Research article

Application of XGBoost model and multi-source data for winter wheat yield prediction in Henan Province of China

  • Received: 25 March 2025 Revised: 19 May 2025 Accepted: 01 June 2025 Published: 18 June 2025
  • Accurate prediction of winter wheat yield is essential for precision agricultural management. Traditional methods, which primarily rely on vegetation indices, often fall short in achieving high prediction accuracy. In this study, we proposed an innovative approach by integrating multi-source data, including vegetation, geographical, and soil features, with 61 features extracted across growth stages. Using these features, the XGBoost regression model was applied to predict winter wheat yield in Henan Province, China. The model was trained on data from 2016 to 2020 and validated using the 2021 dataset. The XGBoost model achieved a mean absolute error (MAE) of 371.36 kg/hm2, a root mean square error (RMSE) of 516.97 kg/hm2, and a coefficient of determination (R2) of 0.85. These results significantly outperformed other models, such as random forest (RF), gradient boosting decision tree (GBDT), and Lasso. Notably, the XGBoost model provided high accuracy 2–3 growth stages before harvest, enabling earlier yield prediction than traditional methods. To enhance interpretability, SHAP (shapley additive explanations) was employed to quantify the influence of each input variable on yield and determine the direction of influence. This study underscores the potential of multi-source data and advanced machine learning techniques for accurate and interpretable winter wheat yield prediction, providing a valuable solution for precision agriculture.

    Citation: Hua Li, Jingyi Gao, Yadan Guo, Xianzhi George Yuan. Application of XGBoost model and multi-source data for winter wheat yield prediction in Henan Province of China[J]. Big Data and Information Analytics, 2025, 9: 29-47. doi: 10.3934/bdia.2025002

    Related Papers:

    [1] Suping Wang . The random convolution sampling stability in multiply generated shift invariant subspace of weighted mixed Lebesgue space. AIMS Mathematics, 2022, 7(2): 1707-1725. doi: 10.3934/math.2022098
    [2] R. Mareay, Radwan Abu-Gdairi, M. Badr . Soft rough fuzzy sets based on covering. AIMS Mathematics, 2024, 9(5): 11180-11193. doi: 10.3934/math.2024548
    [3] Necati Can Açıkgöz, Ceren Sultan Elmalı . Nearly Menger covering property via bitopological spaces. AIMS Mathematics, 2024, 9(12): 34042-34066. doi: 10.3934/math.20241623
    [4] Mohammed Abu Saleem . On soft covering spaces in soft topological spaces. AIMS Mathematics, 2024, 9(7): 18134-18142. doi: 10.3934/math.2024885
    [5] Jeong-Kweon Seo, Byeong-Chun Shin . Reduced-order modeling using the frequency-domain method for parabolic partial differential equations. AIMS Mathematics, 2023, 8(7): 15255-15268. doi: 10.3934/math.2023779
    [6] Justin Tzou, Brian Wetton . Optimal covering points and curves. AIMS Mathematics, 2019, 4(6): 1796-1804. doi: 10.3934/math.2019.6.1796
    [7] Panpan Jia, Jizhu Nan, Yongsheng Ma . Separating invariants for certain representations of the elementary Abelian p-groups of rank two. AIMS Mathematics, 2024, 9(9): 25603-25618. doi: 10.3934/math.20241250
    [8] Zhenzhen Jiang, Jun Yue, Xia Zhang . Polychromatic colorings of hypergraphs with high balance. AIMS Mathematics, 2020, 5(4): 3010-3018. doi: 10.3934/math.2020195
    [9] Imran Shahzad Khan, Choonkil Park, Abdullah Shoaib, Nasir Shah . A study of fixed point sets based on Z-soft rough covering models. AIMS Mathematics, 2022, 7(7): 13278-13291. doi: 10.3934/math.2022733
    [10] Muhsin Incesu . LS (3)-equivalence conditions of control points and application to spatial Bézier curves and surfaces. AIMS Mathematics, 2020, 5(2): 1216-1246. doi: 10.3934/math.2020084
  • Accurate prediction of winter wheat yield is essential for precision agricultural management. Traditional methods, which primarily rely on vegetation indices, often fall short in achieving high prediction accuracy. In this study, we proposed an innovative approach by integrating multi-source data, including vegetation, geographical, and soil features, with 61 features extracted across growth stages. Using these features, the XGBoost regression model was applied to predict winter wheat yield in Henan Province, China. The model was trained on data from 2016 to 2020 and validated using the 2021 dataset. The XGBoost model achieved a mean absolute error (MAE) of 371.36 kg/hm2, a root mean square error (RMSE) of 516.97 kg/hm2, and a coefficient of determination (R2) of 0.85. These results significantly outperformed other models, such as random forest (RF), gradient boosting decision tree (GBDT), and Lasso. Notably, the XGBoost model provided high accuracy 2–3 growth stages before harvest, enabling earlier yield prediction than traditional methods. To enhance interpretability, SHAP (shapley additive explanations) was employed to quantify the influence of each input variable on yield and determine the direction of influence. This study underscores the potential of multi-source data and advanced machine learning techniques for accurate and interpretable winter wheat yield prediction, providing a valuable solution for precision agriculture.



    In information sciences, in order to extract useful information hidden in voluminous data, many methods were proposed, including classical logics and classical mathematics. Rough set theory, introduced by Z. Pawlak in [1], plays an important role in applications of these methods (see [2,3,4,5,6,7,8,9], for example). In the classical rough set theory, Pawlak approximation spaces are based on partitions of the universe of discourse, but this requirement is not satisfied in some situations [10]. In the past years, Pawlak approximation spaces were extended to covering approximation spaces, and used in information sciences (see [10,11,12,13,14,15,16,17,18,19,20,21,22], for example).

    It is easy to see that topological spaces on finite universes are special covering approximation spaces. Thus, some topological properties are considered in covering approximation spaces. In addition, Separations play important roles in topological spaces. Recently, Si-separations (i=0,1,2,3,d,r) in covering approximation spaces draw our attention. And some interesting results were obtained ([12,13]).

    However, just as separations in topological spaces [23,24,25], the theoretical research framework about separations in covering approximation spaces can be constructed by relations, characterizations and invariance. More precisely, this framework can be described as follows.

    Framework 1. The theoretical research framework about separations in covering approximation spaces consists of the following three parts.

    Part 1: Establish some relations among separations in covering approximation spaces. These relations mainly include implications, not implications, and equivalences.

    Part 2: Give some equivalent characterizations of separations in covering approximation spaces. These equivalent characterizations are mainly shown by elements of covering, the covering upper approximation of subsets and the covering lower approximation of subsets, respectively.

    Part 3: Study invariance of separation in covering approximation spaces. The invariance of separation mainly includes invariance of separation about reducts of covering, subspaces and transformations.

    Part 1 and Part 2 of Framework 1 were researched in [12,13]. However, there are not results on Part 3 of Framework 1. Obviously, this research can be performed by means of investigations of the following questions.

    Question 1. Let i=0,1,2,3,d,r.

    (1) Is Si-separation in covering approximation spaces invariant to reducts of coverings?

    (2) Is Si-separation in covering approximation spaces invariant to covering approximation subspaces?

    (3) Is Si-separation in covering approximation spaces invariant under some transformations of covering approximation spaces?

    This paper investigates covering upper approximations and covering lower approximations of subsets in covering approximation spaces and their reduction spaces (resp. their subspaces, their transformations), and give some relations among covering approximations of these subsets. Based on these results, we answer Question 1 positively.

    The remainder of this paper is organized as follows. Section 2 reviews some definitions and remarks. The invariance of separations in covering approximation spaces are considered in Sections 3–5. In section 6, some concluding remarks are provided.

    To begin with, the definition of covering approximation space is reviewed.

    Definition 1. (cf. [1,22]) Let U, the universe of discourse, be a set and C be a family of subsets of U.

    (1) C is called a covering of U if {K:KC}=U. Furthermore, C is called a partition on U if also KK= for all K,KC, where KK.

    (2) The pair (U;C) is called a covering approximation space (resp. a Pawlak approximation space) if C is a covering (resp. a partition) of U.

    All results this paper proposed are based on the assumption that the universe U is a finite set. Next we review the following upper approximation operator ¯C and lower approximation operator C_ on (U;C) [17].

    Definition 2. (cf. [17]) Let (U;C) be a covering approximation space. For each XU, put

    C_(X)={K:KC and KX},

    ¯C(X)=UC_(UX).

    (1) ¯C:2U2U is called a covering upper approximation operator, simply denoted by X.

    (2) C_:2U2U is called a covering lower approximation operator, simply denoted by X.

    Before introducing concepts of separations, we give some notations as follows.

    Notation 1. Let (U;C) be a covering approximation space. Throughout this paper, we use the following notations, where F is a family of subsets of U and xU.

    (1) F=FFF.

    (2) F=FFF.

    (3) Cx={K:xKC}.

    (4) N(x)=Cx.

    (5) ¯N(x)={K:KCx}.

    (6) D(x)=U(CCx).

    Now we define separations in covering approximation spaces.

    Definition 3. (cf. [12]) Let (U;C) be a covering approximation space. Some separations are defined as follows.

    (1) S0-separation: if x,yU and xy, then there exists KC such that K{x,y}={x} or K{x,y}={y}.

    (2) S1-separation: if x,yU and xy, then there exist Kx,KyC such that Kx{x,y}={x} and Ky{x,y}={y}.

    (3) S2-separation: if x,yU and xy, then there exist Kx,KyC such that xKx, yKy and KxKy=.

    (4) S3-separation: if xU and xXU, then there exists KC such that xK and KX=.

    (5) Sd-separation: if xU, then there exists KC such that {x}=KD(x).

    (6) Sr-separation: if xKC, then D(x)K.

    Some explanations of separations are provided in the following.

    Remark 1. Let (U;C) be a covering approximation space.

    (1) If (U;C) satisfies S0-separation, then for every pair of distinct x,yU there is KC containing exactly one of x and y.

    (2) If (U;C) satisfies S1-separation, then for every pair of distinct x,yU there is KC containing x and KC containing y such that K does not contain y and K does not contain x.

    (3) If (U;C) satisfies S2-separation, then every pair of distinct x,yU can be separated from elements of C.

    (4) If (U;C) satisfies S3-separation, then for each xU and each subset X of U not containing x there is KC containing x such that K and X are separated.

    (5) If (U;C) satisfies Sd-separation, then for each xU there is KC containing x such that whenever yU and yx, either K does not contain y or K contains y for KC satisfying xK.

    (6) (U;C) satisfies Sr-separation, i.e., for each xU and each KC containing x, if yU such that KC does not contain x implying K does not contain y, then K contains y.

    Let i=0,1,2,3,d,r. Throughout this paper, we call a covering approximation space (U;C) with Si-separation an Si-covering approximation space. For short, Si-covering approximation spaces are called Si-spaces (Note: Si-separations and Si-spaces in [13] are called Gi-separations and Gi-spaces, respectively).

    The following example illustrates the concepts of separation.

    Example 1. (1) Let U={a,b,c} and C={{a,b},{c}}. Then (U,C) is a Pawlak approximation space. Each Pawlak approximation space is an Sr-space ([12,Remark 3.4]). Hence, (U,C) is an Sr-space. On the other hand, N(a)={a,b}{a}, so (U,C) is not an S1-space ([12,Theorem 3.6]).

    (2) Let U={a,b,c} and C={{a,c},{b,c}}. It is easy to check that (U,C) is an S0-space. Since D(c)=U and KD(x)=K{c} for each KC, (U;C) is not an Sd-space.

    (3) Let U={a,b} and C={{a,b},{a}}. Since D(a)={a,b} and D(b)={b}, we have {a}={a}D(a) and {b}={a,b}D(b), Note that {a}, {a,b}C. So (U,C) is an Sd-space. On the other hand, N(b)={a,b}{b}, so (U,C) is not an S1-space ([12,Theorem 3.6]).

    (4) Let U={a,b,c} and C={{a,b},{a,c},{b,c}}. Since N(a)={a}, N(b)={b} and N(c)={c}, (U,C) is an S1-space. Whenever K,KC, KK, So (U,C) is not an S2-space.

    (5) Let U={a,b,c,d} and C={{a,b},{a,c},{c,d},{b,d}}. It is not difficult to check that (U,C) is an S2-space. Since {a}C, (U,C) is not an S3-space ([12,Theorem 3.13]).

    The following lemma shows the relation among all separations in covering approximation spaces.

    Lemma 1. [13] The following implications are known and none of these implications can be reversed by Example 1.

    (1) S3-space S2-space S1-space Sd-space S0-space.

    (2) S1-space Sr-space.

    However, we have the following relations among separations in Pawlak approximation spaces, which further illustrates the concepts of separation.

    Corollary 1. Let (U;C) be a Pawlak approximation space. Then the following are equivalent:

    (1) (U;C) is an S3-space,

    (2) (U;C) is an S2-space,

    (3) (U;C) is an S1-space,

    (4) (U;C) is an Sd-space,

    (5) (U;C) is an S0-space.

    Proof. By Lemma 1, we have (1) (2) (3) (4) (5). So we only need to prove that (5) (1).

    Suppose (U;C) is an S0-space. Since (U;C) is a Pawlak approximation space, C is a partition of U. Let xU and xXU. Take KC satisfying xK. It suffices to prove that KX=. In fact, if KX, then there is yKX. So we obtain yx. Since (U;C) is an S0-space, there is KC such that K{x,y}={x} or K{x,y}={y}. Without loss of generality, we assume K{x,y}={x}. Then yK. It is not hard to see that KK and xKK. This contradicts that C is a partition of U. So KX=.

    Remark 2. Separations in covering approximation spaces play an important role in not only applications but also theoretical research of rough set theory. For example, in the fields of rough set data analysis, it is an interesting question how to characterize the conditions under which {N(x):xU} forms a partition of U for a covering approximation space (U;C) ([17,26]). Reference [12] proves that a covering approximation space (U;C) is an Sr-space if and only if {N(x):xU} forms a partition of U. Thus, we can investigate the above question by discussing Sr-spaces. Furthermore, one concludes that {N(x):xU} forms a partition of U if C is a partition of U.

    The following results are known.

    Lemma 2. (cf. [13,17,22]) Let (U;C) be a covering approximation space. Then the following results hold:

    (1) U=U=U and ==,

    (2) if XU, then XXX,

    (3) if XYU, then XY and XY,

    (4) if XU, then (X)=X and (X)=X,

    (5) if KC, then K=K,

    (6) if XU, then (UX)=UX and (UX)=UX,

    (7) if xU, then D(x)={x}.

    The characteristics about all separations in covering approximation spaces are provided.

    Theorem 1. (cf. [13]) Let (U;C) be a covering approximation space. Then the following results hold:

    (1) (U;C) is an S0-space if and only if {x}{y} for each pair x,yU with xy,

    (2) (U;C) is an S1-space if and only if {x}=N(x) for each xU,

    (3) (U;C) is an S2-space if and only if {x}=¯N(x) for each xU,

    (4) (U;C) is an S3-space if and only if {x}C for each xU,

    (5) (U;C) is an Sd-space if and only if whenever xU, {x}=XY for some X,YU,

    (6) (U;C) is an Sr-space if and only if x{y}y{x} for each pair x,yU.

    To consider the invariance of separation in covering approximation spaces, the reduct of covering is introduced firstly.

    Definition 4. (cf. [20,21]) Let U be the universe of discourse and C be a covering of U.

    (1) KC is called reducible in C if K=F for FC{K}. Otherwise, K is called irreducible in C.

    (2) C is called irreducible if K is irreducible in C for each KC; otherwise C is called reducible.

    (3) C is called the reduct of C if C is obtained by deleting all reducible elements in C. It is clear that C is an irreducible subcovering of C.

    Proposition 1. Let C be a covering of the set U. Each set in C is a union of some irreducible elements of C.

    However, Proposition 1 does not hold for infinite set U. A counter example is provided to show it.

    Example 2. Let U be the set of all positive integers. Assume C={{1,2},{3},{4},{5},{6},}{{2}{3,4,5,6,},{2}{4,5,6,},{2}{5,6,},} is a covering of U. It is easy to see that {1,2},{3},{4},{5},{6}, are irreducible elements of C. But {2}{n,n+1,n+2,} is a reducible element of C because {2}{n,n+1,n+2,}={2}{n+1,n+2,}{n} holds, where nU and n>2. Thus, {2}{n,n+1,n+2,} is not a union of some irreducible elements of C.

    Let (U;C) be a covering approximation space, C be the reduct of C and XU. According to Definition 4, C is the unique reduct of C. In this section, the covering lower approximation of X and the covering upper approximation of X in covering approximation space (U;C) are denoted by X# and X#, respectively. For each xU, D(x)=U(CCx), N(x)={K:KCx}=Cx and ¯N(x)={K#:KCx}. The following propositions can be provided.

    Proposition 2. Let (U;C) be a covering approximation space and C be the reduct of C. For each xU, if KCx, then KK for some KCx.

    Proof. (1) Let xU and KCx. If KCx, then KK for K=KCx. If KCx, then K is a reducible elements in C. It is not difficult to see that there exists FC such that K=F. Since xK, we choose KF satisfying xK. Thus, one derives KCx and KK.

    Proposition 3. [21] Let (U;C) be a covering approximation space and C be the reduct of C. For XU, X=X# and X=X#.

    Proposition 4. Let (U;C) be a covering approximation space, C be the reduct of C and xU. Then the following conclusions hold:

    (1) D(x)=D(x),

    (2) N(x)=N(x),

    (3) ¯N(x)=¯N(x).

    Proof. (1) It suffices to prove that (CCx)=(CCx). Assume y(CCx). Then yK for some KCCx. By Proposition 2 (1), there is KC such that yKC and KK. KCCx because xK and xK, i.e., KCx. It follows that y(CCx). Conversely, we assume y(CCx). Then yK for some KCCx. So KCC and xK. Thus KCCx. It follows that y(CCx). It proves that (CCx)=(CCx).

    (2) Since CC, CxCx, one sees N(x)=CxCx=N(x). On the other hand, we assume yN(x). Whenever KCx, by Proposition 2 (1), there is KCx such that KK, hence yN(x)=CxKK. It proves that yN(x). So N(x)N(x). Consequently, we have N(x)=N(x).

    (3) Since CC, CxCx, we have ¯N(x)={K:KCx}{K#:KCx}=¯N(x). On the other hand, if y¯N(x), then xK# for each KCx. Whenever KCx, by Proposition 2 (1), there is KCx such that KK. Based on Lemma 2 (3) and Proposition 2 (2), it is not hard to see that y¯N(x)K#=KK. This proves that y¯N(x). So ¯N(x)¯N(x). Consequently, one gets ¯N(x)=¯N(x).

    Now we give the main theorem of this section.

    Theorem 2. Let (U;C) be a covering approximation space and C be the reduct of C. Then the following conclusions hold:

    (1) (U;C) is an S0-space if and only if (U;C) is an S0-space,

    (2) (U;C) is an S1-space if and only if (U;C) is an S1-space,

    (3) (U;C) is an S2-space if and only if (U;C) is an S2-space,

    (4) (U;C) is an S3-space if and only if (U;C) is an S3-space,

    (5) (U;C) is an Sd-space if and only if (U;C) is an Sd-space,

    (6) (U;C) is an Sr-space if and only if (U;C) is an Sr-space.

    Proof. (1) By Proposition 2 (2), {x}{y} if and only if {x}#{y}# for each pair x,yU with xy. By Theorem 1 (1), (U;C) is an S0-space if and only if (U;C) is an S0-space.

    (2) By Proposition 4 (2), N(x)=N(x) for each xU. It follows that {x}=N(x) if and only if {x}=N(x). By Theorem 1 (2), (U;C) is an S1-space if and only if (U;C) is an S1-space.

    (3) By Proposition 4 (3), ¯N(x)=¯N(x) for each xU. It follows that {x}=¯N(x) if and only if {x}=¯N(x). By Theorem 1 (3), (U;C) is an S2-space if and only if (U;C) is an S2-space.

    (4) By Definition 4, {x}C if and only if {x}C for each xU. By Theorem 1 (4), (U;C) is an S3-space if and only if (U;C) is an S3-space.

    (5) By Proposition 2 (2), whenever xU, {x}=XY if and only if {x}=X#Y# for some X,YU. By Theorem 1 (5), (U;C) is an Sd-space if and only if (U;C) is an Sd-space.

    (6) By Proposition 2 (2), for each pair x,yU, x{y}y{x} if and only if x{y}#y{x}#. By Theorem 1 (6), (U;C) is an Sr-space if and only if (U;C) is an Sr-space.

    Remark 3. For each i=0,1,2,3,d,r, Theorem 2 shows that Si-separation in covering approximation spaces is invariant with respect to the reduct of coverings. It indicates that we can reduce some elements of C without influencing separations. In network applications, some existing approaches deal with the covering directly. In contrast, we deal with C', which is the reduct of C. Therefore, both time and space requirements are reduced, and network securities are preserved.

    The following definition presents covering approximation subspaces.

    Definition 5. (cf. [11]) Let (U;C) be a covering approximation space and UU. Take C={KU:KC}, then (U;C) is a covering approximation space. (U;C) is called a subspace of (U;C).

    Let (U;C) be a covering approximation space, (U;C) be a subspace of (U;C) and XU. In this section, the covering lower approximation of X and the covering upper approximation of X in covering approximation space (U;C) are denoted by X# and X#, respectively. For each xU, D(x)=U(CCx), N(x)={K:KCx}=Cx and ¯N(x)={K#:KCx}. The following two results are proved.

    Proposition 5. Let (U;C) be a covering approximation space and (U;C) be a subspace of (U;C). Then the following conclusions hold:

    (1) If XU, then (XU)#XU.

    (2) If XU, then XU(XU)#.

    Proof. (1) Because (XU)#=U(UXU)#=U(UX)# and XU=(U(UX))U=UU(UX), it suffices to prove that U(UX)(UX)#. Assume xU(UX). Then xU and there is LC such that xLUX. It follows that xLUUX. Note that LUC. So x(UX)#. It proves that U(UX)(UX)#.

    (2) Assume xXU. Then xU and there is KC such that xKX. Thus xKUXU. Note that KUC. So x(XU)#. It proves that XU(XU)#.

    Remark 4. In Proposition 5 (1) and (2), "" can not be replaced by"=" (see Examples 2 and 4). However, we have the following result.

    Proposition 6. Let (U;C) be a covering approximation space and (U;C) be a subspace of (U;C). If XU, then X#=XU. Specially, {x}#={x}U if xU.

    Proof. Assume XU. By Proposition 5 (1), X#XU, we only need to prove XUX#. Note that XU=(U(UX))U=U(UX)U and X#=U(UX)#. It suffices to prove that (UX)#(UX)U. Assume y(UX)#. Then there is KC such that yKUX, i.e., there is KC such that yKUUX. Thus (KU)X=, i.e., KX=. Hence, we derive yKUX. It follows that y(UX)U. It proves that (UX)#(UX)U.

    The following examples show that "" is not replaced by "=" in Proposition 5 (1) and (2).

    Example 3. Let U={a,b,c}, C={{a,b},{c}}, U={b,c}, C={{b},{c}} and X={a,c}. Then XU(XU)#. In fact, (XU)#={c}#=U(U{c})#=U{b}#=U{b}={c}. On the other hand, X=U(UX)=U{b}=U=U. It follows that XU=UU=U={b,c}. So XU(XU)#.

    Example 4. Let U={a,b}, C={{a,b}}, U={a}, C={{a}} and X={a}. Then (XU)#XU. In fact, XU= and (XU)#={a}.

    To show the invariance of separation, the following results are proved.

    Proposition 7. Let (U;C) be a covering approximation space, (U;C) be a subspace of (U;C) and xU. Then the following conclusions hold:

    (1) D(x)U=D(x),

    (2) N(x)U=N(x),

    (3) ¯N(x)U¯N(x).

    Proof. By computing directly, we have the following results.

    (1) D(x)U=(U(CCx))U=UU((CCx))=UU({K:xKC})=U{KU:xKC}=U{K:xKUC}=U(CCx)=D(x).

    (2) N(x)U=({K:xKC})U={KU:xKC}={K:xKC}=N(x).

    (3) ¯N(x)U=({K:xKC})U={KU:xKC}{(KU)#:xKC}={K#:xKC}=¯N(x).

    Now we give the main theorem of this section.

    Theorem 3. Let (U;C) be a covering approximation space and (U;C) be a subspace of (U;C). Then the following conclusions hold:

    (1) if (U;C) is an S0-space, then (U;C) is an S0-space,

    (2) if (U;C) is an S1-space, then (U;C) is an S1-space,

    (3) if (U;C) is an S2-space, then (U;C) is an S2-space,

    (4) if (U;C) is an S3-space, then (U;C) is an S3-space,

    (5) if (U;C) is an Sd-space, then (U;C) is an Sd-space,

    (6) if (U;C) is an Sr-space, then (U;C) is an Sr-space.

    Proof. (1) Assume (U;C) is an S0-space. By Theorem 1 (1), {x}{y} for any x,yU satisfying xy. We claim that {x}#{y}#. In fact, if {x}#={y}#, then {x}U={y}U from Proposition 6. Thus x{x}U={y}U{y}. That is {x}{y}. Similarly, we obtain {y}{x}. It follows that {x}={y}. This is a contradiction.

    (2) Assume (U;C) is an S1-space. By Theorem 1 (2), N(x)={x} for each xU. Thus, we have N(x)U={x}. From Proposition 7 (2), one sees N(x)=N(x)U={x}. According to Theorem 1 (2), (U;C) is an S1-space.

    (3) Assume (U;C) is an S2-space. By Theorem 1 (3), ¯N(x)={x} for each xU. Thus, we have ¯N(x)U={x}. By Proposition 7 (3), {x}¯N(x)¯N(x)U={x}, i.e., ¯N(x)={x}. By Theorem 1 (3), (U;C) is an S2-space.

    (4) Assume (U;C) is an S3-space. By Theorem 1 (4), {x}C for each xU. Thus, we have {x}={x}UC. By Theorem 1 (4), (U;C) is an S3-space.

    (5) Assume (U;C) is an Sd-space. Take xU. Then there is KC such that {x}=KD(x). Thus, we obtain {x}=(KU)(D(x)U). By Proposition 7 (1), D(x)U=D(x). It follows that {x}=(KU)D(x). Note that KUC. So (U;C) is an Sd-space.

    (6) Assume (U;C) is an Sr-space. Let x,yU and x{y}#. By Theorem 1 (6), we only need to prove that y{x}#. Since {y}#={y}U from Proposition 6, x{y}. By Theorem 1 (6), y{x} because (U;C) is an Sr-space. It follows that y{x}U={x}#.

    Remark 5. For each i=0,1,2,3,d,r, Theorem 3 shows that Si-separation in covering approximation spaces is invariant to covering approximation subspaces. It indicates that some dynamic methods [27] are applicable to covering based rough sets theory. We can employ different subsets of U to produce different coverings, and obtain more stable rules.

    In this section, we use the following notations for covering approximation space (U;C). For XU, the covering lower approximation of X and the covering upper approximation of X in covering approximation space (U;C) are denoted by X# and X#, respectively. For xU, D(x)=U(CCx), N(x)={K:KCx}=Cx and ¯N(x)={K#:KCx}. The definition of transformation is proposed.

    Definition 6. Let (U;C) and (U;C) be two covering approximation spaces. If f:UU is a bijective mapping, then f is called a transformation from (U;C) to (U;C).

    Remark 6. It is clear that if f is a transformation from (U;C) to (U;C), then f1 is a transformation from (U;C) to (U;C).

    Remark 7. Let (U;C) and (U;C) be two covering approximation spaces, f be a transformation from (U;C) to (U;C). Whenever A,BU and X,YU. Then the following conclusions hold:

    (1) f(f1(X))=X,

    (2) f1(f(A))=A,

    (3) f(AB)=f(A)f(B),

    (4) f1(XY)=f1(X)f1(Y),

    (5) f(AB)=f(A)f(B),

    (6) f1(XY)=f1(X)f1(Y),

    (7) f(AB)=f(A)f(B),

    (8) f1(XY)=f1(X)f1(Y),

    (9) ABf(A)f(B),

    (10) XYf1(X)f1(Y).

    In general, Si-separation in covering approximation spaces is not invariant under transformations of covering approximation spaces for i=0,1,2,3,d,r.

    Example 5. Let (U;C) and (U;C) be two covering approximation spaces, where U={a,b,c,d}, U={x,y,z,w}, C={{a},{b},{c},{d},{a,b},{b,c},{c,d}} and C={{x,y},{x,y,z},{z,w}}. Take f:UU, where f(a)=x, f(b)=y, f(c)=z and f(d)=w. Then f is a transformation from (U;C) to (U;C).

    (1) Since {u}C for each uU, (U;C) is an S3-space from Theorem 1 (4). By Remark 1, (U;C) is an Si-space for each i=0,1,2,3,d,r.

    (2) Note that {x}#=U(U{x})#=U{y,z,w}#=U{z,w}={x,y} and {y}#=U(U{y})#=U{x,z,w}#=U{z,w}={x,y}. So we derive {x}#={y}#. Hence, (U;C) is not an S0-space from Theorem 1 (1). By Remark 1 (1), (U;C) is not an Si-space for each i=0,1,2,3,d.

    (3) It is clear that N(z)={z} and N(w)={z,w}. So zN(w) and wN(z). By Theorem 1 (6), (U;C) is not an Sr-space.

    However, which classes of transformations preserve separations in covering approximation spaces? It is one of important topics in research framework about separations in covering approximation spaces. In order to answer this question, we introduce the following definitions.

    Definition 7. Let (U;C) and (U;C) be two covering approximation spaces, and f be a transformation from (U;C) to (U;C).

    (1) f is called a covering lower approximation-preserving transformation (abbr. CLAP-transformation) if f(A)=(f(A))# for all AU.

    (2) f is called a covering upper approximation-preserving transformation (abbr. CUAP-transformation) if f(A)=(f(A))# for all AU.

    (3) f is called a covering approximation-preserving transformation (abbr. CAP-transformation) if f is a covering both lower and upper approximation-preserving transformation.

    The following proposition is obvious.

    Proposition 8. Let (U;C) and (U;C) be two covering approximation spaces, and f be a transformation from (U;C) to (U;C). Then the following conditions are equivalent:

    (1) f is a CLAP-transformation,

    (2) f is a CUAP-transformation,

    (3) f is a CAP-transformation.

    According to Proposition 8, one obtains Lemma 3.

    Lemma 3. Let (U;C) and (U;C) be two covering approximation spaces. Then the following conditions are equivalent:

    (1) f is a CAP-transformation from (U;C) to (U;C),

    (2) f1 is a CAP-transformation from (U;C) to (U;C).

    Proof. We only need to prove (1) (2) because f=(f1)1.

    Assume f be a CAP-transformation from (U;C) to (U;C). Then f1 is a transformation from (U;C) to (U;C) by Remark 6. Whenever XU, we get (f(f1(X)))#=f((f1(X))) because f is a CLAP-transformation. Combined with Lemma 7 (1) and (2), f1(X#)=f1((f(f1(X)))#)=f1(f((f1(X))))=(f1(X)). It shows that f1 is a CLAP-transformation. By Proposition 8, f1 is a CAP-transformation.

    Lemma 4. Let (U;C) and (U;C) be two covering approximation spaces, and f be a CAP-transformation from (U;C) to (U;C). If aU and x=f(a)U, then the following conclusions hold:

    (1) f(N(a))=N(x),

    (2) f(¯N(a))=¯N(x).

    Proof. For each KCa, K=K from Lemma 2 (5). Then one sees f is a CAP-transformation. So we get x=f(a)f(K)=f(K)=(f(K))#. Hence, there is HKCx such that HKf(K).

    (1) f(N(a))=f({K:KCa})={f(K):KCa}{HK:KCa}N(x). Similarly, f1(N(x))N(a) because f1 is a CAP-transformation from Proposition 3. By Lemma 7 (1), N(x)=ff1(N(x))f(N(a)). Consequently, we obtain f(N(a))=N(x).

    (2) f(¯N(a))=f({K:KCa})={f(K):KCa}{H#K:KCa}¯N(x). By the same method, f1(¯N(x))¯N(a) because f1 is a CAP-transformation from Proposition 4.7. By Lemma 4.2(1), ¯N(x)=ff1(¯N(x))f(¯N(a)). Consequently, f(¯N(a))=¯N(x).

    Based on the analysis above, the following results are proved.

    Theorem 4. Let (U;C) and (U;C) be two covering approximation spaces, and f be a CAP-transformation from (U;C) to (U;C). Then the following conclusions hold:

    (1) (U;C) is an S0-space if and only if (U;C) is an S0-space,

    (2) (U;C) is an S1-space if and only if (U;C) is an S1-space,

    (3) (U;C) is an S2-space if and only if (U;C) is an S2-space,

    (4) (U;C) is an S3-space if and only if (U;C) is an S3-space,

    (5) (U;C) is an Sd-space if and only if (U;C) is an Sd-space,

    (6) (U;C) is an Sr-space if and only if (U;C) is an Sr-space.

    Proof. By Proposition 3, we only need to prove "only if" parts.

    (1) Assume (U;C) is an S0-space. For any x,yU satisfying xy, there are a,bU with ab such that x=f(a) and y=f(b). Because (U;C) is an S0-space, {a}{b} from Theorem 1 (1). It follows that f({a})f({b}) from Lemma 3 (9). Since f is a CAP-transformation, one sees {f(a)}#=f({a}) and {f(b)}#=f({b}). Consequently, {x}#={f(a)}#=f({a})f({b})={f(b)}#={y}#. By Theorem 1 (1), (U;C) is an S0-space.

    (2) Assume (U;C) is an S1-space. For each xU, there is aU such that x=f(a). Because (U;C) is an S1-space, we have {a}=N(a) from Theorem 1 (2). By Lemma 4 (1), one gets {x}={f(a)}=f({a})=f(N(a))=N(x). It follows that (U;C) is an S1-space from Theorem 1 (2).

    (3) Assume (U;C) is an S2-space. For each xU, there is aU such that x=f(a). Because (U;C) is an S2-space, {a}=¯N(a) from Theorem 1 (3). By Lemma 4 (2), we have {x}={f(a)}=f({a})=f(¯N(a))=¯N(x). It follows that (U;C) is an S2-space from Theorem 1 (3).

    (4) Assume (U;C) is an S3-space. For each xU, there is aU such that x=f(a). Because (U;C) is an S3-space, we obtain {a}C from Theorem 1 (4). Hence, one gets {a}={a}. Since f is a CAP-transformation, it is easy to see that f({a})=(f({a}))#. Consequently, {x}={f(a)}=f({a})=f({a})=(f({a}))#={f(a)}#={x}#. It follows that {x}C. By Theorem 1 (4), (U;C) is an S3-space.

    (5) Assume (U;C) is an Sd-space. For each xU, there is aU such that x=f(a). Because (U;C) is an Sd-space, there are A,BU such that {a}=AB from Theorem 1 (5). Since f is a CAP-transformation, one sees f(A)=(f(A))# and f(B)=(f(B))#. Combined with Lemma 7 (5), we have {x}={f(a)}=f(AB)=f(A)f(B)=(f(A))#(f(B))#. Note that f(A),f(B)U. So (U;C) is an Sd-space from Theorem 1 (5).

    (6) Assume (U;C) is an Sr-space. For any x,yU satisfying x{y}#, there are a,bU such that x=f(a) and y=f(b). Thus, we find a=f1(x) and b=f1(y). Since f is a CAP-transformation from (U;C) to (U;C), f1 is a CAP-transformation from (U;C) to (U;C) by Proposition 3. So f1({y}#)=(f1({y})). Consequently, a=f1(x)f1({y}#)=(f1({y}))={b}. Because (U;C) is an Sr-space, we obtain b{a} from Theorem 1 (6). It follows that y=f(b)f({a})=(f({a}))#={x}# because f is a CAP-transformation. Hence, (U;C) is an Sr-space from Theorem 1 (6).

    Remark 8. In Theorem 4, "CAP" can be replaced by "CLAP" or "CUAP" from Proposition 8 and "f be a CAP-transformation from (U;C) to (U;C)" can be replaced by "f1 be a CAP-transformation from (U;C) to (U;C)" from Proposition 3.

    Remark 9. For each i=0,1,2,3,d,r, Theorem 4 shows that Si-separation in covering approximation spaces is invariant under CAP-transformations of covering approximation spaces. It indicates that some transformations methods are applicable to covering-based rough sets theory. We can employ different the universe of discourse with different coverings to produce the same separations.

    This paper studies some invariance of separations in covering approximation spaces. Three main theorems, Theorems 2–4, are obtained. These are important results in the theoretical research framework about separations in covering approximation spaces. These results give answers to questions posed in the background section.

    It is a strong assumption that the universe is finite, because usually coverings are defined as a collection of nonempty subsets of an arbitrary universe. In this work we have restricted to finite universes, but we will prove whether the results this paper presented hold for covering approximation spaces with an infinite universe in the future. In addition, since there are more than 20 covering covering approximation operators, the invariance of separation can also be considered in other types of covering approximation spaces.

    The authors thank the anonymous reviewers'constructive suggestions. This work was supported by the National Natural Science Foundation of China (No. 11871259, No. 61379021), and the Educational Research Project for Young and Middle-aged teachers of Fujian Province (JAT190371).

    The authors declare that they have no conflict of interest.



    [1] Mo Z, Liu ZQ, Wu YC, (2008) The present situation analysis of agricultural production monitoring and output forecasting system at home and abroad. Chin Agric Sci Bull 24: 434–437.
    [2] Yang BJ, Lu DH, Pei ZY, Zhao HJ, Wu YY, (1997) The structure design of a national crop condition monitoring system. Trans Chin Soc Agricu Eng 13: 16–19.
    [3] Zhang L, Zhang Z, Luo Y, Cao J, Xie R, Li S, (2021) Integrating satellite-derived climatic and vegetation indices to predict smallholder maize yield using deep learning. Agric For Meteorol 311: 108666. https://doi.org/10.1016/j.agrformet.2021.108666 doi: 10.1016/j.agrformet.2021.108666
    [4] Bolton DK, Friedl MA, (2013) Forecasting crop yield using remotely sensed vegetation indices and crop phenology metrics. Agric For Meteorol 173: 74–84. https://doi.org/10.1016/j.agrformet.2013.01.007 doi: 10.1016/j.agrformet.2013.01.007
    [5] Qu G, Shuai Y, Shao C, Peng X, Huang J, (2023) County scale corn yield estimation based on multi-source data in Liaoning Province. Agronomy 13: 1428. https://doi.org/10.3390/agronomy13051428 doi: 10.3390/agronomy13051428
    [6] Qiao L, Tang W, Gao D, Zhao R, An L, Li M, et al. (2022) UAV-based chlorophyll content estimation by evaluating vegetation index responses under different crop coverages. Comput Electron Agric 196: 106775. https://doi.org/10.1016/j.compag.2022.106775 doi: 10.1016/j.compag.2022.106775
    [7] Zhou W, Liu Y, Ata-Ul-Karim ST, Ge Q, Li X, Xiao J, (2022) Integrating climate and satellite remote sensing data for predicting county-level wheat yield in China using machine learning methods. Int J Appl Earth Obs Geoinformation 111: 102861. https://doi.org/10.1016/j.jag.2022.102861 doi: 10.1016/j.jag.2022.102861
    [8] Lang P, Zhang L, Huang C, Chen J, Kang X, Zhang Z, et al. (2023) Integrating environmental and satellite data to estimate county-level cotton yield in Xinjiang Province. Front Plant Sci 13: 1048479. https://doi.org/10.3389/fpls.2022.1048479 doi: 10.3389/fpls.2022.1048479
    [9] Pede T, Mountrakis G, Shaw SB, (2019) Improving corn yield prediction across the US Corn Belt by replacing air temperature with daily MODIS land surface temperature. Agric For Meteorol 276: 107615. https://doi.org/10.1016/j.agrformet.2019.107615 doi: 10.1016/j.agrformet.2019.107615
    [10] He Y, Qiu B, Cheng F, Chen C, Sun Y, Zhang D, et al. (2023) National scale maize yield Estimation by integrating multiple spectral indexes and temporal aggregation. Remote Sens 15: 414. https://doi.org/10.3390/rs15020414 doi: 10.3390/rs15020414
    [11] Zhang L, Zhang Z, Luo Y, Cao J, Tao F, (2019) Combining optical, fluorescence, thermal satellite, and environmental data to predict county-level maize yield in China using machine learning approaches. Remote Sens 12: 21. https://doi.org/10.3390/rs12010021 doi: 10.3390/rs12010021
    [12] Bian C, Shi H, Wu S, Zhang K, Wei M, Zhao Y, et al. (2022) Prediction of field-scale wheat yield using machine learning method and multispectral UAV data. Remote Sens 14: 1474. https://doi.org/10.3390/rs14061474 doi: 10.3390/rs14061474
    [13] Di Y, Gao M, Feng F, Li Q, Zhang H, (2022) A new framework for winter wheat yield prediction integrating deep learning and Bayesian optimization. Agronomy 12: 3194. https://doi.org/10.3390/agronomy12123194 doi: 10.3390/agronomy12123194
    [14] Cao J, Zhang Z, Tao F, Zhang L, Luo Y, Han J, et al. (2020) Identifying the contributions of multi-source data for winter wheat yield prediction in China. Remote Sens 12: 750. https://doi.org/10.3390/rs12050750 doi: 10.3390/rs12050750
    [15] Wang J, Si H, Gao Z, Shi L, (2022) Winter wheat yield prediction using an LSTM model from MODIS LAI products. Agriculture 12: 1707. https://doi.org/10.3390/agriculture12101707 doi: 10.3390/agriculture12101707
    [16] Ren Y, Li Q, Du X, Zhang Y, Wang H, Shi G, et al. (2023) Analysis of corn yield prediction potential at various growth phases using a process-based model and deep learning. Plants 12: 446. https://doi.org/10.3390/plants12030446 doi: 10.3390/plants12030446
    [17] Maimaitijiang M, Sagan V, Sidike P, Hartling S, Esposito F, Fritschi FB, (2020) Soybean yield prediction from UAV using multimodal data fusion and deep learning. Remote Sens Environ 237: 111599. https://doi.org/10.1016/j.rse.2019.111599 doi: 10.1016/j.rse.2019.111599
    [18] Kumar C, Mubvumba P, Huang Y, Dhillon J, Reddy K, (2023) Multi-stage corn yield prediction using high-resolution UAV multispectral data and machine learning models. Agronomy 13: 1277. https://doi.org/10.3390/agronomy13051277 doi: 10.3390/agronomy13051277
    [19] Zhang J, Zhao Y, Hu Z, Xiao W, (2023) Unmanned aerial system-based wheat biomass estimation using multispectral, structural and meteorological data. Agriculture 13: 1621. https://doi.org/10.3390/agriculture13081621 doi: 10.3390/agriculture13081621
    [20] Zhou X, Zheng HB, Xu XQ, He JY, Ge XK, Yao X, et al. (2017) Predicting grain yield in rice using multi-temporal vegetation indices from UAV-based multispectral and digital imagery. ISPRS J Photogramm Remote Sens 130: 246–255. https://doi.org/10.1016/j.isprsjprs.2017.05.003 doi: 10.1016/j.isprsjprs.2017.05.003
    [21] Clevers J (1989) Application of a weighted infrared-red vegetation index for estimating leaf area index by correcting for soil moisture. Remote Sens Environ 29: 25–37. https://doi.org/10.1016/0034-4257(89)90076-X doi: 10.1016/0034-4257(89)90076-X
    [22] Wang XQ, Wang MM, Wang SQ, Wu YD, (2015) Extraction of vegetation information from visible unmanned aerial vehicle images. Trans Chin Soc Agric Eng 31: 152–159. https://doi.org/10.3969/j.issn.1002-6819.2015.05.022 doi: 10.3969/j.issn.1002-6819.2015.05.022
    [23] Jiang H, Hu H, Zhong R, Xu J, Xu J, Huang J, et al. (2020) A deep learning approach to conflating heterogeneous geospatial data for corn yield estimation: A case study of the US Corn Belt at the county level. Global Change Biol 263: 1754–1766. https://doi.org/10.1111/gcb.14885 doi: 10.1111/gcb.14885
    [24] Cao J, Zhang Z, Tao F, Zhang L, Luo Y, Zhang J, et al. (2021) Integrating multi-source data for rice yield prediction across China using machine learning and deep learning approaches. Agric For Meteorol 297: 108275. https://doi.org/10.1016/j.agrformet.2020.108275 doi: 10.1016/j.agrformet.2020.108275
    [25] Shahhosseini M, Martinez-Feria RA, Hu G, Archontoulis SV, (2019) Maize yield and nitrate loss prediction with machine learning algorithms. Environ Res Lett 14: 124026. https://doi.org/10.1088/1748-9326/ab5268 doi: 10.1088/1748-9326/ab5268
    [26] Chen T, Guestrin C, (2016) Xgboost: A scalable tree boosting system. in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining 2016: 785–794. https://doi.org/10.1145/2939672.2939785
    [27] Li H, Cao Y, Li S, Zhao J, Sun Y, (2020) XGBoost model and its application to personal credit evaluation. IEEE Intell Syst 35: 52–61. https://doi.org/10.1109/MIS.2020.2972533 doi: 10.1109/MIS.2020.2972533
    [28] He XH, Luo HT, Qiao MJ, ZH Tian, GS Zhou, (2021) Yield estimation of winter wheat in China based on CNN-RNN network. Trans Chin Soc Agric Eng 37: 124–132. https://doi.org/10.11975/j.issn.1002-6819.2021.17.014 doi: 10.11975/j.issn.1002-6819.2021.17.014
    [29] Zhang YB, Li X, Man WD, Liu MY, Fan JH, HR Hu, et al. (2024) Research on yield estimation method of winter wheat based on Sentinel-1/2 data and machine learning algorithms. Acta Agric Zhejiangensis 36: 2812–2822. https://doi.org/10.3969/j.issn.1004-1524.20231368 doi: 10.3969/j.issn.1004-1524.20231368
    [30] Lundberg SM, Lee SI (2017) A unified approach to interpreting model predictions. Adv Neural Inf Process Syst 30.
  • This article has been cited by:

    1. Liying Yang, Jinjin Li, Yiliang Li, Qifang Li, Sub-base local reduct in a family of sub-bases, 2022, 7, 2473-6988, 13271, 10.3934/math.2022732
  • Reader Comments
  • © 2025 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(324) PDF downloads(19) Cited by(0)

Figures and Tables

Figures(8)  /  Tables(8)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog