关键词：论文写作讨论结论章法



对于新手而言，讨论的写作永远都是最令人头大的一件事。学生们的反馈也总是，“到底该怎么写啊，太难了”。与我在引言的写作时所说一致，讨论的写作困难主要还是心理因素。

根本上是新手们未经专门的训练，自行看论文，也难以有效地总结出章法，导致的畏难情绪。

讨论是一个允许作者自由发挥的空间，想把讨论写出来其实并不算难，但是想把讨论写出彩就需要极大的功力了，至少对我来说还难忘项背。

对于一项研究来说，有太多的内容可以讨论了，诸如：本研究的优势与特点、实验数据或样本的代表性、方法的可行性适用性、方法中部分细节设置的合理性、未来可以进一步改进的地方、结果的解释、结果的局限性、结果与现有研究的一致性与差异性、结果说明了什么以及对未来研究的影响、等等。

"结论"与"摘要"具有很大的不同，摘要是论文各个部分的浓缩，而结论重在陈述本文做了什么事情以及发现了什么结果，更像是结果和讨论的浓缩。

讨论与结论hadoop_讨论与结论hadoop

莫奈一家在阿让特伊的花园里The Monet family in their garden at Argenteuil

1. 讨论要有重点，不能放飞自我

尽管可讨论的内容多种多样，甚至数据方法与结果中的任意一点理论上都可以讨论，但是限于篇幅和论文重侧重，我们要有重点的讨论，不要讨论那些无关痛痒的问题。"Discussion"单独成章时，其写作逻辑，我总结了四个讨论一下。

  • (1) 将本研究相对于前人研究的特点与创新性再次讨论一下：读者经过数据、方法和结果部分的阅读，了解完研究的细节后，是时候再次总结强调一下研究的特色与对研究缺口的填补了，目的是再次突出研究的价值。例如：
  • (Liu et al., 2017)：A synthetic learning methodology was developed in this study to estimate GAI from ground-based LiDAR observations over wheat .…… This approach allows circumventing the limits of training the neural network over actual experimental measurements that are prone to uncertainties and are limited by the number and range of cases that can be sampled. Another important advantage of the method proposed is that it takes explicitly into account the sensor specifications to simulate the measurements asit is actually acquired. …………
  • (Yang et al., 2022)：In this paper, we combined high spatial resolution UAS imagery with high spectral resolution imaging spectroscopy (IS) data from AVIRIS-NG and developed robust machine learning models to upscale and estimate the fractional cover (FCover) of key Arctic PFTs in the Seward Peninsula, western Alaska. We show that using this multi-platform scaling approach, the FCover of 12 dominant Arctic tundra PFTs can be accurately mapped with IS data from AVIRIS-NG (RMSE 0.14–0.22) and outperformed traditional spectral unmixing analysis (RMSE: 0.22–0.31). In particular, high-resolution UAS data provided an important avenue to obtain the quantitative training samples necessary to develop accurate scaling models. …………
  • (2) 结合前人研究，将本文中的重要结果与发现讨论一下：任何研究都是在前人的基础上开展的，所以结合前人文献，对本研究中的重要发现和结果进行对比讨论分析，以突出本研究的特色与价值。这部分比较考验作者对前人研究的掌握程度，想写出彩主要靠这一部分。
  • (Macfarlane et al., 2011)：The main conclusion of this study is that the method used to classify mixed pixels in canopy images is of little importance, once homogeneous regions of sky and canopy have been identified. Previous studies have often tested a single image classification metho. but even studies that have compared multiple methods have relied heavily on subjective visual inspection of images to rate the performance of methods although in the latter study an attempt was also made to ‘score’ methods based on more objective criteria. …………
  • (Calders et al., 2018)：Within this paper, we compared three different ground-based instruments and quantified aspects of their measurement uncertainty. Similar to Woodgate et al. (2015b), Ryu et al. (2010b), our results indicated that the agreement between these instruments does not meet the 5% accuracy specified by WMO (2012). Ryu et al. (2010b) also reports higher estimates from LAI-2x00 measurements compared to DHP in savanna ecosystems. …………
  • (3) 将本研究存在的不足以及潜在的改进方向讨论一下：任何研究都存在一定的不足与局限性，让读者明确本研究的不足，对于充分衡量本研究的结果非常重要，对于开展未来的研究也十分重要。在写作时，将论文中的一些关键假设和局限性点出来，并尽量讨论一下未来可以改进的方法。
  • (Hadi et al., 2017)：Despite our best efforts in providing ecologically realistic range of the inputs—on which results of sensitivity analysis are dependent—used in the BRF models, the difficulty, and consequently lack of available spectral measurements in humid tropical forest of South East Asia caused unavoidable uncertainties in our modelling experiments. …………
  • (Tang et al., 2019)：4.2. Estimation error and algorithm scales. In some cases, the estimation errors of remote sensing systems can contribute more than the definitional difference to the observed cross product comparisons. For example, errors in the Landsat-based tree cover datasets were clearly larger than those in the lidar datasets, and all three Landsat products had only moderate agreement with each other. The GLCF TC product used the multi-year MODIS VCF data as training data. …………
  • (4) 将本研究的应用潜力或前景讨论一下：一项好的研究一定会对某个应用或者某个小领域或者某个学科产生或多或少的推动作用，适当讨论一下本研究方法或结论的应用前景，能够让读者更加明确研究的价值。
  • (Liu et al., 2017)：The success of the method will highly bear on the realism and plasticity of the 3D canopy model used. The inter-plant variability is simulated stochastically in the current version of ADEL-Wheat model, neglecting the possible interactions between neighboring individuals. ……. Finally, the training database used to relate the LiDAR 3D point cloud features to GAI should represent well the possible variability of canopy structure, including the range of genotypic and environmental conditions.…… Nevertheless, the results of the available experiments should be organized to facilitate their compilation for the generation of the prior distributions and co-distributions of the derived model parameters.
  • (Zhang et al., 2023)：4.3. Future improvements of SIFtotal and GPPSIF. It has been shown that the relationship between SIF and GPP can become non-linear at the satellite and canopy levels under a wide range of conditions but in particular at finer scales. Clearly, consideration of the spatial and temporal scale dependencies of different SIF data is critical to assimilate the information embedded in SIF and infer GPP. …………
  • (Brüllhardt et al., 2020)：4.4. Potential areas of application. Applications across a wide range of questions benefit from spatially distinct data on radiation regimes. These span microclimate mapping over larger scales to the unravelling of patterns and interrelations of plant growth and to the improvement of ecological niche models. …………Based on our findings, the applicability of light estimation from synthetic hemispherical pictures opens the possibility to use low cost aerial photography instead of often more expensive LiDAR. ……

当然，这四个要素并不一定全部包含，有的论文中仅对(2)和(3)进行讨论，这取决于作者的习惯与研究内容的可讨论性。 Ecarnot et al., (2015) 也对讨论的写法进行了总结，我觉得还是挺不错的，可以借鉴一下。

部分文章中将“Discussion”和“Results”合并，也有部分文章中将“Discussion”和"Conclusion"合并。

2. 结论要凝练重要结果，争取短促有力

如前所述，结论重在陈述本研究做了一项什么工作以及获得了什么样的结果，所以并不需要再介绍背景和方法思路了，其内容可以在"Results"和"Discussion"中提炼。关于其写作逻辑，我总结了三个凝练：

  • (1) 将本研究的主旨或主要目标用一句话凝练出来：基本句式为本研究分析了……、展现了……，发展了……，探索了……，比较了……，提出了……，证明了……，等等。 例如：
  • (Richardson et al., 2009)：The present study investigated the applicability of various models to estimate effective LAI from aerial discrete-return LIDAR using a unique data set of ground-based LAI collected from a heterogeneous forest with a large range of LAI.
  • (Hofle et al., 2007)：In this paper two independent methods for correcting airborne laser scanning intensities were presented. The first approach – data-driven method – performs a least squares adjustment for a given empirical model including intensity and range, …………The second approach – model-driven method – is derived from the radar equation, which describes the loss of emitted pulse power.
  • (2) 将本研究的重要结果和关键讨论凝练出来：可能论文中对应多个结果小节，将每个小节的主要结果总结凝练出来，可以用数字量化指标（也可以不用），同时一些重要的讨论也可以放进去。例如：
  • (Beland et al., 2011)：The level of agreement between the leaf area estimates obtained from leaf harvesting and the TLS measurements shows the great potential value of TLSs as stand-alone tools for measuring spatially explicit individual tree leaf area with a high level of detail. The differences found suggest that for some crown shapes a 2-scan configuration may cause errors related to grazing of the leaves by the laser beams (which then generate very low intensity returns), and that the main limitation in the case of high LAI trees is occlusion effects (i.e. limited beam penetration inside the crown).
  • (Kayad et al., 2022)：The main findings can be summarized as follows: 1. PROSAIL model inversion was capable to retrieve maize LAI, presenting reasonable R2 value (0.5) and considerably low RMSE value (0.8); 2. maize yield estimation showed better performance at the late vegetative growth stages (V16), around 73 DAS; 3. maize biomass and grain yield estimations through NDRE showed higher performance in comparison with retrieved LAI, GNDVI and NDVI, when applied to an independent dataset.

(3) 将本研究的应用前景或未来展望凝练出来：与讨论中的前景展望相对应，将本研究对未来研究或应用的潜在价值点出来。例如：

  • (Aasen et al., 2015)：The described method allows the gathering of complementary hyperspectral and 3D information with only one single lightweight UAV imaging system. ………… Additionally, we hope that the described way of embedded quality assurance information will help to establish best practice procedures for small lightweight sensors to make UAV sensing systems a reliable source for remote sensing data.
  • (Chianucci et al., 2016)：Among the canopy attributes, the method allows objective evaluation of forest canopy cover, without requiring additional parameters (e.g., foliage projection coefficient, leaf angle distribution); while this option holds great potential for forest inventory purposes, the high-resolution canopy cover estimates obtainable from UAV platforms could also be used for calibrating metrics obtained from coarser-scale remote sensing products and/or analyses that use morphological processing (rather than relying only on vegetation indices), avoiding the need of ground measurements.

通常“Conclusion”的体量为1-2段，也有论文写的比较长了（其实或多或少包含了一些讨论的内容），我的个人建议依然是，结论短促有力最好。

其实如果能够克服“拿来主义”，自己反复琢磨，得到一些自己的心得，比我在这里反复讲可能更好。就怕有的同学连伸手去拿的力气都懒得花，恨不得老师们递到嘴里，那就实在无能为力了。