Journal of Geo-information Science >
Using Machine Learning Algorithms to Monitor Near-surface Freeze/Thaw State by Considering Soil Moisture in Tibetan Plateau
Received date: 2022-04-21
Revised date: 2022-07-20
Online published: 2023-02-25
Supported by
Second Tibetan Plateau Scientific Expedition and Research Program (STEP)(2019QZKK0603)
As the largest alpine permafrost area in the middle and low latitudes, permafrost and seasonally frozen soil are widely distributed in the Tibetan Plateau (TP). Accurate spatiotemporal observation of surface freeze/thaw state in the TP is important for quantifying surface energy balance, carbon and nitrogen exchange, and soil freeze-thaw erosion. However, land surface freeze/thaw state can hardly be detected in this area because of its harsh and complex geographical environment. This study aimed to employ the LightGBM algorithm and random forest algorithm to identify near-surface freeze/thaw state, based on four soil temperature & moisture observational networks. Previous studies have shown that soil moisture could significantly affect the seasonal variation characteristics of near-surface soil freeze-thaw cycles. In this study, soil moisture was introduced as a discriminant feature. In order to illustrate the contribution of microwave brightness temperature, discriminant index, and soil moisture, four different feature combination schemes were designed. We utilized AMSR2 brightness temperature data and ERA5-Land soil moisture data to identify the surface freeze/thaw state using these two machine learning algorithms. By evaluating the importance of different features based on the training set, we found that the importance score of soil moisture was high in both LightGBM and random forest algorithms, which indicates that soil moisture is a very important feature that affects freeze-thaw discrimination. To evaluate the performance of our algorithms, we compared LightGBM and random forest algorithms with a traditional freeze-thaw discriminant algorithm. Results show that the accuracy of the two machine learning algorithms was higher than that of the traditional method, and the overall rate of correct classification for daytime and nighttime was increased by 12.09%, 14.45%, respectively using LightGBM, and 13.23%, 14.96%, respectively using random forest. Nearly 80% of the misclassification occurred when the surface soil temperature was between -4.0 ℃ and 4.0 ℃. So the two machine learning algorithms are able to identify stable soil freeze/thaw state. In addition, the average RMSE of the freeze-thaw conversion days obtained by the LightGBM algorithm and the random forest algorithm decreased by 112.82, 117.00, respectively; the average RMSE of the frozen days decreased by 47.87, 53.96, respectively; and the average RMSE of the thawed days decreased by 37.10, 39.80, respectively. Based on random forest algorithm, we calculated the number of frozen days, number of thawed days, and number of freeze-thaw conversion days from July 2014 to June 2015. The accuracy assessment was carried out using the map of permafrost classification as the reference, and the total classification accuracy of frozen days within the permafrost zone was 96.78%.
XU Fubao , FAN Jianrong , ZHANG Xiyu , YANG Chao , LIU Jiali . Using Machine Learning Algorithms to Monitor Near-surface Freeze/Thaw State by Considering Soil Moisture in Tibetan Plateau[J]. Journal of Geo-information Science, 2022 , 24(12) : 2404 -2419 . DOI: 10.12082/dqxxkx.2022.220211
表1 数据源信息Tab. 1 The information of data used in this study |
数据类型 | 空间分辨率 | 时间分辨率 | 时间范围 |
---|---|---|---|
AMSR2亮温数据 | 0.1° | 1 d | 2013.01.01—2015.12.31 |
土壤湿度数据 | 0.1° | 1 h | 2013.01.01—2015.12.31 |
土壤温湿度逐时观测数据 | 站点数据 | 1 h | 2013.01.01—2015.12.31(那曲) |
2013.01.01—2015.12.31(玛曲) | |||
2013.01.01—2015.12.31(阿里) | |||
2015.06.21—2015.12.31(帕里) |
表2 训练集和测试集的数量情况Tab. 2 Statistics of training set and test set (个) |
数据集 | 时间 | 实测数据总数 | 实测数据冻融状态 | |
---|---|---|---|---|
冻结数/个 | 融化数/个 | |||
训练集 | 白天 | 15 923 | 5021 | 10 902 |
夜间 | 17 981 | 7031 | 10 950 | |
测试集 | 白天 | 31 769 | 9640 | 22 129 |
夜间 | 31 781 | 11 785 | 19 996 |
表3 4种方案冻融判别精度统计Tab. 3 Statistics of freeze-thaw discrimination accuracy for 4 schemes |
时间 | 算法 | 冻结判对率/% | 融化判对率/% | 总体判对率/% | F1分数 | |
---|---|---|---|---|---|---|
白天 | LightGBM | 方案1 | 64.39 | 90.09 | 82.34 | 0.69 |
方案2 | 71.49 | 90.10 | 84.45 | 0.74 | ||
方案3 | 67.18 | 90.24 | 83.24 | 0.71 | ||
方案4 | 73.81 | 89.37 | 84.65 | 0.74 | ||
随机森林 | 方案1 | 71.56 | 88.24 | 83.21 | 0.72 | |
方案2 | 77.14 | 89.53 | 85.77 | 0.77 | ||
方案3 | 75.19 | 87.62 | 83.85 | 0.74 | ||
方案4 | 79.80 | 88.40 | 85.79 | 0.77 | ||
夜间 | LightGBM | 方案1 | 84.83 | 86.94 | 86.16 | 0.82 |
方案2 | 86.33 | 88.49 | 87.74 | 0.84 | ||
方案3 | 85.07 | 87.39 | 86.53 | 0.82 | ||
方案4 | 87.62 | 88.19 | 87.98 | 0.84 | ||
随机森林 | 方案1 | 86.26 | 85.99 | 86.09 | 0.82 | |
方案2 | 88.22 | 88.56 | 88.43 | 0.85 | ||
方案3 | 89.79 | 86.40 | 86.17 | 0.82 | ||
方案4 | 88.38 | 88.55 | 88.49 | 0.85 |
表4 植被和裸土覆盖下冻融判别总体判对率Tab. 4 Accuracies of three algorithms under vegetation and bare soil |
地表覆盖类型 | 白天 | 夜间 | |||||
---|---|---|---|---|---|---|---|
判别式算法 | LightGBM | 随机森林 | 判别式算法 | LightGBM | 随机森林 | ||
植被 | 78.55 | 82.55 | 82.08 | 83.17 | 80.26 | 85.50 | |
裸土 | 73.01 | 82.70 | 82.15 | 84.20 | 80.48 | 84.12 |
表5 3种冻融判识算法在有效像元和重构像元处的精度Tab. 5 Accuracies of three algorithms at valid pixels and reconstructed pixels |
观测 时间 | 冻融判别 算法 | 冻结判对率/% | 融化判对率/% | 总体判对率/% | F1分数 | ||||
---|---|---|---|---|---|---|---|---|---|
有效像元 | ATC像元 | 有效像元 | ATC像元 | 有效像元 | ATC像元 | 有效像元 | ATC像元 | ||
白天 | 判别式算法 | 17.92 | 7.85 | 97.26 | 98.34 | 72.96 | 71.55 | 0.29 | 0.14 |
LightGBM | 71.89 | 78.89 | 89.85 | 88.16 | 84.35 | 85.41 | 0.74 | 0.76 | |
随机森林 | 77.41 | 86.16 | 89.00 | 86.86 | 85.45 | 86.66 | 0.77 | 0.79 | |
夜间 | 判别式算法 | 97.35 | 99.97 | 59.82 | 57.14 | 73.79 | 72.88 | 0.73 | 0.73 |
LightGBM | 85.52 | 92.98 | 88.25 | 88.02 | 87.24 | 89.84 | 0.83 | 0.87 | |
随机森林 | 86.68 | 92.71 | 88.75 | 88.06 | 87.98 | 89.77 | 0.84 | 0.87 |
表6 基于不同算法得到的日冻融转换、冻结天数和融化天数RMSE和BIAS(算法结果与实测站点数据的差值)Tab. 6 Overall performance of the three algorithms in detecting the number of daily freeze-thaw conversion days, freezing days, thawing days |
观测网 | 指标 | 判别式算法 | LightGBM算法 | 随机森林算法 | |||
---|---|---|---|---|---|---|---|
RMSE | BIAS | RMSE | BIAS | RMSE | BIAS | ||
阿里 | 日冻融转换天数 | 150.88 | 112.50 | 29.41 | 5.08 | 23.34 | 0.17 |
冻结天数 | 120.91 | 91.50 | 60.15 | 15.17 | 49.74 | 8.58 | |
融化天数 | 55.76 | 28.33 | 50.41 | -18.60 | 52.69 | -21.00 | |
玛曲 | 日冻融转换天数 | 122.44 | 101.13 | 17.80 | 10.50 | 15.20 | 7.75 |
冻结天数 | 58.43 | 41.13 | 19.92 | -7.69 | 16.62 | -5.88 | |
融化天数 | 78.37 | 64.25 | 28.92 | 10.94 | 22.95 | 4.13 | |
那曲 | 日冻融转换天数 | 151.00 | 132.89 | 38.65 | 25.39 | 34.78 | 23.32 |
冻结天数 | 69.80 | 59.93 | 25.47 | -4.25 | 20.90 | 2.43 | |
融化天数 | 89.02 | 75.39 | 32.53 | 13.36 | 28.10 | 8.61 |
注:各指标最小值均用加粗字体表示。 |
[1] |
|
[2] |
|
[3] |
|
[4] |
|
[5] |
|
[6] |
郭阳, 张廷军, 曹琳, 等. 黑河上游地表冻融指数与径流关系[J]. 水土保持通报,中国科学院水利部水土保持研究所|水利部水土保持监测中心, 2018, 38(3):222-227.
[
|
[7] |
|
[8] |
|
[9] |
|
[10] |
曹梅盛, 张铁钧. 青海高原春秋季地表土冻融的微波遥感监测[J]. 遥感学报, 1997, 1(2):139-144.
[
|
[11] |
|
[12] |
|
[13] |
|
[14] |
刘源, 秦军, 阳坤, 等. 3种土壤冻融判别算法在青藏高原的分类精度评价[J]. 地球信息科学学报, 2018, 20(8):1178-1189.
[
|
[15] |
|
[16] |
|
[17] |
|
[18] |
|
[19] |
|
[20] |
宁晓刚, 常文涛, 王浩, 等. 联合GEE与多源遥感数据的黑龙江流域沼泽湿地信息提取[J]. 遥感学报, 2022, 26(2):386-396.
[
|
[21] |
|
[22] |
|
[23] |
|
[24] |
冉有华, 李新. 中国多年冻土制图:进展、挑战与机遇[J]. 地球科学进展, 2019, 34(10):1015-1027.
[
|
[25] |
王健, 蒋玲梅, 寇晓康, 等. 根河地区冻融监测和降尺度算法的验证分析[J]. 遥感学报, 2019, 23(6):1209-1222.
[
|
[26] |
刘闻慧, 文军, 陈金雷, 等. 青藏高原土壤冻融过程关键参量时空分布特征分析[J]. 高原气象,中国科学院西北生态环境资源研究院, 2022:11-23.
[
|
[27] |
|
[28] |
张镱锂, 李炳元, 刘林山, 等. 再论青藏高原范围[J]. 地理研究, 2021, 40(6):1543-1553.
[
|
[29] |
|
[30] |
|
[31] |
|
[32] |
张天一, 苏华, 杨欣, 等. 基于LightGBM的全球海洋次表层温盐遥感预测[J]. 遥感学报, 2020, 24(10):1255-1269.
[
|
[33] |
|
[34] |
|
[35] |
|
[36] |
|
[37] |
胡同喜, 赵天杰, 施建成, 等. AMSR-E与AMSR2被动微波亮温数据交叉定标[J]. 遥感技术与应用, 2016, 31(5):919-924.
[
|
[38] |
|
[39] |
|
[40] |
寇晓康, 张玉芝, 靳梦杰, 等. 基于多层土壤温度的地表冻融变化被动微波遥感验证分析[J]. 地理与地理信息科学, 2020, 36(3):49-55.
[
|
[41] |
周幼吾, 郭东信, 邱国庆, 等. 中国冻土[M]. 北京: 科学出版社, 2000.
[
|
/
〈 | 〉 |