训练它学习新任务。
5. 好吧，就让它做我原来的工作吧，那工作本来就不是人该干的。
6. 哇，机器人正在干我以前做的工作，我的新工作不仅好玩多了，工资还高！

7. 真高兴，机器人（电脑）绝对干不了我现在做的事情。
   重复……

8. Potential techniques that push ecological researches

We video foraging Charadriiformes, then quantify how many times they peck the mud, eat, poop, vomit, conflict during a period, plus information on food types. These data need to be recorded by manually watching videos and striking keys. If the hundreds of 5-minute videos are replayed in 0.2 speed, then the time for analysis will require at least 5 times original total length – 200 hours! If the mass, boring working load could be accomplished by a machine, then we’d be far relieved.

A cover article on “Nature” illustrated analyzing a flock’s trace in a video to find who the head bird is. So artificial intelligence (AI) already has been applied on ethology. Analysis on videos of foraging shorebirds, I think, should also be switched from human to machine. Extracting frame by frame, computers can note each time a beak pecks the mud, and identify each mollusk or worm eaten. The computer’s performance is stable, free of mood, while a human types when staring at the screen, mistakes inescapable. We also scan the flock for their behaviors, learning the proportion of foraging ones at different time, which can also be analyzed by machine. A computer can identify in what posture the birds are foraging or resting, and get the number and species.

What is artificial intelligence / neural networks? In common words, we feed a machine 10 photos of shoes and tell it they are shoes. Then we show the machine the 11th shoes, it will tell us that’s shoes. Someone may ask why we let the machine identify what a man with even a little common knowledge can identify. So what if when an object is too ambiguous to distinguish, for example, telling whether a sound track is a bird’s song or a man’s whistle (or listen to a poem recitation and tell whether it’s a man’s sound or a machine’s), or when the information is too multiple to immediately calculate the result, for example, in which part of the library a newly-cataloged book should be assigned, then it’ll be reasonable to have a machine do it. On the just-ended ImageNet contest, the algorithm hit 97.3% accuracy in object discriminating, beating humans’ 94.9%. I believe, after accumulating enough data, one day iNaturalist can automatically identify species according to photos, off the unnecessary manual ID. When machines come better than humans in some aspect, they can replace humans in that role. But pessimistically speaking, how soon this AI tech can be universally applied on ecological researches, or analyzing bird foraging videos, it’s difficult to predict at present. The reason is the lack of economic drive. To make an AI product good, a great deal of resources should be devoted to training and debugging during the development phase. Samples of only 10 pairs of shoes are far from ideal. What Google and Amazon make use of are big data from global users. Efficiency and quality of sample training depend on the GPUs on cloud and the datasets. Even resources of labs of graduate schools can’t rival that of big corporations, so the former’s developing efficiency is supposed beaten by decades by latter’s. Eco-protection isn’t an industry with a promising market and big corporations’ support as autopilot and smart home. Even autopilot and smart home haven’t been as developed and public as imagined, so from now, it seems still a faraway dream to free labors by revolving techs. Honestly I know nothing about neural networks at all. All that I write in here is heard and summarized from the IT engineers around, so there must be a lot of mistakes. Nonprofessional please don’t lay superstition. Professional please point any mistakes you find. Thank you!

If we set the obstacle aside and further envisage, then it’s better to have machines manage all fieldwork. Everyday we had at most 4 people going onto the mudflat to shoot birds. We carried a tripod, a telescope and other equipment to search for flocks. After succeeding in getting close, we would do a behavior scan, count birds and video individuals, and write down time, location, etc. If there were Unmanned Aerial Vehicles (UAV), fixed cameras on mudflat and auto-controlled spy cameras united carrying out the task, then UAV would survey birds’ distribution quick, and robots could be set to video targets with the instant information from UAV. The compose of the flock, scale, time, coordinates, etc. would have been detected and quantified by UAV, so the cruise cameras would just need to focus on video taking. If the videos are uploaded to cloud, then the flock’s foraging situations would be analyzed and quantified at once. If a spy camera finds a banded bird, then it would also note it down. The fixed cameras would wait for their prey, and they could be specifically set according to the results of benthos survey. An advantage of auto videoing would be avoiding anthropogenic disturbance on birds’ behavior. Particularly in the late stage when both birds and food lacked, it’d be very hard to get close to a flock, as they would suddenly fly away with no clue. Sometimes the flock saw guys coming, they would forage towards the opposite direction. As for benthos survey, machines could dig and filter the mud, identify worms and live shellfishes, same technique. If machines hadn’t been advanced enough to do benthos survey, then in assistant ways, AI would be expected to optimize routes. Combining environmental factors as landform, tide, wind, etc. and difficulty levels of different types of mud, whether to bypass a fishnet, to cross a channel safe and sound, the computer could schedule the path for benthos survey. When any change occurs on real time circumstances (e.g. wind influences where the water line is), or the surveyor doesn’t follow the scheme or has a change of the plan then needs a new guide, AI should be flexible and resourceful. This calls for an integrated product with a critical function on synchronizing (Google Home?). We went to survey the mudflat where we took videos yesterday, we would need to note down its coordinates (usually the guy took videoed wasn’t the guy dig mud). It’d be embarrassing if the one doing benthos survey forgot to note the coordinates, or found the 49 sampling points haven’t been imported to her/his GPS device from the laptop when the new task came.

Perhaps somebody may ask, “if all the things are assigned to machines, what are we human beings supposed to do?” We definitely always have things to do! From the invention of tools in primitive society to today’s information system, the revolution of technology pushes forward productivity while we humans have been always working. In ecology, the process of research and protection never feels too fast. We don’t know what we’ll do after machines replace us, because our recognition is limited by the lifestyle nowadays. At last may I cite the repeating mode summarized in book “The Inevitable” (Kevin Kelly 2016):

1. A robot (computer) isn’t able to do my job.
2. Well, it can do a lot, but it can’t do all my job.
3. Well, it can do all what I do, but bugs often occur, so I need to handle at that time.
4. Well, it never makes a mistake in usual work, but I need to train it to learn new tasks.
5. Well, let it do what I did, as that work shouldn’t be done by a human.
6. Wow, robots are doing what I did, and my new job isn’t only more fun, but also more well-paying!

7. Happy, a robot (computer) definitely isn’t able to do what I’m doing now.
   [repeating…]